A governed portfolio of AI Agents for Kazyon’s next phase of operating leverage.
The proposed model assigns a Director AI Agent to each operating domain. Each Director AI Agent coordinates focused AI Agents that monitor defined signals, recommend interventions and, where policy and integration allow, execute human-approved actions with an audit trail. Investment can be staged by domain and measured against a pre-agreed baseline. ANCHOR provides the shared data, governance and integration architecture needed as the portfolio scales.
Computer vision applications
Six proposed AI Agents use computer vision or image-based OCR.
These capabilities use existing CCTV, staff-phone cameras and dock cameras to convert visual activity into evidence, alerts and human-approved actions across shrink, store execution, receiving and merchandising.
LENS · FLOW · TAG · BLUEPRINT · HARVEST · GATE
Proposed capability map
Six Director AI Agents and fifteen domain AI Agents
The portfolio is modular. Each domain can be assessed independently, while the full model creates a coordinated decision layer across the business. Expand a domain to review its mandate, AI Agents, implementation requirements and KPIs.
Security Director AI AgentSENTRY · Computer vision
- Shoplifting Detection Officer · LENS Computer vision
- Transaction Fraud Investigator · TRACE
Operations Director AI AgentPULSE
- Waste & Markdown Analyst · HARVEST Image OCR
- Price & Promo Compliance Officer · TAG Computer vision
- Floor & Queue Supervisor · FLOW Computer vision
- Facilities & Cold-Chain Engineer · FROST
Logistics Director AI AgentCOMPASS
- Route & Load Optimization AI Agent · VECTOR
- Receiving Dock Inspector · GATE Vision + OCR
Commercial Director AI AgentMARGIN
- Competitive Intelligence Analyst · RADAR
- Supplier Negotiation Analyst · LEVERAGE
- Planogram & Space Analyst · BLUEPRINT Computer vision
- Retail Media Manager · SPOTLIGHT
Customer Experience Director AI AgentECHO
- Call & Chat Quality Analyst · TONE
- WhatsApp Concierge · THREAD
Expansion Director AI AgentSCOUT
- Catchment & Cannibalization Analyst · MAPPER
Dept 1 · Loss Prevention The Security Director AI Agent SENTRYComputer vision Computer vision and POS evidence for integrated shrink-risk assessment · 2 AI Agents
Director AI Agent · loss prevention
The Security Director AI Agent SENTRYComputer vision-led
Combines camera and POS signals into one evidence-based view of shrink risk by location, till and daypart. SENTRY prioritises cases for review and directs its AI Agents and loss-prevention teams toward the highest-value interventions.
Primary decision surface: a shrink-risk map across products, aisles, tills and dayparts, supported by linked incident evidence.
Full technical profile
ProblemShrink comes from two blind spots at once — floor theft the cameras record but nobody reviews, and till fraud buried in POS logs. Owned separately, neither adds up to a picture of where margin actually leaks.
ProductA Director AI Agent layered over LENS (camera event detection) and TRACE (POS anomaly detection). It combines camera and transaction signals into one risk view and prioritises human review and loss-prevention deployment where potential value at risk is greatest.
What gets installedShrink data layer joining LENS incidents and TRACE cases; unified risk map (product × aisle × daypart × till); guard/loss-prevention deployment console; case management with evidence.
Main KPIs
Total shrink %Validated incidentsRecovery valueIntervention timeShrink per store cluster
AI Agents reporting to SENTRY
Shoplifting Detection Officer LENSComputer vision
Computer vision on existing CCTV identifies defined concealment and walkout patterns, then sends a discreet alert and short clip for human confirmation.
How it works
- Signal
- Existing CCTV over aisles and self-checkout.
- How it decides
- Computer vision scores concealment and walkout gestures against a theft-likelihood threshold.
- Output
- A discreet alert with a short clip to the guard or security desk for human confirmation.
Full technical profile
ProblemShoplifting is often identified only during inventory counts or retrospective CCTV review, delaying intervention and obscuring the source of shrink.
ProductAn AI Agent sharing the same CCTV analytics appliance as FLOW (or a second analytics channel on it). It identifies defined concealment and exit-without-payment patterns, then sends a discreet alert and short clip for human confirmation. Aggregated incidents support risk mapping by SKU, aisle and daypart.
What gets installedComputer-vision theft module on the existing CCTV/NVR AI box; discreet mobile/earpiece alerts with a 10–20s clip; incident log linked to camera timestamp; risk heatmap by product, aisle and daypart.
Main KPIs
Theft shrinkageValidated incidentsRecovered merchandiseFalse positivesIntervention time
Transaction Fraud Investigator TRACE
Anomaly detection on POS records — void abuse, discount manipulation, sweethearting — the loss cameras cannot see.
How it works
- Signal
- POS transaction logs: voids, discounts, refunds, no-sales and cashier IDs.
- How it decides
- Rules catch known fraud patterns; anomaly detection flags tills and cashiers that deviate from their peer baseline.
- Output
- A ranked investigation queue with supporting evidence — a lead for a human, not an accusation.
Full technical profile
ProblemTill-level fraud — void abuse, discount manipulation, sweethearting, phantom refunds — is invisible to cameras and buried in transaction logs.
ProductAn AI Agent monitoring the POS data stream. A rules layer identifies known fraud patterns, while an anomaly-detection model flags material deviations from peer and historical baselines. It prepares an evidence-backed investigation queue and can cross-reference relevant LENS camera timestamps.
What gets installedPOS data-stream connector (voids, discounts, refunds, no-sales, cashier IDs); rules + anomaly-detection engine; loss-prevention case queue with evidence and camera cross-reference.
Main KPIs
Cashier shrinkValidated fraud casesRecovery valueFalse positivesTime-to-investigate
Dept 2 · Store Operations The Operations Director AI Agent PULSE Human-approved Action Plans across store operations · 4 AI Agents
Director AI Agent · store operations
The Operations Director AI Agent PULSE
Monitors sales, stock, pricing, promotions, queues, cold-chain conditions and deliveries. PULSE ranks material exceptions and prepares a human-reviewable Action Plan showing the evidence, expected impact and proposed steps. Approved actions can be routed to connected systems under defined controls and permissions.
Primary decision surface: a daily store-health view that ranks stores by risk, explains likely causes and assigns accountable actions.
Full technical profile
ProblemOperational signals are distributed across alerts, messaging, paper processes and separate systems. Material issues can therefore remain unresolved until they affect availability, labour productivity or sales.
ProductA Director AI Agent that monitors the principal store-operating signals, identifies exceptions and prepares an Action Plan with supporting evidence, estimated impact and implementation steps. Managers approve, edit or reject each plan. Approved actions are sent to connected systems subject to role-based permissions and a complete audit trail.
What gets installedAI Retail OPS Agent (cloud) connected to SAP/POS/WMS/pricing/promos; manager console with Approve/Edit/Reject; staff phone app receiving approved tasks (scan/photo close-out); execution connectors (label printers, POS price file, replenishment/transfer tickets, labour roster); audit trail of plan vs actual.
Main KPIs
Action Plans approvedTime-to-executeOn-shelf availabilitySales recoveredPrice / promo complianceForecast vs actual
AI Agents reporting to PULSE
Waste & Markdown Analyst HARVESTImage OCR
Uses staff-phone image OCR to capture expiry dates, then recommends markdown, transfer, return or removal before stock becomes waste.
How it works
- Signal
- Staff-phone images of expiry dates, interpreted with OCR, combined with POS sell-through and ERP stock data by SKU and store.
- How it decides
- Predicts days-to-expiry against likely sell-through, then ranks each item to markdown, transfer, return or write off.
- Output
- A prioritised action list and a printable markdown ticket the manager approves.
Full technical profile
ProblemProducts are spotted too late, driving avoidable markdowns, waste and expired stock on thin fresh margins.
ProductAn AI Agent that reads the near-expiry register — populated as staff scan the barcode and OCR the expiry date — and recommends markdown, transfer, return or removal, printing the ticket on approval. It later feeds order adjustments back to ERP so the store stops over-ordering the same SKU.
What gets installedExpiry-scan module in the store app (phone/handheld); cloud near-expiry register linked to WMS; markdown/transfer/return ticket workflow that prints a label and closes in POS; predictive waste model writing to SAP.
Main KPIs
Waste %Expiry write-offsSell-through before expiryMarkdown recoverySupplier returns
Price & Promo Compliance Officer TAGComputer vision
Pushes price and promo jobs to the floor and verifies the shelf tag matches the POS by photo and computer vision.
How it works
- Signal
- Daily price and promo files plus a phone photo of the shelf tag.
- How it decides
- Computer vision reads the printed tag and matches it against what POS is actually charging.
- Output
- A task queue of tags to fix; the compliance rate feeds PULSE and MARGIN.
Full technical profile
ProblemShelf prices drift from POS prices and promotions get installed late or wrong — breaking the discount promise.
ProductAn AI Agent that issues a job to the store app whenever a price or promo changes centrally — print the label, place it, scan the product, photograph the shelf edge — then reads the photo with computer vision, compares the printed price to the POS file, and closes the task or flags a mismatch on a live compliance board.
What gets installedPrice/promo job module in the store app; connection to existing label printers for one-tap print; photo + barcode proof flow with a CV model comparing tag to POS; head-office compliance dashboard (store × SKU exceptions).
Main KPIs
Price accuracyComplaintsRefundsPromotion compliancePromotional sales uplift
Floor & Queue Supervisor FLOWComputer vision
Reads existing CCTV for queue length, closed tills and unattended zones, and alerts the manager’s phone before a line becomes a complaint.
How it works
- Signal
- The store’s existing CCTV feeds over tills and aisles.
- How it decides
- Computer vision counts people in line and idle tills, comparing against a wait-time threshold.
- Output
- A push alert to the duty manager’s phone before the queue turns into a complaint.
Full technical profile
ProblemLong queues, closed tills and unattended areas reduce service quality and throughput during peak periods across a high-volume store network.
ProductAn AI Agent running on a small edge appliance (or cloud connector) attached to the store’s existing CCTV/NVR — no new cameras. It uses computer vision to count the queue, detect open versus closed tills and flag empty aisles, cross-checked against POS transaction rate per till, and sends a proposed intervention to the manager’s phone. The same feed can support shift-staffing recommendations.
What gets installedAI edge box / NVR plugin on existing camera streams; POS till-status feed into the same console; manager mobile alerts; store/area dashboard for queue time and till utilisation.
Main KPIs
Queue timeTill utilisationResponse timeUnattended minutesTransactions per labour hour
Facilities & Cold-Chain Engineer FROST
Uses sensors in fridges and cold rooms to open a controlled work order when temperature drifts beyond policy and to recommend energy settings against footfall and tariffs.
How it works
- Signal
- IoT temperature sensors in fridges and cold rooms, energy meters and footfall.
- How it decides
- Flags sustained temperature drift and models energy use against footfall and tariff windows.
- Output
- A maintenance work order with the affected asset and a nightly setpoint recommendation.
Full technical profile
ProblemEquipment failures are caught late, causing spoilage, food-safety risk and emergency maintenance across 20,000 m² of cold storage and 1,150+ stores.
ProductAn AI Agent fed by wireless temperature and humidity sensors inside cold rooms, fridges and freezers, plus optional energy meters on compressors/HVAC, via a store/DC gateway. It opens a controlled work order when readings drift beyond policy, identifies the affected asset and can later support failure prediction and energy optimisation.
What gets installedIoT temperature/humidity sensors in cold assets; optional clamp/smart meters on refrigeration and HVAC; store/DC gateway + cloud console; auto work-order tickets with SLA timers.
Main KPIs
Temperature incidentsSpoilageDowntimeMaintenance responseEnergy consumption
Dept 3 · Fleet & Distribution The Logistics Director AI Agent COMPASS On-time delivery, cost per drop and AI-assisted dispatch · 2 AI Agents
Director AI Agent · fleet & distribution
The Logistics Director AI Agent COMPASS
Coordinates delivery performance and cost per drop across the fleet. COMPASS combines live vehicle positions, ETAs, store demand and warehouse orders to recommend route and load changes, while linking delivery exceptions to their potential availability impact.
Primary decision surface: a live dispatch view with ETAs, proof of delivery, route exceptions and recommended interventions.
Full technical profile
ProblemLate or poorly scheduled deliveries cause store stockouts, dock congestion and high transport cost — and the 300-truck fleet runs without a live view.
ProductA Director AI Agent layered over VECTOR (route and load optimisation). GPS or existing telematics, a driver application and WMS order data feed a live dispatch view. COMPASS recommends route and load changes at planning time and during the day, and can flag cold-chain exceptions from reefer temperature probes.
What gets installedGPS trackers (or telematics API); driver Android app (stops, POD, exceptions); dispatch console (live map, ETAs, capacity/load planning); AI routing engine fed by store demand/WMS; optional cold-chain probe on reefers.
Main KPIs
On-time deliveryCost per deliveryCapacity utilisationFuel useDelivery-related stockouts
AI Agents reporting to COMPASS
Route & Load Optimization AI Agent VECTOR
Continuously re-sequences delivery routes and truck loads against live traffic, store demand and dock congestion, and pushes the updated run straight to the driver app.
How it works
- Signal
- Live GPS positions, traffic conditions, store order windows and dock congestion at each DC.
- How it decides
- A routing AI Agent re-solves the vehicle routing and load plan in near real time, balancing distance, delivery windows, truck capacity and driver hours.
- Output
- An updated turn-by-turn run pushed to the driver app, with the time and fuel saved versus the original plan.
Full technical profile
ProblemStatic, pre-planned routes cannot react to traffic, last-minute order changes or a blocked dock door, so trucks run longer than necessary and some stores wait later than they should.
ProductA routing AI Agent that re-optimises each truck’s route and load against live traffic, store demand and dock congestion. It updates the plan during the day when defined triggers occur and, following dispatch approval, sends the revised run to the driver application with the expected impact recorded.
What gets installedLive traffic/GPS feed integration; vehicle-routing optimization engine (capacity, time windows, driver hours); real-time re-planning triggers from WMS/demand and dock status; driver app updates with the revised run.
Main KPIs
Distance per deliveryFuel useOn-time deliveryRoute re-plans executedCost per drop
Receiving Dock Inspector GATEVision + OCR
Camera and OCR at each DC dock door count cases, read expiry dates and match every pallet against the PO before it enters the warehouse.
How it works
- Signal
- A camera and OCR at each dock door.
- How it decides
- Counts cases, reads expiry dates and matches the tally against the purchase order.
- Output
- An instant discrepancy flag before putaway, feeding the goods-receipt note and ERP.
Full technical profile
ProblemShort deliveries, wrong cases and near-dated stock slip into the warehouse because manual dock checks are slow and inconsistent.
ProductAn AI Agent reading fixed cameras and OCR at each DC dock door. It counts cases, reads expiry dates and matches each pallet against the purchase order before putaway, creating an evidenced receiving checkpoint.
What gets installedDock-door cameras + OCR at each DC bay; case-count and expiry-read model; PO-match service feeding the goods-receipt note and ERP; discrepancy log.
Main KPIs
Receiving accuracyShort/over deliveries caughtNear-dated stock rejectedDock throughput
Dept 4 · Pricing & Merchandising The Commercial Director AI Agent MARGIN Evidence-based margin Action Plans · 4 AI Agents
Director AI Agent · pricing & merchandising
The Commercial Director AI Agent MARGIN
Combines competitor pricing, supplier economics, shelf productivity and retail-media demand into ranked commercial Action Plans. Recommendations are evaluated against margin, price position, availability and customer-value constraints before human approval.
Primary decision surface: a commercial view that consolidates four AI Agent outputs into a prioritised set of margin opportunities.
Full technical profile
ProblemPricing, supplier terms, shelf productivity and media demand are managed in separate silos, so the commercial team never sees the full margin picture or the trade-offs between moves.
ProductA Director AI Agent layered over RADAR (competitor prices), LEVERAGE (supplier economics), BLUEPRINT (shelf productivity) and SPOTLIGHT (media demand). It consolidates the four AI Agent feeds into commercial Action Plans, ranked by estimated margin impact and reviewed against pricing and availability constraints.
What gets installedCommercial cockpit joining the four AI Agent feeds; margin-impact ranking model; Action Plan workflow (approve to route into pricing/negotiation/space/media); private-label and price-gap views.
Main KPIs
Gross marginPrivate-label sharePrice index vs competitorsTrade termsMedia revenue
AI Agents reporting to MARGIN
Competitive Intelligence Analyst RADAR
Tracks competitor prices and promotions daily per SKU, alerts on moves, and arms buyers with negotiation benchmarks.
How it works
- Signal
- Daily competitor price and promo capture per SKU.
- How it decides
- Matches items to Kazyon’s basket and tracks the price gap against a threshold.
- Output
- A daily price-gap alert and a negotiation benchmark sheet for buyers.
Full technical profile
ProblemKazyon’s price position versus BIM and Carrefour is checked ad hoc, so it reacts late when a competitor moves on a key SKU.
ProductAn AI Agent that reads competitor websites and applications into a price board matched to Kazyon’s basket. It flags relevant movements on watched SKUs and prepares current benchmarks for buyer review and supplier negotiation.
What gets installedDaily competitor price/promo scraper; SKU-matching engine to Kazyon’s basket; commercial price board with gap alerts; export to the negotiation benchmark sheet.
Main KPIs
Price-gap vs competitorsReaction time to movesMargin on watched SKUsWin rate in negotiation
Supplier Negotiation Analyst LEVERAGE
Puts sell-through, shelf-share versus sales and delivery reliability by supplier on screen — in the negotiation room.
How it works
- Signal
- Sell-through, shelf-share versus sales, fill rate and delivery reliability by supplier.
- How it decides
- Compares the space and terms each supplier is given against the sales they actually earn.
- Output
- A one-page supplier scorecard, live on screen in the negotiation room.
Full technical profile
ProblemSupplier negotiations can proceed without a consistent view of the relationship between commercial terms, shelf allocation, service levels and the sales each supplier generates.
ProductAn AI Agent connected to sell-through, shelf-share, fill-rate and delivery-reliability data by supplier. It compares commercial support and shelf allocation with realised performance and prepares a concise supplier scorecard for negotiation.
What gets installedSupplier data pipeline (sell-through, shelf-share, fill rate, on-time delivery); scoring model comparing terms/space to sales; negotiation-room scorecard view.
Main KPIs
Trade terms improvementFill rateOn-time deliveryMargin per supplierShelf-space ROI
Planogram & Space Optimization Analyst BLUEPRINTComputer vision
Estimates the sales impact of a shelf change before physical rollout and uses a virtual store simulator to compare layout scenarios.
How it works
- Signal
- Sales-per-facing history, the storebook rules and a proposed layout change.
- How it decides
- A forecasting model simulates the change in a virtual store before any physical reset.
- Output
- A predicted uplift or downside and a compliance score, delivered before staff touch the shelf.
Full technical profile
ProblemThe storebook is a rigorous rule set, but it cannot see the future: shelf changes ship on judgement, and compliance is a manual photo audit.
ProductAn AI Agent coordinating four linked modules: inputs, forecast, proposed layout and compliance audit. The Sales Forecast Engine estimates sales per facing; the Virtual Store Simulator compares layout scenarios; and the computer-vision scanner assesses a shelf image against the approved planogram. Outputs can be exported to DotActiv or Quant Retail.
What gets installedSales forecast model on sales-per-facing history; virtual store simulator (web) with A/B save + export; CV compliance scanner in the store app; store-tier classification service.
Main KPIs
Sales per meterCompliance rateOOS SKUsForecast vs actualReset cost avoided
Retail Media Manager SPOTLIGHT
Sells in-store screen and Kazyon Plus placements to suppliers with proof of sales lift — a new high-margin revenue line.
How it works
- Signal
- Screen and app placement inventory plus sales data for the promoted SKUs.
- How it decides
- Matches each supplier booking to the before-and-after sales lift it produced.
- Output
- A sales-lift report that turns screen space into a sellable, evidenced media product.
Full technical profile
ProblemKazyon’s footfall and Kazyon Plus audience are a high-margin media asset that today earns nothing for suppliers who would pay to reach them.
ProductAn AI Agent that manages digital-screen inventory and sponsored placements in Kazyon Plus, links each booking to measured sales lift and supports a performance-based retail-media proposition for suppliers.
What gets installedIn-store digital screens + CMS; Kazyon Plus sponsored-slot inventory; booking/billing tool for suppliers; sales-lift measurement tied to POS.
Main KPIs
Media revenueSupplier-funded spendSales lift per campaignFill rate of slots
Dept 5 · Customer Care The Customer Experience Director AI Agent ECHO Interaction quality, CSAT and first-contact resolution · 2 AI Agents
Director AI Agent · customer experience
The Customer Experience Director AI Agent ECHO
Assesses hotline, WhatsApp and delivery interactions against a consistent service framework. ECHO identifies recurring pain points, supports coaching and coordinates routine service automation subject to policy, system access and human escalation rules.
Primary decision surface: interaction-quality trends, service-representative scorecards, recurring issues and critical-call flags.
Full technical profile
ProblemAs Kazyon Plus and delivery grow, every hotline and WhatsApp thread is both a quality risk and an unread customer insight — and manual QA can only sample a fraction.
ProductA Director AI Agent layered over TONE (quality assessment) and THREAD (WhatsApp commerce). It consolidates service KPIs and recurring issues, then coordinates routine requests through approved ERP, CRM and order-management workflows, escalating exceptions to human service representatives.
What gets installedInteraction intelligence layer over TONE + THREAD; CSAT/FCR/AHT dashboard; pain-point feed to other AI Agents; later an AI co-pilot connected to SAP, CRM and order management for approved routine requests and human escalation.
Main KPIs
CSAT / NPSFCRAHTRepeat complaintsManual QA hoursAutomated interactions
AI Agents reporting to ECHO
Call & Chat Quality Analyst TONE
Transcribes and assesses interactions in dialect-aware Arabic, builds service-representative scorecards, supports coaching and flags potential policy breaches for review.
How it works
- Signal
- Hotline, WhatsApp and delivery-call recordings, with coverage expanded toward the full interaction base and transcribed in dialect-aware Arabic.
- How it decides
- Scores each interaction against a quality rubric and flags high-risk conversations as they happen.
- Output
- A service-representative scorecard with coaching notes plus a live flag for supervisor intervention.
Full technical profile
ProblemHotline, WhatsApp, delivery and complaint interactions are reviewed manually — if at all — so service quality, representative performance and recurring pain points remain difficult to assess consistently.
ProductAn AI Agent connected to the call recorder, WhatsApp Business API and delivery-call lines. It transcribes interactions in dialect-aware Arabic, applies an approved quality rubric and writes the result to a service-representative scorecard. Managers see service trends, CSAT/NPS sentiment, FCR, AHT and recurring themes; critical calls trigger a supervisor flag.
What gets installedConnectors to hotline recorder, WhatsApp Business and delivery-call lines; speech-to-text + LLM assessment engine designed for full interaction coverage; manager console with scorecards, coaching guidance, pain-point dashboards and critical-call flags.
Main KPIs
FCRAHTEscalation rateCSAT / NPS sentimentManual QA hoursRepeat complaints
WhatsApp Concierge THREAD
Conversational commerce in Arabic on a channel customers already use: search, basket, order, loyalty — no app download.
How it works
- Signal
- Inbound WhatsApp messages plus live catalogue, stock and loyalty data.
- How it decides
- An Arabic conversational model interprets the request against live ERP data.
- Output
- A completed order, answer or loyalty action inside the same chat thread — no app download.
Full technical profile
ProblemMany Kazyon shoppers will not download an app, so a large, price-sensitive base is hard to reach for ordering, loyalty and re-engagement.
ProductAn AI Agent operating on WhatsApp that supports product search, basket building, order placement and tracking, and loyalty enquiries in Arabic, grounded in current catalogue, stock, pricing and loyalty data.
What gets installedWhatsApp Business API integration; Arabic conversational model; live connectors to catalogue, stock, pricing and loyalty; order + payment/handoff flow.
Main KPIs
Orders via WhatsAppBasket sizeRepeat order rateLoyalty engagementCost per order
Dept 6 · Growth The Expansion Director AI Agent SCOUT Site scoring before lease commitment · 1 AI Agent
Director AI Agent · expansion & growth
The Expansion Director AI Agent SCOUT
Scores every candidate location on catchment wealth, cannibalization risk and day-one assortment before the lease is signed — then benchmarks each new store against its predicted potential once it opens.
Primary decision surface: a site-investment map with demand, cannibalization, assortment and go/no-go evidence for each candidate location.
Full technical profile
ProblemRapid expansion toward 5,000 stores risks signing leases that cannibalize nearby Kazyon stores or under-perform their catchment — decisions made before the data is in.
ProductA Director AI Agent layered over MAPPER that assesses candidate locations using catchment demand, cannibalization risk and proposed opening assortment before lease commitment, then benchmarks post-opening performance against the investment case.
What gets installedSite-selection map UI for the expansion team; catchment + cannibalization scoring (via MAPPER); day-one assortment recommender; post-open actual-vs-predicted benchmarking.
Main KPIs
New-store sales vs forecastNet-new vs cannibalizedPayback periodSite-decision hit rate
AI Agents reporting to SCOUT
Catchment & Cannibalization Analyst MAPPER
Runs the catchment wealth index and cannibalization simulation so a new store grows the network, not just the footprint.
How it works
- Signal
- Governorate wealth data, the existing branch network and each candidate site.
- How it decides
- Scores catchment demand and simulates how much of a new store’s sales would be pulled from nearby Kazyon stores versus genuinely new.
- Output
- A go / no-go score and a net-new-sales estimate before the lease is signed.
Full technical profile
ProblemOn the path to 5,000 stores, new sites are chosen partly on instinct — risking cannibalization of nearby Kazyon stores rather than net-new demand.
ProductAn AI Agent used before lease commitment to assess catchment demand, estimate sales transferred from nearby Kazyon stores versus net-new demand, and recommend an opening assortment for the local catchment.
What gets installedCatchment wealth index (governorate/district data); existing-branch network model; cannibalization simulation; day-one assortment recommender; site scoring map UI.
Main KPIs
New-store sales vs forecastNet-new vs cannibalized salesPayback periodSite-decision hit rate
Enabling architecture and governance
The Systems Integrator AI Agent ANCHOR
Enabling AI Agent · portfolio infrastructure
The Systems Integrator AI Agent ANCHOR
Creates the governed data and integration layer used across the AI Agent portfolio. ANCHOR connects SAP, POS, WMS, CCTV, GPS and customer-interaction systems; applies common data definitions and access controls; and records how each AI Agent reads data, recommends an action and writes back to an operating system.
Investment implication: individual AI Agents can be piloted against existing systems, while ANCHOR is developed in parallel and expanded as validated capabilities move into production. This limits up-front platform investment without creating a separate data stack for every use case.
Full technical profile
ProblemKazyon’s relevant data is distributed across SAP, POS, WMS, CCTV, GPS and call platforms. Without shared definitions, access rules and integration standards, each AI initiative would create its own pipeline, controls and version of the truth.
ProductAn enabling AI Agent and governed data layer that standardises how Director AI Agents and domain AI Agents access operational data, exchange context, submit actions for approval and record outcomes.
What gets installedCloud data warehouse; ingestion connectors for core operating systems; data-quality controls; role-based access, lineage and audit services; shared schemas and integration interfaces for the AI Agent portfolio.
Main KPIs
Sources integratedData quality / completenessFreshness (latency)AI Agents servedTime to onboard an AI AgentAuditable actions
Portfolio extensions
Additional AI Agents to consider as the model matures
The model is designed to extend. Once the core capabilities are established and delivering measurable results, these additional AI Agents become viable — each building on the data and systems already in place.
Customer Service Co-Pilot COPILOT
An extension of ECHO: integrated with ERP and CRM to resolve approved routine orders, changes and refunds, with exceptions routed to human service representatives.
Checkout-Free Experience Lead EXPRESSComputer vision
A pilot for queue-free, walk-out shopping, using the same camera intelligence that FLOW and LENS already operate.
Micro-Fulfilment AI Agent SWIFT
Dark-store fulfilment for Kazyon Plus delivery at scale, with picking and staging optimised by the same demand models.
Cross-Market Playbook Analyst ATLAS
Adapts the Egypt-trained models for Morocco and Saudi Arabia with local calibration, extending the model to new markets.
Portfolio summary
AI Agent roles and mandates at a glance
| Tier | Department | Title | Codename | Mandate |
|---|---|---|---|---|
| Director AI Agent | Loss Prevention | The Security Director | SENTRY | Integrated shrink-risk assessment |
| AI Agent | Loss Prevention | Shoplifting Detection Officer | LENS | Computer vision · Event detection on existing CCTV |
| AI Agent | Loss Prevention | Transaction Fraud Investigator | TRACE | POS anomaly detection |
| Director AI Agent | Store Operations | The Operations Director | PULSE | Human-approved Action Plans across store operations |
| AI Agent | Store Operations | Waste & Markdown Analyst | HARVEST | Image OCR · Near-expiry markdown / transfer / return |
| AI Agent | Store Operations | Price & Promo Compliance Officer | TAG | Computer vision · Shelf tag matches POS, verified by photo |
| AI Agent | Store Operations | Floor & Queue Supervisor | FLOW | Computer vision · Queue, till and zone alerts from existing CCTV |
| AI Agent | Store Operations | Facilities & Cold-Chain Engineer | FROST | Fridge sensors, work orders, energy optimisation |
| Director AI Agent | Fleet & Distribution | The Logistics Director | COMPASS | On-time delivery, cost per drop and AI-assisted dispatch |
| AI Agent | Fleet & Distribution | Route & Load Optimization AI Agent | VECTOR | Continuous route and load re-optimization against live traffic and demand |
| AI Agent | Fleet & Distribution | Receiving Dock Inspector | GATE | Vision + OCR · PO verification at the DC |
| Director AI Agent | Commercial | The Commercial Director | MARGIN | Evidence-based margin Action Plans |
| AI Agent | Commercial | Competitive Intelligence Analyst | RADAR | Daily competitor price / promo tracking |
| AI Agent | Commercial | Supplier Negotiation Analyst | LEVERAGE | Supplier economics for negotiation |
| AI Agent | Commercial | Planogram & Space Optimization Analyst | BLUEPRINT | Computer vision · Predict and audit shelf changes |
| AI Agent | Commercial | Retail Media Manager | SPOTLIGHT | Supplier-funded media with measured sales lift |
| Director AI Agent | Customer Care | The Customer Experience Director | ECHO | Interaction quality, CSAT and first-contact resolution |
| AI Agent | Customer Care | Call & Chat Quality Analyst | TONE | Transcription, quality assessment and coaching support |
| AI Agent | Customer Care | WhatsApp Concierge | THREAD | Conversational commerce in Arabic |
| Director AI Agent | Growth | The Expansion Director | SCOUT | Site scoring before lease commitment |
| AI Agent | Growth | Catchment & Cannibalization Analyst | MAPPER | Catchment index and cannibalization simulation |