CT x CCH | Strategic AI Partnership

Process	What CCH Says (verbatim)	CT Fit	Pillar
Next Gen OTC	“End-to-end customer experience from order to delivery, deliver mDSD to enable track & trace and touchless settlement, increase DIFOT AI and NPS.”	Smart Order AI, Where-Is-My-Order Agent, IOM exceptions	Lead
Promo360	“Deploy the holistic End-To-End Promo Platform across CCH, aligning promo process, tools, data and org. capabilities.”	Promo Activation Assistant, TPO, Pricing Co-pilot	Lead
RGM — Pricing Co-pilot & OBPPC	“Advance our RGM tools using AI to improve the speed and quality of decision making”; PVP, Pricing Co-pilot, OBPPC Automation.	TPO, Customer Contract Simulation, AI-driven RGM	Lead
Next Gen Bespoke (Segmented Execution)	“Segmented Execution… Omnichannel Suggested Orders & Recommended Activities, RGM toolkit, INC, Digital Twin, Contract Mgmt, AI-enabled audit, People Analytics.”	Perfect Store Suggested Order, NBA, CT Mobile	Lead
MDM Oxygen	“Material MDG deployment and agentic enabled scaled DQ.” Customer, Material, Finance, Vendor & Employee.	Digital Data Steward (agentic MDM, ARJO live)	Lead
Customer Contract Management	“Connect Customer Contract process from commercial policy to P&L simulations… decrease customer waiting time.”	Deal Lifecycle Copilot / CPQ (EnerSys reference)	Lead
Customer Service & NPS	NPS 78 (up from 66); target BP26 NPS 80. “99% of customer issues within 48 hours.”	Where-Is-My-Order Agent, invoice dispute agent	Lead
Connected Enterprise Planning	“Drive financial planning to increase efficiency… step up scenario planning… full utilisation of BlueYonder.”	Planning data & AI integration support	Support
Manufacturing IIOT & Digital Twin	“Connect production process and lines to enable virtual representation of production process, improve line utilisation.” Predictive maintenance on 55 lines, +15 in 2025.	Predictive maintenance ML, QC computer vision, energy MPC	Support
Connected Coolers / IoT Hub	BP26 Priority 1: “Coffee & Connected Coolers — IoT Hub Automation.”	Field service virtual agent (Agentforce + RAG on manuals)	Support
P&C Service Desk with AI (Hellen+)	“Introduce P&C Service Desk and new virtual agent capabilities on Refresh for BDs & Operators.”	Hellen+ HR virtual agent (AIUC_10 Bacardi reference)	Lead
Sales Academy Metaverse	“Immersive learning for all BDs. Drive performance through deployment. AI coach.”	Active CT scope (17 HC); Extend: AI performance analytics	Lead
Enterprise Insights Transformation	“Transiting from BW to forward looking reporting, powered by AI.”	Talk-to-your-data NLQ, Data Mesh 2.0 support	Support
iGRC & AI-enabled audit	“Continue implementation of iGRC solution to empower cross-functional and enterprise-wide risk visibility”; AI-enabled audit is BP26 Priority 1.	AI-enabled internal audit & SOX controls (Takeda reference)	Lead
SAP S/4HANA upgrade (2027 cut-off)	“Migration of in-sourced payroll & mini-master data for ERP to S/4 as no support by SAP after 2027.”	SAP Institutional Knowledge Copilot (AB InBev reference)	Lead

KBI	BP26 Target	FY25	What closes the gap
eB2B Monthly Active Users (Customer Portal + SIRVIS)	152k	120k	SIRVIS Conversational Commerce, Where-Is-My-Order Agent, loyalty (KOF Juntos+ pattern — 1.3M)
RED Image Recognition coverage	86%	77%	CT Vision Store Check (already in CCH environment)
Segmented Execution (SO, RA, WHS)	60%	43%	Perfect Store Suggested Order, NBA visit steps, Talk-to-your-data
Leads → Customers (New Customers)	72k	42k	Menu Data Bot for HoReCa, Credit Scoring Agent, Deal Lifecycle Copilot
Overall Customer Satisfaction (NPS)	80	78	Where-Is-My-Order Agent, invoice dispute agent, SIRVIS copilot
INC Mean-Time-To-Resolve (MTTR)	2 days	TBD (was 4)	CT AI-Driven AMS Context Memory Layer — 5× MTTR improvement
Reduce # of incidents	-20%	TBD	Code Quality Dashboard, GenAI validator suite, Predictive Maintenance

Metric	CCH	CCEP	KOF
Revenue	€11.6bn	€20.9bn	US$14.9bn
EBIT / Op margin	11.7%	13.4%	14.7%
Revenue / employee (rough)	~€346k	~€536k (+55%)	~US$160k
Digital B2B MAU	120k (target 152k)	MyCCEP: €2.5bn portal revenue	1.3M (56% customers monthly)
Sales copilot revenue contribution	n/d	“Sales Force of the Future”	1.9% of total sales (Juntos+ Advisor)
Disclosed digital/AI savings	n/d	€350–400m by 2028; -290bps opex/rev	US$136m in 2025
Voice picking productivity	93.5% usage; uplift n/d	n/d	+16% productivity
Dynamic routing coverage	BP26 priority, no % disclosed	n/d	90% of last-mile routes
Predictive maintenance	BP26 priority, 55 lines rolling out	n/d	Live (ML-enabled)
Digital capex share	16% of capex (≈€132m)	~28% of capex mix	Part of 9.0%/rev

#	Gap	Peer benchmark	CT asset to close
1	B2B portal MAU scale	KOF Juntos+ 1.3M MAU (56% monthly buyers); CCEP MyCCEP €2.5bn portal revenue	SIRVIS Conversational Commerce + loyalty + NBA recommender (UC #15)
2	Agentic GenAI sales copilot at scale	KOF Juntos+ Advisor = 1.9% of sales (ML + GenAI)	Vocal C360 Briefing, NBA Visit Steps, Perfect Store Suggested Order (UC #1, #7, #8)
3	AI-driven RGM / promo simulation in production	KOF “AI simulations for market share/revenue/volume” in BR, MX, CO. CCEP price-modelling for promo ROI.	Trade Promotion Optimisation, Customer Contract Simulation (UC #39, #42)
4	ML predictive maintenance at scale	KOF: ML-PdM live, reducing unplanned downtime	Predictive Maintenance for lines & coolers (UC #37)
5	Warehouse productivity / voice picking uplift	KOF Voice Picking +16% productivity, 93% Brazil coverage	Layer CT Vision + Voice on warehouse workflows (adjacent)
6	Last-mile dynamic routing	KOF: 90% of last-mile routes on digital platforms	Route Optimisation & DSD Exception Handling (UC #38)
7	Agentic GenAI in Shared Services	CCEP Manila ISS: “Agentic & GenAI further automates process & reporting”	Hellen+ HR virtual agent, Where-Is-My-Order, Invoice dispute agent (UC #44, #16, #18)
8	Quantified opex reduction from digital	CCEP -290bps opex/rev in 4 years; KOF US$136m 2025	CT AI-Driven AMS (40-60% cost reduction), GenAI validator suite (UC #26–35)
9	D2C digital platform	KOF En Tu Hogar: 135k monthly buyers, 2× digital ticket	Conversational commerce + D2C playbook (SIRVIS adjacent)
10	Scale of agile digital org	KOF: 30+ agile digital teams, ~600 specialists	CT AI-Driven AMS + 53 HC already embedded
11	Loyalty as execution lever	KOF Premia: 1.6M enrolled, 82% redemption, drives cooler placement	SIRVIS loyalty layer (UC #15 + NBA)
12	ML demand planning	KOF Demand Planning 360	Demand Forecasting & Replenishment Copilot (UC #36)

Slide 1 — 71 Use Cases

The Bottler AI Use Case Library — What We Have, What We Can Build, What Needs a Partner

71 practical use cases across 7 domains (66 core + 5 RGM Command Center engines). Every card is tagged with one of four status labels so CCH can see at a glance what is production-ready vs. what still needs work. Every card has a concrete practical example — a named person, a named place, a real scenario — so you know exactly what the thing actually is.

Reading the status labels

Have — Live

CT has this asset in production at a customer today. CCH-specific tuning required, but the core asset is built and running.

Have — Template

CT has the asset / code / pattern, but it needs CCH-specific configuration and data wiring. Build effort stated per card.

Don't Have — Can Build

No pre-built CT asset, but CT has the skills, platform partnerships, and pod to build it. Rough weeks-of-effort given per card.

Don't Have — Partner

Outside CT's lane. Requires a partner (BlueYonder, Siemens, Microsoft Security Copilot, Watershed, SAP EWM, Geotab, etc.). CT integrates and plays orchestration layer.

Have — Live

Have — Template

Don't Have — Can Build

Don't Have — Partner

Commercial & Sales

UC #1–15, #49–56

Order-to-Cash & Service

UC #16–20, #57

Data, MDM & Knowledge

UC #21–25, #58

GenAI for IT / Delivery

UC #26–35, #59

Manufacturing & Supply

UC #36–43, #60–64

Digital Employee

UC #44–48, #65–66

Domain 1 — Commercial & Sales

Cards 1–15 • 49–56

UC 1Vocal Customer 360 Briefing

Have — Live

Digital Employee / Digital Consumer

What it is: Before a BD walks into an outlet, the CT Mobile AI Assistant reads a 30–45 sec voice brief: last 3 orders, share-of-throat vs competitor, open tickets, promo compliance from last photo, and the one “quick win” for today. Hands-free on Agentforce + Data Cloud.

Problem: BDs arrive cold; reps miss the most critical insight per visit.

Tech: CT Mobile + Agentforce + Salesforce Data Cloud + Azure OpenAI TTS/STT; RAG over visit/order history + RED compliance; offline-capable.

📍 Practical exampleTuesday 9:12am. Maria Papadopoulou, a BD in Thessaloniki, parks outside Café Aroma on Tsimiski Street. Her phone plays a 40-second brief in Greek: “Last visit 12 days ago. Owner Nikos Georgiadis mentioned building a new terrace. Monster 500ml order was half size last week. RED score on the back cooler dropped 6 points. Quick win today: propose the 2-for-1 Sprite Zero end-cap — margin €14, Nikos said yes to this format in March.” Maria walks in already knowing what to say.

P&L: +2–4% NSR per visited outlet. Formula: 1.2M outlets × 25% touched/qtr = 300k × €180 avg order × 2% uplift ≈ €4.3m/qtr (~€17m/yr).

Start: 200 BDs Romania on Next Gen OTC

UC 2Talk-to-Your-Data Natural Language Filtering

Have — Template

Digital Employee

What it is: Rep says “Show me HoReCa outlets in Sofia under 50% loyalty that haven't ordered Monster in 6 weeks.” Agent parses, builds the SOQL query, and filters the account list offline.

Problem: Reps never touch 40+ SFA filters; analytical data becomes shelfware.

Tech: CT Mobile + CT Zen; on-device SLM for intent parsing; Azure OpenAI fallback; Salesforce metadata-aware query builder. Build effort: ~6–8 weeks.

📍 Practical exampleWednesday 2:40pm in Bucharest. Andrei Ionescu, an ASM, has 20 minutes between meetings. He taps the mic on CT Mobile and says in Romanian, “Arată-mi cafenelele din Sectorul 3 care n-au comandat Fuze Tea în ultimele patru săptămâni.” Two seconds later the list shows 17 accounts sorted by last order value. He forwards the top 5 to his BD Elena via Teams with a note “azi după-amiază.”

P&L: -30–50 min/rep/day admin. Formula: 45 min × 220 days × €45/hr × 2,500 reps = ~€9.3m/yr oxygen released.

Start: next CT Mobile wave, IT or GR

UC 3AI-Assisted Visit Scheduling & Real-Time Next-Stop

Have — Live

Digital Employee / Data, Insights & AI

What it is: On a cancellation or gap, CT Mobile recommends the next best outlet within X km: “Café Centrale, 1.2 km, 18 days overdue, €480 avg order.” Also builds optimised multi-day routes.

Problem: Gaps become wasted hours; tail accounts under-visited, A-accounts over-visited. Travel is 20–25% of the working day.

Tech: CT Mobile + SF Maps + Einstein/Databricks ML scoring.

📍 Practical exampleThursday 11:47am in Naples. Giuseppe Romano is a BD whose 12:00 meeting at Trattoria da Mario just cancelled — Mario is at the hospital with his son. Giuseppe taps “What's next” on CT Mobile. The agent says: “Bar Vesuvio, 800 metres, 21 days overdue, avg €340, owner Francesco is a Napoli fan — today is a match day, push the Powerade 6-pack promo.” Giuseppe walks, closes €390 in 15 minutes, and is back on schedule.

P&L: +1–2 productive visits/rep/week. Formula: 1.5 visits × 45 wks × 2,500 reps × 30% conv × €160 = ~€8.1m/yr NSR + ~€1.2m/yr travel.

Start: pair with Dynamic Routing / BlueYonder BP26

UC 4Smart Order Assistant — Photo/Screenshot → Order Draft

Have — Live

Digital Consumer & Customer

What it is: Customer hands BD a paper list, WhatsApp fridge photo, or Excel PO. “New order from photo” — agent OCRs, matches SKUs, fills quantities, checks price/quota, shows reviewable draft in <10 sec. CT Smart Order Assistant GA Oct-2025.

Problem: “Human error rates in order-taking 5–15%”. Every retype = 2–4 min, every error = dispute.

Tech: CT Smart Order Assistant on CT Mobile; GPT-4o multimodal OCR + SKU match; price/quota guardrail.

📍 Practical example11:03am in Sofia. A Bulgarian HoReCa owner, Dimitar from Mehana Vodenitsata, sends CCH's tele-sales a photo of a handwritten fridge count on a napkin. The rep Yana taps “New order from photo” — 9 seconds later the cart shows 14 SKUs matched against the Bulgarian price book, 2 flagged for quota swap (Schweppes Tonic → Schweppes Bitter Lemon), and the delivery slot for Thursday. One tap: submit. Dimitar gets the SMS confirmation before he finishes his cigarette.

P&L: -8–12% dispute rate. Formula: 10% error rate × €18/order × ~15m photo-addressable orders/yr = ~€2.7m/yr dispute avoidance + 625k hrs back.

Start: Next Gen OTC pilot RO+BG

UC 5Smart Order Assistant — Mass Actions & Template Cloning

Have — Live

Digital Consumer & Customer

What it is: “Clone last month's order, double the Energy SKUs, remove discontinued packs, add the current Sprite Zero promo” — one utterance, no scrolling 60 lines.

Problem: 40–80 line carts eat visit time.

Tech: CT Smart Order Assistant + cart mutation tool calls (bulkAdd/bulkReplace/applyPromo); pricing & quota guardrails.

📍 Practical exampleMonday 8:20am in Dublin. Siobhan O'Brien, a BD covering south Dublin pubs, is at The Long Hall. The landlord Patrick wants “the usual, but drop the Fanta Orange, add two cases of Monster Ultra, and put me down for whatever Christmas POS you have.” Siobhan says exactly that to CT Mobile. Six seconds later a 37-line cart appears — Fanta removed, 2×Monster Ultra added, 2026 Christmas POS bundle pre-attached. Patrick nods; she submits.

P&L: -4–6 min/repeat order. Formula: 5 min × 60% order-bearing visits × 18 visits/day × 220 days × 2,500 reps ÷ 60 = ~495k hrs/yr capacity.

Start: ship with UC 4 in same OTC wave

UC 6Quota Check & Smart Substitution in the Cart

Have — Live

Digital Consumer & Customer / Data, Insights & AI

What it is: Before Submit, agent scans cart, flags every SKU breaching quota/allocation/policy, offers one-click swap: “Line 7: Coke 1.5L quota exceeded — suggest Coke 1.25L (same margin, in stock).”

Problem: Blocked SKUs reach depot → ship-refuse → dispute → re-pick.

Tech: CT Smart Order Assistant + CT Mobile quota engine + SAP real-time allocation.

📍 Practical exampleFriday 4:18pm in Vienna. Klara Huber, a KAM, is closing a 52-line order with the purchasing manager of the Hotel Sacher group. As she taps Submit, the agent flashes: “Line 23 — Almdudler 1L exceeds Q2 allocation by 48 cases. Suggest Almdudler 0.75L, margin +€0.22/case, in stock Wiener Neudorf depot, on the Sacher approved list since 2023.” Klara one-taps the swap, the order goes through clean, no 6pm phone call from depot.

P&L: -3–5 pts rejection rate → +1–1.5 pts DIFOT. Formula: 4% reject rate × 120m lines × €3.20 handling ≈ €15m → half = ~€7.5m/yr.

Start: Next Gen OTC touchless workstream

UC 7Perfect-Store Suggested Order (Recommended Order Engine)

Don't Have — Can Build

Data, Insights & AI

What it is: For every outlet, an ML model produces a pre-filled order draft before the visit — sell-in, sell-out, promos, weather, events. Rep walks in with 75%+ accept-rate cart; non-visited outlets get the pre-order pushed to SIRVIS.

Problem: Manual order construction misses promos, under-orders fast movers. BP26 P1: “Omnichannel Suggested Orders.”

Tech: Databricks / Data Mesh 2.x feature store + time-series + uplift model; CT Mobile + SIRVIS delivery. Build effort: 12–16 weeks per market.

📍 Practical exampleMonday morning in Cork. Declan Murphy is about to visit Centra Ballincollig. His CT Mobile already has a draft 28-line cart waiting: the usual Coke Zero and 7UP base, +3 cases Powerade because Munster Rugby has a home fixture Saturday, +2 cases Deep RiverRock 500ml because the forecast is 19°C, no Innocent smoothies because the outlet has 11 cases on hand from the last visit photo. Declan walks in, shows the draft to the store manager Aoife, she changes one line (-1 Powerade), taps approve.

P&L: +3–8% order size. Formula: 1.2m outlets × 60% reach × +5% size × €180 × 8 orders/yr = ~€52m/yr NSR. Baseline: CCEP WhatsApp recommender +5–20%.

Start: IE+AT where EPOS cleanest; Segmented Execution 60%

UC 8Next Best Action / Visit Step Personalisation

Have — Template

Digital Employee / Data, Insights & AI

What it is: Per visit, the agent ranks steps: “Step 1: cooler re-placement. Step 2: Sprite Zero 2-for-1 (margin +€14). Step 3: replace expired POS. Step 4: ask about new on-trade license.” Outlet-specific, not generic.

Problem: Store-check scripts are identical across 500 outlets; reps disengage.

Tech: CT Mobile “Recommended Visit Steps” + Einstein NBA + Databricks; needs CCH data wiring to RED + promo calendar.

📍 Practical exampleTuesday 10:05am in Warsaw. Kasia Nowak walks into Żabka Mokotów with a ranked 4-step card: (1) fix the Cappy Pulpy facing — slipped to 2 of 4 slots last visit; (2) push the 1.75L Kinley Tonic bundle — store manager Tomasz tested it in March and liked the margin; (3) swap the torn Coca-Cola Zero shelf-strip; (4) ask Tomasz about the new store in Ursynów. Kasia does steps 1–3 in 14 minutes and uses the saved time for step 4. Closes a second-location commitment.

P&L: +3–6% perfect-store compliance. Formula: 1.2m outlets × €2.40 incremental NSR/compliance-pt-month ≈ €8–15m/yr.

Start: layer on existing RED 86% coverage

UC 9Voice-to-Visit-Report (“Voice-to-Action”)

Have — Template

Digital Employee

What it is: Walking out of the outlet, rep speaks freely; agent fills visit report fields, creates follow-up task, logs competitor intel, drafts KAM email. Offline, synced on next connect.

Problem: 20–30 min/day paperwork, one-handed in the car park.

Tech: CT Mobile + Whisper STT + GPT-4o structured extraction to SF objects. MVP on roadmap.

📍 Practical example5:42pm in Lagos. Chinedu Okafor, a BD for CCBA Nigeria, walks out of Shoprite Ikeja City Mall into heat and traffic noise. He presses the mic in CT Mobile and says: “Store manager Funke pushed back on the new Schweppes price — agreed to trial 10 cases Schweppes Pineapple for two weeks, needs a POS refresh by Friday, competitor is running a Pepsi-and-Gala-sausage bundle till month end, follow up next Tuesday.” The agent files it into Salesforce, creates the POS task, logs the Pepsi bundle, and drafts the email for his KAM. Chinedu drives home without touching the phone again.

P&L: -20 min/rep/day. Formula: 20 min × 220 days × 2,500 reps ÷ 60 = 183k hrs/yr; half redeployed ≈ €24.7m/yr NSR upside.

Start: “Releasing Oxygen” BP26 theme

UC 10Manager's Briefing / Daily Territory Memo

Have — Template

Digital Employee

What it is: Every morning each ASM gets a 1-page AI-generated memo reading all team visit reports from yesterday, flagging exceptions, proposing 3 coaching conversations.

Problem: ASMs drown in 20–40 reports/day and skim them.

Tech: Azure OpenAI + CT Institutional Knowledge pattern; scheduled pipeline reading SF visit reports + RED + anomalies.

📍 Practical exampleWednesday 7:50am in Belgrade. Miloš Petrović, an ASM managing 14 BDs across Vojvodina, opens Teams on his way to the office. A card titled “Yesterday in your territory” says: (1) BD Jelena had 3 outlets refuse the Schweppes price increase in Novi Sad — all Serbian-flag independents, same objection wording. Suggest: joint visit today. (2) RED score dropped 9 points at Maxi Liman. (3) Monster 500ml stockout at the Petrovaradin stadium kiosk before Saturday's match. Miloš books two 15-minute coaching calls before he reaches his desk.

P&L: +10–15% faster reaction. Formula: 30 min/ASM/day × 400 ASMs × 220 days = 44k hrs/yr redeployed to coaching.

Start: Sales Academy Metaverse cohort

UC 11Promo Activation Assistant

Have — Template

Digital Consumer & Customer

What it is: In-visit, rep asks “What promos are live here today?” Agent returns applicable promos, POS assets, placement, and one-tap adds promo SKUs to cart.

Problem: Promo leakage — reps can't remember which promos apply where. BP26 Promo360 is the programme; this is its field layer.

Tech: CT Mobile + Promo360 integration + CT Smart Order Assistant; RAG over promo docs + eligibility rules + POS CDN. Integration 6–10 weeks.

📍 Practical exampleSaturday 10:30am in Milan. Francesca Bianchi is in a small alimentari on Via Padova. The owner, Signora Rizzo, asks “che promozioni ci sono oggi?” Francesca taps her phone and reads back in Italian: “Three promos apply today — Fanta 1.5L 3-for-2 until 28 April, Lurisia 0.75L end-cap bundle with €18 margin per case, and a free branded ice-bucket with 15 cases of Kinley. Shall I add all three?” Signora Rizzo picks two; one tap adds 22 lines to the cart.

P&L: +15–25% promo redemption. Formula: 1,500 promos × €180k avg uplift × 20% leakage recovery = ~€54m/yr NSR.

Start: marquee field deliverable for Promo360 P1

UC 12AI-Powered Store Check & Planogram Compliance (CT Vision)

Have — Live

Data, Insights & AI

What it is: Rep points the phone at the cooler or shelf; CT Vision recognises every SKU, calculates share-of-shelf vs planogram, scores distribution, flags OOS, files evidence offline.

Problem: Manual store-check on a 12-door cooler takes 6–9 min and is inconsistent between reps.

Tech: CT Vision (already live at CCH on RED) + edge inference + cloud retraining.

📍 Practical exampleThursday 3:15pm in Athens. Dimitris Karalis is in a kiosk in Kolonaki. He raises the phone to the 6-door cooler, sweeps it once, and CT Vision returns in 4 seconds: “Share of shelf Coca-Cola 48% (target 55%), Coke Zero missing from door 3, planogram compliance 71% (-4 vs. last visit). Two out-of-stocks on 330ml cans.” The rep fixes door 3 on the spot and the evidence is logged against the RED KBI.

P&L: +2–5 pts RED KBI → +1–3% volume. Formula: RED 86%→91% × 1.2m outlets × 1.5% NSR lift = €30–45m/yr.

Start: close 86→90% gap; Egypt Commercial BP26 P2

UC 13Menu Data Bot — HoReCa Intelligence

Have — Template

Data, Insights & AI

What it is: Continuous crawler pulls menu PDFs/web pages for bars, restaurants, cafés; GPT-4o extracts beverage brands, prices, categories. Output feeds CT Mobile as prioritised visit lists.

Problem: HoReCa acquisition depends on knowing who lists what — today done by rep eyeballs.

Tech: Azure Functions crawler + Azure OpenAI extraction + Snowflake/Databricks. UAT complete for a global CPG beverage customer. CCH language packs 6–10 weeks.

📍 Practical exampleMonday 7:00am in Rome. The HoReCa acquisition lead, Stefano Conti, opens his Power BI dashboard. Overnight the crawler pulled 2,340 updated menus across Trastevere and Testaccio. Twelve restaurants added Pepsi listings last week; seven cocktail bars are listing San Pellegrino where the CCH contract says Lurisia. The top 20 leads drop into BD Laura's CT Mobile list before she starts her day.

P&L: +5–10% HoReCa acquisition. Formula: 3–5k incremental qualified × 55% conv × €3,200 first-year value = €5.3–8.8m/yr.

Start: GR+IT on-trade, Leads→Customers KBI

UC 14Deal / Contract Lifecycle Copilot (CPQ for Bottlers)

Have — Live

Digital Consumer & Customer / Digital Enterprise

What it is: For HoReCa contracts, the agent guides the KAM through structure, proposes pricing tiers grounded in comparable deals, simulates P&L scenarios, drafts the contract. Adapted from EnerSys DLM / Optimus.

Problem: BP26 explicitly: “decrease customer waiting time” on Customer Contract Management. Big HoReCa contracts take 2–4 weeks.

Tech: Salesforce CPQ + CT DLM accelerator + Agentforce; RAG over past contract corpus + pricing rules; Databricks P&L simulator.

📍 Practical exampleTuesday 4:00pm in Zurich. Patrick Meier, a national account KAM, is negotiating a 3-year deal with the SV Group restaurant chain. The CPQ copilot pre-populates: Sprite 1L at CHF 0.78, Schweppes listing fee waiver in exchange for 18-month exclusivity on tonic, a cooler loan over 40 outlets. It runs the P&L: year-1 margin -0.4 pts, year-2 +1.1 pts, 3-year NPV positive at CHF 1.9m. Patrick tweaks the tonic clause, the contract PDF re-renders in 11 seconds. He sends it the same day instead of three weeks later.

P&L: -40–60% cycle time; +5–10% pricing discipline. Formula: 10k contracts × 16hrs × €110/hr ≈ €17.6m/yr + €2m/yr pricing.

Start: HoReCa chains CH+IE; Contract Mgmt BP26 P1

UC 15Customer Portal Conversational Commerce (SIRVIS Copilot)

Have — Live

Digital Consumer & Customer

What it is: Outlet owner opens SIRVIS, voices or types their reorder in their own language. Agent completes the order, checks stock/slot, applies promos, confirms.

Problem: BP26 SIRVIS MAU target 152k — typing baskets on mobile is painful for small outlet owners.

Tech: Azure OpenAI + multilingual guardrails + Agentforce (AIUC_01 pattern) + ARJO Laura 2.0; SAP + SIRVIS.

📍 Practical example8:45pm in Limassol. A kiosk owner, Andreas, opens SIRVIS on his iPhone. He speaks in Greek: “Δώσ' μου την παραγγελια της περασμένης Τετάρτης, άλλαξε τη Φάντα με Schweppes Lemon, φέρ' τη μέχρι την Πέμπτη.” The bot confirms Thursday 11am–1pm delivery, applies the current pack promo, shows the new total. Andreas taps yes and goes back to closing up. No phone call to the Nicosia service centre.

P&L: +20–30% SIRVIS MAU; -15–25% L1 calls. Formula: overshoot 25k MAU × €420 order share × 14% margin = €1.5m/yr EBIT + 300k calls × €4.80 = €1.4m/yr OPEX.

Start: 1 Developing + 1 Established Q2; KBI = eB2B MAU

UC 49Coupon & Rebate Fraud Detection

Don't Have — Can Build

Data, Insights & AI / Digital Enterprise

What it is: ML layer on the rebate settlement flow flagging suspicious claims — duplicate submissions across chains, rebate volumes inconsistent with delivered cases, implausible geographic concentration, photo-evidence forgery (same shelf photo submitted for two outlets).

Problem: Rebate / promo-claim fraud and over-claiming is 1–3% of trade spend in most CPG systems.

Tech: Databricks anomaly detection + CT Vision (photo dedupe + forgery detection) + Agentforce + SAP rebate ledger. Build 10–14 weeks.

📍 Practical exampleTuesday 10:12am in Athens. Claim-handler Sofia Dimitriou opens her queue. Top item: a chain submitted a rebate claim for 840 cases of Fanta Lemon at the Patras branch — but CT Vision matches the “shelf evidence” photo to a photo already submitted by the Larissa branch two weeks ago, same angle, same cracks in the tile floor. The agent drafts a hold-the-claim email with both photos side by side. Sofia escalates to the KAM.

P&L: recover 30–50% fraudulent claims. Formula: €700m trade spend × 1.5% fraud × 40% recovery = ~€4.2m/yr.

Start: GR+IT modern trade (largest trade-spend)

UC 50Shopper Marketing Personalisation on the Consumer CRM

Have — Template

Digital Consumer & Customer

What it is: For CCH's direct-to-shopper programmes (Coke app, loyalty codes under the cap), ML segments shoppers, predicts next-best-brand, personalises offers per cohort.

Problem: Shopper CRM activations today are broadcast — same email to 10m consumers. Response rates <1%.

Tech: Salesforce Data Cloud + Einstein Personalization + Azure OpenAI copy generation. Build 10–14 weeks.

📍 Practical exampleWednesday evening 8:00pm across Poland. CCH's “Summer Coke Zero” push goes out. Instead of one email, the system sends 47 variants: a Kraków cohort of 18–24s gets a TikTok-style video with a Kraków-specific geo-code, a Warszawa working-parent cohort gets a 3-for-2 family-pack offer, a Silesian over-50 cohort gets a classic-glass-bottle nostalgia angle. Response rate jumps from 1.4% to 4.9%.

P&L: +3–5× response rate. Formula: 10m CRM × 2% base response × 3× lift × €0.60 margin/response × 4 campaigns = ~€14m/yr.

Start: PL+IT (largest consumer base)

UC 51Competitive Price Scraping & Real-Time Alerting

Have — Template

Data, Insights & AI

What it is: Daily scraper pulls retailer sites, e-commerce catalogues, Glovo/Wolt/Getir across 29 markets, extracts prices and promos for CCH SKUs and competitors, alerts RGM on guardrail breaches.

Problem: RGM today relies on weekly/monthly price surveys — often outdated before they land. Competitor promos run for 10 days before CCH notices.

Tech: Azure Functions crawler + Azure OpenAI extraction (same lineage as UC 13) + Databricks + Power BI alerting. Build 6–10 weeks.

📍 Practical exampleMonday 6:00am in Budapest. RGM manager Zsófia Nagy opens her Power BI alert: “Lidl Hungary dropped Pepsi 1.5L from HUF 449 to HUF 379 overnight, 18% below our current list. 87 stores affected.” At 7:15am she's in a standup with the Hungarian commercial director; by 10am a targeted counter-promo on Coca-Cola 1.5L has been approved and pushed to the Promo360 backlog.

P&L: 4–7 days faster reaction. Formula: 1 pt share on €10.7bn = €107m; defending 0.4 pts/yr → ~€42m directional; realistic capture 10% = ~€4m/yr.

Start: HU+RO+PL modern trade

UC 52Outlet Churn & Dormancy Prediction

Don't Have — Can Build

Data, Insights & AI / Digital Consumer & Customer

What it is: ML model scores every one of 1.2m outlets weekly for churn probability (“68% probability of becoming dormant within 60 days”) based on order frequency, avg size, complaints, disputes, RED compliance.

Problem: Churn today is reactive — CCH notices after the customer has stopped ordering for a quarter. Reactivation is 3–5× harder than retention.

Tech: Databricks ML (gradient boosting + time-series) + CT Mobile + Agentforce retention playbook. Build 10–14 weeks.

📍 Practical exampleThursday morning in Naples. BD Roberto Esposito opens his CT Mobile and sees a new “retention” tab — 11 accounts at risk this week. Top of the list: Café del Sole on Via Caracciolo, 68% churn probability — orders halved over 5 weeks, three delivery disputes unresolved, RED score fell. The agent proposes a retention playbook: personal visit, escalate the disputes, offer a 30-day loyalty boost. Roberto visits on Friday and saves the account.

P&L: -15–25% dormancy on targeted cohort. Formula: 4% churn × 1.2m outlets × €1,200 avg NSR = €57.6m at risk × 20% recovery = ~€11.5m/yr.

Start: IT+GR traditional trade

UC 53Merchandiser Photo Coaching Agent

Have — Template

Digital Employee / Data, Insights & AI

What it is: When a merchandiser takes an RED photo, CT Vision not only scores compliance but gives real-time coaching on how to fix a sub-par execution. Gamified streak + team leaderboard.

Problem: Compliance data becomes actionable only retrospectively. Merchandisers don't know what “good” looks like because the feedback loop is weekly at best.

Tech: CT Vision + Azure OpenAI coaching layer + CT Mobile UI. Build 6–8 weeks on top of the live CT Vision model.

📍 Practical exampleMonday 11:30am in Cluj-Napoca. Merchandiser Mihai Popa snaps a photo of a Profi store cooler. The phone speaks in Romanian: “Coca-Cola Zero e pe raftul 4 — mut-o pe raftul 2, la nivelul ochilor. Stick-ul roșu e în buzunarul tău. Fă o poză nouă când ai terminat.” Mihai fixes it, re-photos, and the compliance score jumps from 64 to 88. The team leaderboard shows he moved up 3 positions.

P&L: +4–8 pts compliance. Formula: incremental 2–4 pts RED × 1.2m outlets × 1% NSR lift ≈ €10–20m/yr marginal.

Start: RO+BG (existing CT Vision deployments)

UC 54New Product Launch Forecasting (Cold-Start)

Don't Have — Can Build

Data, Insights & AI

What it is: For a net-new SKU with no history, the model produces a cold-start forecast using analog-SKU weighting, consumer-research signals, promo plan, distribution targets, comparable-market launches from the 29-country portfolio.

Problem: New-product forecasting is a gut-call; over-forecast → write-off; under-forecast → stockout in first 4 weeks and the launch is dead.

Tech: Databricks + analog-matching algorithm + SAP IBP + Azure OpenAI. Build 12–16 weeks.

📍 Practical exampleMonday 9:00am in Vienna. Brand manager Julia Mayer is launching a new Monster Mango Loco flavour in Austria in 6 weeks. The cold-start tool shows: expected week-1 sell-in 47k cases (confidence 65%), peak week-4 80k cases, steady-state from week-12 at 35k/week; driven by analog weighting to the 2023 Monster Ultra Paradise Austrian launch (52%), the 2024 Croatian Mango launch (30%), and current category growth (18%). Julia accepts the plan and shares with supply.

P&L: -25–40% launch forecast error. Formula: 40 launches/yr × €400k write-off avoidance = ~€16m/yr.

Start: next 2 major 2026 innovation launches

UC 55Pack-Price Architecture Optimiser (OBPPC)

Don't Have — Can Build

Data, Insights & AI

What it is: ML layer that analyses CCH's Occasion-Brand-Package-Price-Channel (OBPPC) lattice per market and recommends the optimal mix — which pack sizes, at which price points, in which channels.

Problem: OBPPC is reviewed yearly; consumer elasticity shifts faster than that. BP26 P1 includes “PVP/PAAT/OBPPC automation.”

Tech: Databricks causal ML + mixed-integer optimiser + Power BI + CT RGM pod. Build 14–18 weeks.

📍 Practical exampleWednesday 10:00am in Sofia. RGM analyst Petar Ivanov reviews the optimiser output for Bulgarian traditional trade: “Introduce a 1.25L Coke at BGN 2.89 between the 1L and 1.5L, retire the 1.75L which cannibalises the 2L, shift the Fanta mix toward cans in convenience — projected +0.9 pts margin.” Petar shares with the commercial director; the 2026 H2 plan uses the recommendation.

P&L: +0.5–1.5 pts gross margin. Formula: 1 pt × 50% addressable NSR (€5.3bn) = ~€53m directional; 20% capture = ~€10m/yr.

Start: 2 markets inside PVP/PAAT/OBPPC workstream

UC 56HoReCa Contract Renewal Prediction & JBP Prep Copilot

Don't Have — Can Build

Digital Consumer & Customer

What it is: For every chain on a multi-year contract, the agent predicts renewal risk 6 months out, generates a Joint Business Plan prep pack (3-year scorecard, trade spend effectiveness, category captaincy), drafts the KAM's renewal pitch.

Problem: JBPs prepared under time pressure; renewals won or lost in the pack. Renewals at risk spotted too late.

Tech: Agentforce + Databricks + Azure OpenAI + RAG over past JBP decks + Salesforce + SAP. Build 12–16 weeks.

📍 Practical exampleMonday 8:00am in Milan. National KAM Alessandro Conti opens his renewal dashboard: Autogrill contract is up in 9 months, renewal risk 54% (up from 32%) — trade-spend-to-volume ratio drifting unfavourably, two unresolved disputes, Pepsi has been in the account twice in the last quarter. The JBP prep pack is already drafted: a 12-slide deck with the 3-year scorecard, a proposed new bundle-pack, a mockup of a category-captain shelf reorganisation for their motorway stores. Alessandro refines 3 slides and ships.

P&L: -20–30% renewal loss. Formula: €1.5bn renewal base × 4% loss × 25% recovery = ~€15m/yr NSR retained.

Start: IT+AT (complex-chain exposure)

Domain 2 — Order-to-Cash & Customer Service

Cards 16–20 • 57

UC 16“Where Is My Order” Customer Service Agent

Have — Live

Digital Consumer & Customer

What it is: Conversational Agentforce agent in web, SIRVIS, WhatsApp. Handles “status of my order,” “rebook delivery,” “wrong SKU on invoice,” “credit note status” end-to-end using live SAP data.

Problem: These 4 questions are 60–75% of CCH Customer Care inbound.

Tech: Agentforce + MuleSoft to SAP + SF Service Cloud, 29-language.

📍 Practical exampleFriday 9:02am in Kyiv. A café owner named Oksana Koval messages CCH on WhatsApp in Ukrainian: “де моє замовлення, мало прийти вчора?” The bot reads the SAP delivery status, sees the truck was rerouted for fuel-station congestion, offers a same-day 2pm redelivery slot, and drafts a €15 goodwill credit. Oksana accepts in three taps. No human touched the ticket.

P&L: -30–50% L1 calls. Formula: 4m interactions × €5.20 × 35% deflection = €7.3m/yr OPEX + NPS lift.

Start: pilot existing template in 2 markets CCC

UC 17Order Entry Co-pilot for Tele-sales

Have — Template

Digital Consumer & Customer

What it is: Agent listens to the tele-sales call, captures spoken order against catalogue, pre-validates price/quota/stock, shows rep draft cart in real time.

Problem: Tele-sales error rates 5–15%; AHT 8–12 min.

Tech: Azure OpenAI realtime voice + CT Smart Order Assistant + SF Service Cloud. Deployment 4–8 weeks.

📍 Practical exampleTuesday 11:20am in Lviv. Tele-sales agent Natalia Bondar is on a call with a grocery in Ternopil. The store owner reels off “20 cases Coke 1.5L, 10 Sprite cans, 5 Burn 250ml, whatever Fuze Tea Lemon you have.” Natalia's screen fills the cart in real time; one line (Burn 250ml) is flagged out-of-stock with an auto-suggest swap to Burn Apple. She confirms verbally with the owner, clicks submit at call end. The whole call is 4 minutes instead of 9.

P&L: -25–35% AHT. Formula: 1,200 FTE × 4 min/call × 30 calls/day × 220 days ÷ 60 = ~528k hrs/yr capacity; €2–4m/yr error avoidance.

Start: UA or PL tele-sales desk

UC 18Invoice & Claim Dispute Resolution Agent

Don't Have — Can Build

Digital Enterprise / Digital Consumer

What it is: Customer disputes an invoice — agent pulls order, PoD, promo mechanic, pricing master, reconstructs correct invoice, proposes credit note for approval.

Problem: Biggest drag on CS NPS and AR days.

Tech: Azure OpenAI + RAG over SAP FI/SD + pricing rules + Power Automate. Build 10–14 weeks.

📍 Practical exampleWednesday 10:15am in Budapest. A Tesco Hungary category buyer emails the credit team: “Invoice HU-2026-118441 charged full price on Kinley Tonic which was under the Easter 3-for-2 promo — missing 47 cases of Fanta Grape.” By 10:19am the dispute agent has pulled the order, the POD photo showing 47 cases of Fanta on the dock, the Easter promo mechanic document, reconstructed the invoice, and queued a €1,840 credit note for the credit controller Zsuzsanna to approve. She approves at 10:22am.

P&L: -50–60% dispute time. Formula: 200k disputes × 3hr × €38 ≈ €22.8m → half = €11.4m/yr OPEX + €5–8m WC release.

Start: O2C “Releasing Oxygen” workstream

UC 19IOM Alerts & Touchless Exceptions

Don't Have — Can Build

Digital Enterprise

What it is: AI layer on top of CCH's IOM proactively spots at-risk orders and takes autonomous action — split-ship, pull substitute, notify customer via SIRVIS, open a case only if a human is needed.

Problem: Exceptions sit in a queue; DIFOT killer.

Tech: Agentforce + Databricks streaming + BlueYonder/SAP IOM. Build 14–20 weeks.

📍 Practical exampleThursday 4:30am in Abuja. A CCBA Nigeria delivery truck throws a brake-fault alert. The IOM exception agent sees that two of its 14 drops are time-critical (a hotel for a Friday conference, a supermarket awaiting a weekend promo). It split-ships those two to a nearby truck, notifies both customers via SIRVIS with a new ETA, opens a Service Cloud case only for the damaged-truck repair, and re-optimises the remaining 12 drops. Ops manager Adaeze wakes up to a 3-line summary at 7am.

P&L: +2–4 pts DIFOT. Formula: DIFOT +3 pts × €7bn flow × 0.15% margin = €10.5m/yr EBIT + €2–3m expedite.

Start: Next Gen OTC P1; RO + NG (CCBA)

UC 20Credit & Risk Scoring Agent

Don't Have — Can Build

Data, Insights & AI / Digital Enterprise

What it is: For new HoReCa customers and credit reviews, agent pulls registry, payment history, account behaviour, local econ; proposes limit + terms with written rationale.

Problem: Manual review 45–90 min; backlog slows onboarding → hurts Leads→Customers KBI.

Tech: Azure OpenAI + Databricks feature store + public data APIs + SAP FI. Build 10–12 weeks.

📍 Practical exampleMonday 9:30am in Cairo. Credit analyst Hossam El-Sayed has 48 new HoReCa applications in his queue. For applicant “Kebdet El Prince, Imbaba branch,” the agent pulls the Egyptian commercial registry record, 14 months of payment behaviour on the supplier-network pool, the local macro signal, and proposes a credit limit of EGP 85,000 with 14-day terms, citing three comparable outlets. Hossam reads the rationale, raises the limit by EGP 10k based on personal knowledge of the owner, approves. Total time: 4 minutes instead of 50.

P&L: -70% review time; -15–25% bad-debt. Formula: 50k reviews × 1hr × €45 = €2.25m/yr + €5m/yr bad-debt.

Start: EG (Pharos) or NG (CCBA)

UC 57Promotion Claim Settlement Agent

Don't Have — Can Build

Digital Enterprise

What it is: Distinct from fraud detection: handles the volume settlement of legitimate trade promo claims — reconciles the customer's claim doc against CCH's promo mechanic, scan-data, delivered volume, and ledger; proposes settlement amount.

Problem: Promotion claim settlement is a huge manual back-office load in bottling systems. 6–10 week cycles are common.

Tech: Azure OpenAI + CT AMS Context Memory pattern adapted to SAP FI/CO + Agentforce. Build 10–14 weeks.

📍 Practical exampleThursday in Warsaw. A Biedronka trade-claim bundle arrives: 217 claim lines across 14 stores, €48k total. The agent reconciles each line in 12 minutes (instead of 3 days), proposes 196 for auto-approval, 21 for human review. Back-office lead Tomasz approves the auto-batch and reviews the 21 before lunch.

P&L: -50–70% cycle time. Formula: €250m outstanding claims × 2-wk cycle acceleration = ~€10m/yr WC release + ~€1.5m/yr OPEX.

Start: PL+RO modern trade back-offices

Domain 3 — Data, MDM & Knowledge

Cards 21–25 • 58

UC 21Digital Data Steward (DDS) — Agentic MDM

Have — Live

Data, Insights & AI

What it is: Continuous AI-driven monitoring of master data. DDS agent scores every new/changed record, routes suspicious records to a Steward cockpit, proposes the fix; human approves.

Problem: CCH BP26 explicitly: “MDM Oxygen — agentic enabled scaled DQ” P1.

Tech: CT Digital Data Steward framework on Databricks + Azure OpenAI + SAP MDG/SF/Dataverse connectors + Steward cockpit UI. Live at global pharma customer.

📍 Practical exampleTuesday 11:00am in Sofia. Data steward Ivan Petrov opens the DDS cockpit. Top of the queue: a new material master “Coca-Cola Zero 0,5L PET, Bulgaria” created by a junior planner yesterday. DDS flags it as a 94% duplicate of an existing SKU with a different spelling (“Coca Cola Zero 500ml PET BG”), identical EAN, same pricing condition. DDS proposes “merge and archive new.” Ivan reads the 3-line explanation, confirms merge. Total elapsed: 35 seconds. Without DDS the record would have caused 12 failed sales orders before anyone noticed.

P&L: -30–50% time-to-triage. Formula: 30 stewards × €60k × 40% lift = €720k/yr + €4–7m/yr downstream avoidance.

Direct landing: MDM Oxygen Material MDG P1

UC 22Material/Customer MDM Anomaly Detection

Have — Live

Data, Insights & AI

What it is: ML anomaly detection (LSTM, Isolation Forest, ARIMA) cross-checks anomalies across sources — e.g., a stockout is ignored if correlated to planned promo depletion, escalated if vendor delay.

Problem: Static alerts = noise + alert fatigue.

Tech: Databricks ML + CT DDS; global pharma deployment uses same stack.

📍 Practical exampleFriday 6:47am in Prague. The pricing control tower throws an alert: “Sprite 2L Czech Republic list price dropped 14% overnight at 47 outlets.” The ARIMA model sees the drop correlates to a promo plan that went live at midnight — suppressed. Two minutes later, a second alert: “Kofola cooler placements increased 22% in Brno in the last 14 days.” No matching promo — escalated to RGM manager Petr. He opens the cockpit, sees the explanation, and walks into the 9am commercial meeting already knowing.

P&L: -60–80% false positives. Formula: 1-day faster reaction on €10.7bn NSR × 0.05% avoidance ≈ €5m/yr.

Start: pair with DDS on pricing master

UC 23Institutional Knowledge Base — CCH-Wide

Have — Live

Digital Employee

What it is: Custom RAG assistant ingesting SharePoint, Teams, emails, PPT, PDF, Excel, ServiceNow. Employees ask in natural language, get a cited answer. Multimodal, multilingual.

Problem: 33k employees, fragmented knowledge. BP26 “Hellen+” P1 — this is its knowledge layer.

Tech: Azure OpenAI + Azure AI Search + Graph DB + Docker/K8s; RBAC + guardrails. Live at global CPG beverage customer.

📍 Practical exampleWednesday 3:15pm in Chiasso. A new brand manager, Elena Rossi, is preparing her first price review for Lurisia. She types into the Hellen+ chat: “What were the last three price changes for Lurisia 0.75L glass in the Ticino channel, and what was the volume reaction?” The bot cites three Power BI snapshots, two emails from the previous brand manager, and one quarterly review deck, all with links. Elena has her answer in 40 seconds instead of a 2-hour SharePoint hunt.

P&L: 25–45 min/KW/day. Formula: 10k KWs × 30 min/day × 220 days × €50/hr = €55m/yr capacity; realistic 20% = ~€11m/yr.

Start: Sales Academy / Commercial Excellence → enterprise

UC 24SAP Institutional Knowledge for S/4HANA Migration

Have — Live

Digital Enterprise / Enabling Tech

What it is: Secure RAG agent trained on CCH's legacy SAP ECC docs, configs, tickets, change history. Functional analysts ask “why is the Romanian pricing procedure Z_RO_003 set up this way?” → decision rationale with traceability.

Problem: BP26 S/4 Upgrade P1. Biggest risk in S/4 migrations is tribal-knowledge loss.

Tech: CT SAP Knowledge Base solution (AIUC_44) + Azure OpenAI. Live at global brewer customer.

📍 Practical exampleThursday 10:20am in Sofia. SAP functional analyst Borislav Dimitrov is rebuilding the Bulgarian VAT pricing procedure in S/4. He types: “Why does ZPR0_BG_VAT call the custom routine 907 instead of standard 16?” The bot cites a 2019 change request from a former colleague who's since left, a regulatory circular from the Bulgarian Ministry of Finance, and the original approval email. Borislav learns in 90 seconds what would have taken a week of tickets and phone calls.

P&L: -25–40% analyst discovery time. Formula: 80 FTE × 30% lift × €110/hr × 1,700hrs = €4.5m/yr + €2–3m avoided rework.

Start: ingest SAP ECC corpus Q2; land inside SPEED

UC 25AI-Powered Institutional Content Ingestion (Policy & SOP)

Have — Live

Digital Employee

What it is: Auto-ingestion pipeline for CCH policy docs, playbooks, HR SOPs, regulatory circulars. Ingests from any source, de-dupes, summarises, flags changes, makes searchable; draft→review→publish workflow.

Problem: Outdated SOPs on file shares; nobody knows which version is current.

Tech: CT content-ingestion platform (AIUC_45) + Azure OpenAI + governance workflows.

📍 Practical exampleTuesday 2:00pm in Athens. Policy owner Despina Vasiliou uploads a new commercial-policy draft for the 2026 energy-drink guardrails. The pipeline flags that 11 paragraphs duplicate the 2025 policy, one clause contradicts the current Romanian promo rules, and three SharePoint folders reference the old version. Despina reviews the flags, publishes the new version, and the old one is automatically archived. BDs in Romania see the updated guardrail in Hellen+ within the hour.

P&L: compliance + policy-owner productivity. Formula: 20% reduction on ~40 FTE policy ops ≈ €0.8m/yr + risk reduction.

Start: Commercial Policy + Finance (SOX)

UC 58Translation & Localisation at Scale (29 Languages)

Have — Template

Digital Employee

What it is: Enterprise translation service on Azure OpenAI, grounded in CCH-specific glossary (brands never translated, legal preserved, POS terms correct) for employee / customer / field content across all 29 CCH languages.

Problem: CCH publishes in 29 languages. Today external-agency cost + 3–5 week cycles; low-resource languages skipped.

Tech: Azure OpenAI + CT Institutional Knowledge + glossary service + human-in-the-loop for legal. Build 6–8 weeks.

📍 Practical exampleTuesday in Zug. The Group CFO's all-hands recording needs to go out in 29 languages within 48 hours. The pipeline transcribes, translates, reviews, and publishes dubbed + subtitled versions overnight. A human reviewer in each country only has to validate the financial terminology — not translate from scratch.

P&L: -60–80% cost. Formula: CCH translation spend ~€4m/yr × 60% = ~€2.4m/yr.

Start: Corporate Comms + HR training

Domain 4 — GenAI for IT / Delivery

Cards 26–35 • 59

UC 26CT AI-Driven AMS (Context Memory Layer) — the cash lever

Have — Live

Enabling Technology

What it is: Persistent “Context Memory” MCP layer knowing CCH's exact Salesforce environment — every flow, ISV app, LWC, Apex class, past bug, decision. Claude Code + Memory Layer resolves tickets with full context instead of 80% on context archaeology. Senior architect approves every prod change.

Problem: “80% of billable hours = detective work”. CCH BP26 INC MTTR target 2 days (from 4).

Tech: MCP Memory + Claude Code + continuous QA pipeline. AI handles 95% L1, 70% L2, augments L3.

📍 Practical exampleMonday 6:14am in Chișinău. A CCH rep in Moldova raises a ticket: “CT Mobile crash when clicking Order tab after syncing.” By 6:17am the Context Memory layer has cross-referenced the crash stack against three similar tickets from 2024, identified a conflict with a recent ISV update on AccountTriggerHandler, drafted the fix, generated two unit tests and one Playwright E2E, and sent the PR to senior architect Oleg for review. Oleg approves at 7:30am, the fix deploys to sandbox by 8am. Under the old T&M model this would have been a 3-day ticket.

P&L: 40–60% AMS cost reduction, 5× MTTR, 95% L1 auto-resolution, zero key-person risk. Formula: CCH AMS spend ~€8–12m/yr × 40–50% reduction = €3.5–5.5m/yr.

Transform Valser, CPI Support, D365 AMS (~25 HC); tie to INC MTTR KBI

UC 27GenAI Requirements Validator

Have — Live

Enabling Technology

What it is: At PR time, tool compares submitted code against Jira requirements and reports % coverage. Catches half-built tickets before QA.

Problem: Requirements drift causes rework on Next Gen OTC-scale programmes.

Tech: CT GenAI Validator (AIUC_22) + Jira + GitHub/Azure DevOps + Claude/GPT-4.

📍 Practical exampleFriday 5:40pm in Warsaw. Developer Marcin Kowalski raises a PR on CCH-NGOTC-8812 “Add partial-delivery acceptance to cart.” The validator runs at 5:41pm: 4 of 6 acceptance criteria met; 2 missing — “display remaining delivery window” and “allow cancel on remainder.” The PR is blocked. Marcin groans, fixes both before he leaves for the weekend, gets clean approval at 6:12pm. Under the old process those two gaps would have surfaced in UAT three weeks later.

P&L: -15–25% rework. Formula: €1.2m/release × 6 waves × 20% = €1.44m/yr.

Start: Next Gen OTC delivery pipeline

UC 28GenAI Code Validator (PMD / Security / Compliance)

Have — Live

Enabling Technology

What it is: AI PMD scanner validates against PMD rules, security, compliance, style; proposes or auto-applies fixes.

Problem: Manual static-analysis enforcement slows every PR.

Tech: CT GenAI Validator suite (AIUC_23) + Azure OpenAI + CI/CD.

📍 Practical exampleTuesday 9:05am in Kraków. Developer Agnieszka Wójcik pushes a commit on the Valser Service repo containing an inefficient SOQL inside a for-loop and a missing CRUD check on AccountSelector. The validator comments on the PR with the exact PMD rule references, proposes the two fixes as a one-click patch. Agnieszka accepts, the PR turns green 90 seconds after the push.

P&L: -70–80% review toil. Formula: 150 devs × 3 hr/wk × 45 wks × €80/hr × 70% = €1.1m/yr.

Start: bundle with UC 26 AMS transform

UC 29GenAI Unit Test Validator

Have — Live

Enabling Technology

What it is: AI generates @isTest classes with assertions and setup, validates existing. “40–60% time savings (1–2 hours → 20–40 min per class).”

Problem: Unit-test coverage chronically under-invested; SF deploys fail because of it.

Tech: CT GenAI Validator (AIUC_24) + Claude Code.

📍 Practical exampleWednesday 1:30pm in Kyiv. Developer Dmytro Melnyk needs to ship an Apex class “PriceBookSyncBatch” but has no unit test. He runs the validator CLI; in 90 seconds he has a 92%-coverage test class with 11 assertions and 3 setup scenarios — positive path, bulk, negative. He spends 10 minutes reading it instead of 90 minutes writing it.

P&L: -50% authoring; +10–20 pts coverage. Formula: 200-class wave × 150 hrs × €95/hr = ~€14k/wave × 10/yr = €140k/yr/programme.

Start: CT AMS + next CT Mobile wave

UC 30GenAI Digital Worker — Developer

Have — Live

Enabling Technology

What it is: Jira ticket → AI generates code, opens PR, runs auto-fix, drives merge/deploy. Senior architect gates every production change.

Problem: For well-specified small tickets, spec→code→PR is 80% mechanical.

Tech: Claude Code + CT Context Memory + Jira/GitHub.

📍 Practical exampleMonday 10:00am. Ticket CCH-AMS-4417 “Add 'preferred delivery slot' picklist to Account, visible to BDs in Greece + Cyprus only.” The Digital Worker writes the field, the permission sets, the validation rule, the translation bundle (Greek, English), the unit test, and opens the PR — all by 10:09am. Senior architect Elif reviews for 5 minutes and approves.

P&L: 2–3× throughput on small tickets. Formula: 40% of volume × 2× throughput reallocates ~€1.5–2m/yr.

Start: CT AMS pod on Valser / CPI Support

UC 31GenAI Digital Worker — QA Web & Mobile

Have — Live

Enabling Technology

What it is: TestRail manual cases executed by AI (Claude) simulating human input — screenshots, video, error retries, Allure reports, TestRail auto-status — for web + mobile (iOS/Android).

Problem: Regression cycles on CT Mobile + Salesforce take weeks.

Tech: CT Digital Worker GenAI QA (AIUC_27/29) + Copado Robotic (AIUC_08) + Claude Code.

📍 Practical exampleThursday night 11pm in Limassol. The CT Mobile 2026.3 release candidate drops. The QA Digital Worker starts executing 420 regression cases — on iOS, on Android, on the web console — simultaneously. By 6am it has a full Allure report: 404 pass, 13 fail (all repro'd with video), 3 flakes retried and passed. QA lead Maria reviews the 13 fails over her first coffee; under the old process this was a 9-day cycle.

P&L: -60–75% regression time. Formula: 4 waves × 4 wks × 6 QA FTE × €65/hr × 70% = ~€580k/yr.

Start: CT Mobile regression for CCH rollouts

UC 32GenAI Code Quality Dashboard

Have — Live

Enabling Technology

What it is: Continuous static + dynamic code analysis; multi-parameter 1–5 scoring (security, complexity, maintainability); historical trends; AI hardening recommendations.

Problem: No objective, current view of delivery health across a big portfolio.

Tech: CT Digital Worker (AIUC_28) + Azure OpenAI + Grafana/Power BI.

📍 Practical exampleFriday 8:30am in Belgrade. CTO Ivan Jovanović opens the dashboard on his Teams morning-digest card. The CT Mobile Greek fork has dropped from 4.2 to 3.6 on maintainability over the last 6 weeks — 11 new Apex classes, no test coverage on two. The dashboard's AI recommendation: “schedule a 1-day tech-debt fix on BR-GR-2026 before the May release.” Ivan tags the release manager.

P&L: -20% incident rate. Formula: soft; 1-day MTTR reduction contributing to BP26 MTTR 2-day KBI.

Start: all CT-owned CCH repos as governance

UC 33GenAI Tasks-from-Requirements & Requirements Update

Have — Live

Enabling Technology

What it is: Reads a business doc and auto-creates Jira Epic + child tasks with hierarchy and naming. One-click rewrites messy ticket descriptions into structured actionable text.

Problem: “Grooming tax” — 20–30% of a BA's week converting prose to tickets.

Tech: CT GenAI Validator (AIUC_39/40) + Jira plugin.

📍 Practical exampleMonday 11:15am in Dublin. BA Ruairí Ó Briain receives a 14-page Word doc from the commercial team titled “Ireland 2026 On-Trade Rebate Mechanics.” He drops it into the AIUC_39 tool. Out come one Epic, seven Stories, 23 Tasks, neatly worded and pre-estimated, ready for backlog grooming. Ruairí edits 4 titles, accepts the rest — 30 minutes of work instead of two days.

P&L: -25% BA grooming. Formula: 20 BAs × 10 hr/wk × 45 wks × €70 = €630k/yr.

Start: BP26 BAs on Next Gen OTC + Promo360

UC 34GenAI Test Data Generator

Have — Live

Enabling Technology

What it is: Generates realistic test data scripts aligned with CCH's schema and validation rules (customers, orders, pricing, promos), bulk-capable. Replaces “copy from prod and anonymise.”

Problem: Test-data prep is a quiet time-sink and a GDPR risk.

Tech: CT GenAI Validator (AIUC_41).

📍 Practical exampleTuesday 4:00pm in Bucharest. QA engineer Andreea Popescu needs 1,200 realistic Romanian HoReCa accounts, with orders across 8 SKUs, 3 promo types, valid VAT numbers in the Romanian format, for the Next Gen OTC UAT. Run the generator: 90 seconds later her sandbox has the data. Zero production PII.

P&L: -50–70% test-prep. Formula: 10 programmes × 200 hrs × 60% × €65 = €78k/yr + compliance risk reduction.

Start: QA pipeline on CT AMS scopes

UC 35Copado Robotic Test Automation for Salesforce

Have — Live

Enabling Technology

What it is: Automatically builds and maintains Salesforce regression test suites; impact analysis, metadata dependency mapping, auto-doc from real-time metadata.

Problem: CCH's Salesforce estate carries years of legacy config; every release is regression risk.

Tech: Copado Robotic (AIUC_08) + CT wrapper + CT AMS Context Memory.

📍 Practical exampleWednesday 3:00pm in Athens. Release manager Yannis Karagiannis kicks off the Valser Service March release. Copado Robotic runs the full regression in 45 minutes instead of 2 days. The impact-analysis report shows only 18 of 310 test cases touched the changed metadata; the rest can be skipped with confidence. Yannis signs off at 3:50pm.

P&L: -70% authoring; -80% defect escape. Formula: marginal on top of UC 31 ≈ €300–500k/yr SF-specific.

Start: CT Mobile + Valser Service regression

UC 59Cybersecurity Incident Triage (AI-SOC)

Don't Have — Partner

Enabling Technology

What it is: AI layer over CCH's SIEM (Sentinel) triages alerts — classifies severity, correlates related alerts, proposes containment, drafts incident-response playbook. Human gates every containment.

Problem: SOC teams drown in false positives. Alert fatigue is the #1 reason real incidents take hours to escalate.

Tech: Microsoft Sentinel + Defender XDR + Azure OpenAI Copilot for Security + CT integration & runbook build. Core is Microsoft; CT integrates.

📍 Practical exampleSaturday 2:47am in Sofia. The SIEM fires 18 related alerts in 4 minutes — unusual egress traffic from a build server in Athens. The AI-SOC agent correlates them to a single potential incident, suggests isolating the server, drafts the incident ticket, and pages the on-call analyst Boyan with the 5-line summary. Boyan confirms isolation at 2:51am. Under the old process the analyst would have been sifting 18 separate alerts.

P&L: -40–60% triage time. Formula: avoided breach exposure + ~€0.8m/yr SOC productivity.

Start: CCH Group SOC + Sofia/Athens DCs

Domain 5 — Manufacturing, Supply Chain & Planning

Cards 36–43 • 60–64

UC 36Demand Forecasting & Replenishment Copilot

Don't Have — Can Build

Data, Insights & AI

What it is: Hybrid ML + LLM producing SKU-location-day forecasts blending sell-in, sell-out, weather, events, promos, macro. LLM layer explains forecast changes in plain language.

Problem: Forecast error drives 2–4% NSR loss. Planner attention is the bottleneck.

Tech: Databricks + Prophet/LightGBM + Azure OpenAI + SAP IBP/BlueYonder integration. Core engine is partner (IBP/BlueYonder); CT builds the explanation/copilot layer. Build 16–20 weeks/market.

📍 Practical exampleMonday 8:00am in Vienna. Demand planner Stefan Gruber asks the copilot: “Why did the Austrian Coke Zero 1.5L forecast drop 9% for next week?” Answer in 3 seconds: “Vienna Marathon finished Sunday (+11k cases LY one-off), no Bundesliga home fixtures, temperature -4°C vs. LY 2025 week 17, current promo ending Thursday. Net: -9.2%, aligned with historical elasticity.” Stefan accepts the forecast, closes the meeting 20 minutes early.

P&L: +15–30% accuracy. Formula: €10.7bn NSR × 0.5% recovery = €53m/yr directional.

Start: 1 Established + 1 Developing, tie to BlueYonder

UC 37Predictive Maintenance for Production Lines & Coolers

Don't Have — Partner

Data, Insights & AI / Digital Enterprise

What it is: Sensor-driven model predicts failures 24–72 hours ahead and opens preventive work order. LLM answers root-cause questions with evidence ranking.

Problem: BP26 P1 Manufacturing 4.0 + Connected Coolers. Unplanned downtime 5–12% of prod time.

Tech: Azure IoT Hub + Databricks ML + Azure OpenAI. OT integration (PI historian, Siemens Mindsphere, Aveva) sits with plant-engineering partners; CT plays orchestration + ML + LLM explanation.

📍 Practical exampleTuesday 5:40am in Plant Sofia, Bulgaria. Maintenance supervisor Krasimir Todorov is finishing his night round when the predictive model opens a work order: “Line 3 filler bearing — vibration anomaly trending +38% over 6 hrs, 94% probability of failure within 22 hrs, rec. replace bearing and shaft seal, parts in stock bay C-12, estimated downtime 45 min vs. 6.5 hrs on unplanned failure.” Krasimir schedules it into the 11am planned stop.

P&L: -20% downtime. Formula: €60m output × 8% downtime × 50% monetisable × 20 plants = ~€48m/yr.

Start: 2 priority plants inside Manufacturing IIOT BP26

UC 38Route Optimisation & DSD Exception Handling

Don't Have — Partner

Data, Insights & AI / Digital Enterprise

What it is: Optimises DSD routes daily, re-routes live on exceptions, notifies customers via SIRVIS. LLM explains decisions to dispatchers.

Problem: BP26 P2 Dynamic Routing. KOF is at 90% digital-route coverage.

Tech: BlueYonder is the CCH-selected planning engine; CT builds the SIRVIS notification layer + LLM dispatcher copilot on top.

📍 Practical exampleWednesday 6:00am in the Timișoara depot. Dispatcher Alina Dumitrescu has 14 trucks and 312 drops. The engine rebalances overnight given the E70 road closure west of Arad: two trucks swap territories, one drop moves to tomorrow with an auto-SIRVIS notification to the customer, the full re-plan is ready before Alina's coffee. The LLM layer answers her question “why did we move Băcălia Maria from Truck 4 to Truck 7?” in plain Romanian.

P&L: -10–12% transport cost. Formula: €500m dist cost × 10% = €50m ceiling; realistic pilot = ~€5m/yr.

Start: 1 Developing market high-fragmentation

UC 39Trade Promotion Optimisation (TPO)

Don't Have — Can Build

Data, Insights & AI

What it is: Uplift model estimates incremental volume for each promo mechanic per SKU per channel, recommends optimal mechanic for a target objective. Integrated with Promo360.

Problem: 40–60% of promos are margin-negative. BP26 Promo360 P1.

Tech: Databricks causal ML (double ML/uplift trees) + Promo360 + CT Mobile. Build 16–20 weeks.

📍 Practical exampleThursday 2:00pm in Rome. RGM analyst Valeria Esposito is planning the May Sprite Zero promo. The TPO copilot compares 4 mechanics for convenience channel: “-15% price cut: +8.1% volume, margin -€1.2m. 3-for-2: +14.2% volume, margin +€0.4m. Bundle-with-Fanta: +11.7% volume, margin +€0.9m. Digital coupon: +6.3% volume, margin +€1.1m.” Valeria picks the bundle, the copilot pushes it to Promo360 and to BD pre-visit briefs.

P&L: +10–20% promo ROI. Formula: €700m promo spend × 15% = ~€100m ceiling; realistic capture 10% = ~€10m/yr.

Start: Promo360 rollout + PVP/PAAT/OBPPC P1

UC 40Energy Optimisation for Plants

Don't Have — Partner

Data, Insights & AI / ESG

What it is: Model-predictive control layer on plant energy (chillers, compressors, blow-moulding) optimises set-points in real time against price + carbon signal. Plant manager sees explained recommendation.

Problem: Energy is a top-3 plant cost line; ESG needs measurable reductions.

Tech: MPC on SCADA is Siemens/Schneider/Honeywell territory. CT builds the Azure + Databricks + LLM explanation layer and orchestration.

📍 Practical exampleMonday 3:00am in Plant Edelstal, Austria. The MPC layer sees the Austrian electricity day-ahead price drops €84/MWh between 03:00–05:00. It ramps the chiller pre-cool and the PET blow-moulding compressor, then throttles them back during the evening peak. Plant manager Markus Wagner sees in the morning a 6% reduction on the day and €3,800 saved. The LLM narrative in the morning report explains exactly why.

P&L: -10–18% energy. Formula: €180m energy bill × 12% = ~€21.6m/yr.

Start: co-land with Manufacturing 4.0 BP26 P1

UC 41Quality Control Computer Vision on the Bottling Line

Have — Template

Data, Insights & AI

What it is: Cameras on filler/capper/labeller run a CNN flagging defects in real time and diverting defective units. Continuously retrains from human confirmations.

Problem: Research: “AI image recognition reduces defects by 18%” and “recalls by 30%.”

Tech: Edge GPU + CT Vision (shared lineage with store-check) + Azure ML ops. Build 8–12 weeks per line.

📍 Practical exampleFriday 10:22am in Plant Lamia, Greece. Line 2, the 330ml aluminium can filler. A camera catches a dent on a can at 48,000 cans/hr and a pneumatic kicker diverts it in 40 milliseconds. QC lead Nikos Antoniadis sees on the dashboard that the dent pattern has appeared 6 times in the last hour, all at the same station — and the model suggests a mis-aligned infeed guide. Nikos calls maintenance.

P&L: -15–20% defect rate. Formula: €6m/plant × 18% × 20 plants = ~€21.6m/yr.

Start: one plant PoC inside Manufacturing 4.0

UC 42Customer Contract Simulation & What-If P&L

Don't Have — Can Build

Digital Enterprise / Data, Insights & AI

What it is: On-demand simulator for KAMs & RGM: runs “what-if” scenarios through cost-to-serve, gross margin, working capital, SIRVIS adoption and produces a 1-pager.

Problem: Contract discussions happen without a live P&L — gut decisions.

Tech: Databricks financial model + Azure OpenAI UX + Salesforce. Build 12–16 weeks.

📍 Practical exampleWednesday 4:30pm in Dublin. KAM Órla Ní Riain is in a SuperValu contract negotiation. The buyer asks “what if we moved from 4% to 6% rebate on volumes above 400k cases and added the Fanta 1.75L exclusivity?” Órla types the parameters; 8 seconds later the 1-pager shows year-1 margin -0.7 pts, year-2 +1.3 pts, 3-year NPV positive by €1.1m, working capital impact -€240k on the new delivery cadence. She counter-offers with confidence.

P&L: +0.3–0.8 pts margin. Formula: €3bn portfolio × 0.5 pt = ~€15m/yr.

Start: Customer Contract Management BP26; land with UC 14

UC 43Field Service Virtual Agent for Connected Coolers

Don't Have — Can Build

Digital Consumer & Customer / Enabling Tech

What it is: Cooler throws an IoT alert; agent diagnoses from telemetry + manuals, attempts remote reset, dispatches tech with prefilled work order + parts list, notifies outlet via SIRVIS with ETA.

Problem: BP26 P1 Connected Coolers IoT Hub. Parts wrong on first visit → second visit.

Tech: Azure IoT + Databricks + Agentforce + CT Mobile for tech; RAG on manuals. Build 12–16 weeks.

📍 Practical exampleSaturday 11:55am in Budapest. A Frigoglass cooler at a BP petrol station on Andrássy Avenue drifts to +9°C. The agent reads the telemetry, sees the compressor is healthy but defrost is stuck, tries a remote defrost cycle — temp holds. It dispatches technician László Kovács with a prefilled work order listing the likely defrost timer part, ETA 14:30, and sends the station manager a SIRVIS message. László arrives with the right part, fixes it in 20 minutes. No second visit.

P&L: -25–40% second-visit rate. Formula: 250k incidents × 30% improvement × €85 = ~€6.4m/yr.

Start: Connected Coolers IoT Hub BP26 P1

UC 60Sustainability / Scope 1-2-3 Footprint Copilot

Don't Have — Partner

Digital Enterprise / ESG

What it is: For every CCH SKU and customer contract, the copilot computes a Scope 1/2/3 carbon footprint grounded in plant energy, sourcing, logistics, packaging data. Teams can ask for footprint by SKU+lane.

Problem: ESG reporting and modern-retail carbon requests are manual. CSRD/CBAM tightening.

Tech: Core carbon accounting is Watershed/Persefoni/SAP Green Ledger territory. CT builds data integration + LLM explanation + cockpit layer on top.

📍 Practical exampleThursday 11:00am in Zug. Sustainability lead Barbara Huber gets a carbon-data request from a major German retailer ahead of a tender. She asks the copilot: “per case footprint of Coke Zero 1.5L PET produced in Edelstal, delivered to Frankfurt RDC.” Answer in 8 seconds: Scope 1 0.24 kgCO₂e, Scope 2 0.18 (grid-weighted), Scope 3 upstream PET 0.41, Scope 3 logistics 0.09, total 0.92 kgCO₂e — with a methodology citation. Barbara exports, signs, sends.

P&L: CSRD compliance + tender-win uplift. Formula: 0.5% tender-win on €2bn carbon-weighted = ~€10m/yr directional.

Start: Top-20 modern-trade customers' data requests

UC 61Returnable Glass Bottle Tracking & Asset Intelligence

Don't Have — Can Build

Data, Insights & AI / Digital Enterprise

What it is: ML + IoT layer tracking returnable glass bottles (still significant in EG/NG/RO/RS) from plant → outlet → return. Predicts loss hotspots, shrinkage rates, optimal deposit economics. Extension to branded coolers.

Problem: Returnable asset loss in emerging markets is 5–15% yearly shrinkage — millions of euros vanish.

Tech: Databricks + optional IoT (QR at case level, cooler GPS) + Agentforce + CT Mobile audit. Build 10–14 weeks.

📍 Practical exampleTuesday in Cairo. Asset manager Mohamed Farid sees the dashboard flag that bottle returns from the Giza cluster have dropped 11% vs. LY, concentrated in 24 outlets. The agent proposes a targeted field audit route for the merchandising team. Mohamed dispatches; 9 of the 24 outlets are found to be hoarding empties (as a cash reserve). The deposit economics are re-explained on the spot.

P&L: -30–50% shrinkage. Formula: €80m fleet × 10% shrinkage × 35% reduction = ~€2.8m/yr.

Start: EG + NG (largest returnable base)

UC 62Driver Safety & Behaviour Scoring

Don't Have — Partner

Digital Enterprise / HR

What it is: Telematics + AI scoring drivers on harsh braking, speeding, cornering, idling, fatigue; weekly coaching report + real-time in-cab alert.

Problem: Fleet accident rates are a direct cost line + CSR concern + fuel waste.

Tech: Telematics is partner (Geotab / Samsara / Webfleet). CT builds the scoring layer + Power BI + CT Mobile driver UI.

📍 Practical exampleFriday afternoon in Dublin. Transport manager Aoife Byrne gets the weekly driver-score report. Driver Conor Lynch has the second-worst harsh-braking score on the fleet and a 14% worse fuel efficiency than his peer average. Aoife does a 15-minute coaching sit-down on Monday, shows the score trends, the 3 specific routes where the braking events cluster. Three weeks later Conor is in the top quartile.

P&L: -15–25% incidents; -3–6% fuel. Formula: 6k trucks × €3k avoidable fuel × 5% = ~€0.9m/yr fuel + €2–4m/yr insurance.

Start: IE + AT (cleanest fleet data)

UC 63Procurement Spend Copilot

Don't Have — Can Build

Digital Enterprise / CFO agenda

What it is: Agent over procurement spend data (SAP MM, Ariba) that answers spend questions in natural language and proactively flags anomalies — maverick spend, price drift, duplicate suppliers, off-contract.

Problem: Procurement analytics today is a quarterly SAP BW report; insights stale and action slow.

Tech: Azure OpenAI + Databricks + SAP MM + Ariba + Agentforce. Build 10–14 weeks.

📍 Practical exampleMonday 9:00am in Sofia. CPO Bojidar Ivanov asks the copilot: “What's happened to our PET preform spend in Bulgaria over the last 6 months?” Answer in 6 seconds: volume +8% on plan, average price +11.4% vs. contract, supplier concentration shifted from 60/40 to 78/22 in favour of the local supplier after a single buyer started over-indexing. The drift is worth €340k / 6 months. Bojidar schedules a review.

P&L: 1–3% addressable savings. Formula: €1.5bn indirect × 1.5% = ~€22m directional; realistic pilot €2–4m/yr.

Start: Group procurement spend analytics pilot

UC 64Warehouse Slotting & Pick Optimisation

Don't Have — Partner

Digital Enterprise

What it is: ML model continuously re-slots depot warehouses based on velocity, seasonality, promo plans, picker ergonomics. Pickers see an optimised path on a handheld.

Problem: Static slotting wastes picker time; promo peaks clog pick-faces.

Tech: WMS (SAP EWM / Manhattan) is partner. CT builds the optimiser overlay + Power Platform front-end.

📍 Practical exampleWednesday in Athens. The Aspropyrgos depot is re-slotting for the Easter promo. The optimiser proposes moving Fanta 1.5L from aisle 14 to aisle 3, Coca-Cola glass 0.25L from aisle 7 to aisle 4. Depot manager Yannis runs the simulation — picker travel -1.2 km/shift. He approves and the re-slot happens over the weekend.

P&L: -10–15% picker time. Formula: 80 depots × €300k pick labour × 12% = ~€2.9m/yr.

Start: 2 high-volume depots GR+IT

Domain 6 — Digital Employee & Cross-Functional

Cards 44–48 • 65–66

UC 44Hellen+ Employee Virtual Agent (HR / P&C Service Desk)

Have — Live

Digital Employee

What it is: Employee-facing virtual agent (“Hellen+” BP26) handling payroll, leave, policy, IT triage, expenses in natural language. Integrates Workday + ServiceNow + SAP HCM.

Problem: BP26 P1 Digital Employee: “P&C Service Desk and new virtual agent capabilities on Refresh for BDs & Operators.” 33k employees.

Tech: Azure OpenAI + CT Institutional Knowledge (AIUC_10) + Workday/ServiceNow/SAP HCM connectors; multi-language.

📍 Practical exampleMonday 9:40am in Lagos. A merchandiser, Adaeze Nwosu, messages Hellen+ in English: “How many leave days do I still have and can I take next Friday?” The bot checks Workday, sees 11 days remaining, notes that her team lead has approved holidays up to 3 concurrent absences next week and only 1 is booked, drafts the leave request for her approval. Adaeze taps submit. Total elapsed: 25 seconds.

P&L: -35–45% L1 volume. Formula: 300k tickets/yr × €12 × 40% deflection = €1.4m/yr + employee wait-time savings.

Direct landing: BP26 P&C Service Desk + Hellen+

UC 45GenAI Meeting Notes Refinement & Action-Item Capture

Have — Live

Digital Employee

What it is: Free-form meeting / call / visit notes get structured, summarised, and turned into tasks + reminders. Typed or voice-transcribed.

Problem: Meetings create notes nobody re-reads; action items slip.

Tech: CT GenAI Validator (AIUC_42) + Teams, Outlook, SF activities.

📍 Practical exampleFriday 4:30pm in Zug. Commercial director Lisa Keller finishes a 90-minute S&OP review on Teams. The plugin produces: 6 decisions, 11 actions with owners and dates, 4 risks, a 10-line summary, and drops everything into Planner + Outlook invites. Lisa reviews in 4 minutes, hits send.

P&L: org-wide capacity. Formula: 5,000 KWs × 15 min/day × 220 days = 275k hrs/yr.

Start: Commercial Excellence + BP26 PMO

UC 46AI-Enabled Internal Audit & SOX Controls

Have — Template

Digital Enterprise

What it is: Continuous controls monitoring with agentic layer reading SAP FI/CO journals, flagging control exceptions (SoD, unusual patterns, threshold breaches), writing explanation + remediation, auto-preparing the SOX workpaper.

Problem: BP26 P1 “AI enabled audit.” SOX testing today is sampled and reactive.

Tech: Databricks + Azure OpenAI + SAP FI + Power Automate; AIUC_03 DDS pattern on financial data.

📍 Practical exampleThursday 7:00am in Bucharest. Internal auditor Raluca Ionescu opens the cockpit. Overnight, the agent scanned 117,000 SAP FI journals across 5 CCH markets and flagged three items: a €48k manual journal posted and approved by the same user in Hungary (SoD breach), a pattern of vendor invoices posted just under the €10k approval threshold in Romania, and a duplicate payment to a Bulgarian supplier. Each with a drafted workpaper. Raluca triages all three before 8am.

P&L: avoided SOX penalties. Formula: Takeda: “$529k/yr at 1 market; €5.3–10.6m at 10–20.” CCH 29 markets: ~€10–15m/yr directional.

Co-build with Internal Audit + iGRC BP26 P1

UC 47RPA-to-Power-Automate Intelligent Migration

Have — Live

Enabling Tech

What it is: Already live at CCH. Migrate UIpath RPA bots to Power Automate with AI-assisted migration; refactor UI-automation flows into API calls where available; layer AI decisioning.

Problem: UIpath licensing + stale flows. Current CCH scope = 4 HC.

Tech: Power Automate + Power Platform + CT AI-assisted migration accelerators.

📍 Practical exampleOngoing, Sofia + Bucharest. CT engineer Elitsa Dimitrova is migrating a UIpath bot that downloads SAP reports, parses them in Excel and emails the commercial team. The accelerator reads the UIpath XAML, identifies that 7 of 9 steps can be replaced by direct SAP OData calls, drafts the Power Automate flow. Elitsa tests it in a day instead of a week; the licensing line drops off at year end.

P&L: -40–60% licensing + ~€500k/yr OPEX per 20-bot tranche. Formula: Takeda: $600k/yr per market × 5 realistic markets = ~$3m/yr.

Already in flight; expand scope to additional markets

UC 48AI-Powered Internal Search for Field Managers

Have — Live

Digital Employee

What it is: Narrow-scoped variant of institutional KB focused on commercial playbooks, pricing guardrails, promo mechanics, RED photo standards — tuned for BDs and ASMs with mobile-first UX.

Problem: BDs in CT Mobile today can't find commercial docs fast — they call HQ.

Tech: AIUC_05/AIUC_10 pattern narrowly scoped + CT Mobile embedded UI.

📍 Practical exampleTuesday 2:45pm in Crete. BD Manolis Konstantinou is in a hotel bar in Rethymno negotiating a 2026 listing. The owner pushes for a 12% discount on Avra 0.5L glass. Manolis types “2026 Avra 0.5L HoReCa Greece price guardrail” into his CT Mobile search. Answer in 2 seconds: “Max discount 8% without ASM sign-off, 11% with, documented in the Greek 2026 Commercial Policy v3.2 p.14.” Manolis counter-offers 8% on the spot.

P&L: sharper negotiations. Formula: 10 min/BD/day × 2,500 BDs = ~100k hrs/yr = ~€4m/yr capacity.

Start: Sales Academy Metaverse cohort

UC 65Employee Engagement & Attrition Copilot

Have — Template

Digital Employee / HR

What it is: CHRO-facing agent reads engagement survey data, Teams sentiment proxies (with governance), exit-interview text, training participation, manager-span data to identify engagement hotspots and predict attrition risk by team.

Problem: Engagement data is annual; attrition is noticed after it happens. BP26 Digital Employee P1.

Tech: Databricks + Azure OpenAI + Workday + Glint/Qualtrics; aggregate-only guardrails. Build 10–12 weeks.

📍 Practical exampleMonday morning in Zug. CHRO Marta Lopes opens her weekly heatmap. The Bulgaria CCBA-operations team has dropped 11 points on “belonging” since Q1, attrition risk climbed to 14% vs. 6% group average. The agent flags that three key supervisors have left in the last quarter and the team's spans of control have doubled. Marta flags it to the Bulgaria P&C lead for a targeted intervention.

P&L: -10–20% regretted attrition. Formula: 8% voluntary × 33k FTE × 30% regretted × €15k replacement = €11.9m baseline × 15% = ~€1.8m/yr.

Start: one country P&C pilot, 2026 H2

UC 66Legal / Contract Review & Litigation Support Copilot

Have — Template

Digital Enterprise / Legal

What it is: Legal-team copilot reviews inbound contracts (supplier NDAs, HoReCa, M&A docs) against CCH's playbook, flags deviations, drafts redlines, and summarises litigation-risk documents during e-discovery.

Problem: In-house legal teams spend 30–50% of time on contract and document review — high-volume, pattern-based.

Tech: Azure OpenAI + CT Institutional Knowledge (AIUC_05) + document comparison + SharePoint legal corpus. Build 8–12 weeks.

📍 Practical exampleWednesday 3:00pm in Athens. Legal counsel Eleni Kouris receives a 47-page MSA from a new IoT cooler supplier. She drops it into the copilot, which returns in 90 seconds: 11 deviations from CCH's supplier playbook (3 critical — indemnity cap, IP ownership, data residency; 8 minor), a proposed redline for each, and two market-precedent clauses from prior CCH deals. Eleni accepts 9 redlines, modifies 2, sends back the same afternoon.

P&L: -30–50% review cycle. Formula: €15m outside counsel × 15% = €2.25m/yr + in-house productivity €1m/yr.

Start: Group Legal, supplier NDAs + HoReCa master agreements

CCH / CT Strategic Review

Extreme Efficiency with AI
for Top Line & Bottom Line P&L

CCH has publicly committed to "AI Everywhere — Agentic". Customertimes has the agents, the data patterns, and 53 people already embedded.

1. Consumer affordability & downtrading

2. FX & emerging-market volatility

3. Cyber & AI-weaponised attacks (trend: increasing)

4. Business interruption / supply disruption (increasing)

5. Talent attraction, retention & digital skills shift

6. Packaging & commodity cost inflation

7. CCBA integration — biggest single-project exposure Flagship hook

8. Omnichannel retail disruption

9. Product regulation & sugar/DRS/EUDR compliance

10. Climate & water stress on 27 plants

11. Legacy tech debt & SAP 2027 end-of-support

12. Customer waiting time on contracts

Reading the status labels

Domain 1 — Commercial & Sales

Domain 2 — Order-to-Cash & Customer Service

Domain 3 — Data, MDM & Knowledge

Domain 4 — GenAI for IT / Delivery

Domain 5 — Manufacturing, Supply Chain & Planning

Domain 6 — Digital Employee & Cross-Functional

CCH / CT Strategic Review

Extreme Efficiency with AIfor Top Line & Bottom Line P&L

CCH has publicly committed to "AI Everywhere — Agentic". Customertimes has the agents, the data patterns, and 53 people already embedded.

1. Consumer affordability & downtrading

2. FX & emerging-market volatility

3. Cyber & AI-weaponised attacks (trend: increasing)

4. Business interruption / supply disruption (increasing)

5. Talent attraction, retention & digital skills shift

6. Packaging & commodity cost inflation

7. CCBA integration — biggest single-project exposure Flagship hook

8. Omnichannel retail disruption

9. Product regulation & sugar/DRS/EUDR compliance

10. Climate & water stress on 27 plants

11. Legacy tech debt & SAP 2027 end-of-support

12. Customer waiting time on contracts

Reading the status labels

Domain 1 — Commercial & Sales

Domain 2 — Order-to-Cash & Customer Service

Domain 3 — Data, MDM & Knowledge

Domain 4 — GenAI for IT / Delivery

Domain 5 — Manufacturing, Supply Chain & Planning

Domain 6 — Digital Employee & Cross-Functional

Extreme Efficiency with AI
for Top Line & Bottom Line P&L