The Kazyon AI Workforce · Capabilities organised as an operating model

An AI operating model for Kazyon, presented as a team of Directors and Specialists.

This page organises the proposed AI capabilities the way an organisation is structured. Each domain has a Director — an agent that monitors its area, drafts recommended Action Plans and owns the outcome — supported by Specialists that handle narrower tasks within it. The structure is modular: capabilities can be adopted individually, by domain, or as a whole, and extended over time.

A shared data foundation is a prerequisite: the Systems Integrator consolidates Kazyon’s systems into one governed source of truth before any other capability can operate across the business. Each role is shown as a summary card; expanding its full technical profile reveals the problem it addresses, the product, what is installed, an indicative roadmap and the KPIs it moves — all on this page.

How to read this page

The Day Zero foundation is described first, followed by each domain. A Director can operate independently; its Specialists extend the domain’s coverage. Adoption is incremental and can be sequenced against results. Every role corresponds to a concrete, installable system — expand its full technical profile for the detail.

Day Zero · Prerequisite foundation

The Systems Integrator ANCHOR

Foundation · implemented first
The Systems Integrator ANCHOR

Consolidates SAP, POS, WMS, CCTV, GPS and the call systems into one governed, shared data layer that every other agent reads from and writes to. No agent can monitor, plan or execute across the business until this foundation is in place, so it is implemented first.

Day one: ingest and clean the core data sources into a single source of truth with quality controls and access rules, ready for the Directors to build on.

Full technical profile
ProblemKazyon’s data lives in disconnected systems — SAP, POS, WMS, CCTV, GPS, call platforms — so no agent can reason across the business, and every AI initiative would otherwise rebuild its own brittle pipeline.
ProductA governed data warehouse: the foundation every agent plugs into. It ingests and cleans the core sources into one source of truth with quality controls, access rules and a shared schema each Director and Specialist reads from and writes to.
What gets installedCloud data warehouse + ingestion connectors (SAP, POS, WMS, pricing, promos, fleet GPS, loyalty, call/chat); data-quality and validation layer; governance (access rules, lineage, audit); shared feature/schema layer for the agents.
Roadmap
0–3 moIngest and clean the core sources into one source of truth.
4–9 moQuality controls, governance and shared schema for the first Directors.
10–18 moReal-time feeds and feature store as more agents come online.
Main KPIs
Sources integratedData quality / completenessFreshness (latency)Agents servedTime-to-onboard a new agent
The org chart

One foundation, six Directors, fourteen Specialists

Every department below is collapsed by default. Expand one to see its Director and specialists; expand any role’s full technical profile for the problem, product, install list, roadmap and KPIs — all inline. Expand all

The Systems IntegratorANCHOR · Day Zero
Operations DirectorPULSE
  • Waste & Markdown Analyst · HARVEST
  • Price & Promo Compliance Officer · TAG
  • Floor & Queue Supervisor · FLOW
  • Facilities & Cold-Chain Engineer · FROST
Security DirectorSENTRY
  • Shoplifting Detection Officer · LENS
  • Transaction Fraud Investigator · TRACE
Logistics DirectorCOMPASS
  • Receiving Dock Inspector · GATE
Commercial DirectorMARGIN
  • Competitive Intelligence Analyst · RADAR
  • Supplier Negotiation Analyst · LEVERAGE
  • Planogram & Space Analyst · BLUEPRINT
  • Retail Media Manager · SPOTLIGHT
Customer Experience DirectorECHO
  • Call & Chat Quality Analyst · TONE
  • WhatsApp Concierge · THREAD
Expansion DirectorSCOUT
  • Catchment & Cannibalization Analyst · MAPPER
Dept 1 · Store Operations The Operations Director PULSE Action Plans across the floor, executed on approval · 4 specialists
Director · store operations
The Operations Director PULSE

Watches sales, stock, price, promotions, queues, cold chain and deliveries the way a sharp ops lead would, and surfaces what a human would miss. Instead of another dashboard, PULSE drafts an Action Plan — the insight, the why, the expected impact, and a step-by-step implementation across every system. The manager taps Approve, Edit or Reject; on approve, PULSE executes instantly across SAP, POS, printers and staff phones.

Its own desk: the daily store-health control tower — stores ranked by risk with root causes and assigned actions — is part of PULSE’s scope, not a separate capability.

Full technical profile
ProblemManagers drown in alerts but still miss what matters. Decisions that should take minutes — move stock, fix a price, open a till, pull near-expiry, reshuffle labour — wait on WhatsApp, paper and gut feel until the sale or the product is gone.
ProductAn AI Operations Director agent that watches sales, stock, price, promos, queues, cold chain and deliveries and surfaces the miss. Instead of a dashboard it drafts an Action Plan — insight, why, expected impact and a step-by-step implementation across systems. The manager taps Approve / Edit / Reject; on approve it executes instantly across SAP, POS, printers and staff phones. It also owns the daily store-health control tower that ranks stores by risk.
What gets installedAI Retail OPS agent (cloud) plugged into 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.
Roadmap
0–3 moAction Plans for stockouts, price changes and promo setup; approve pushes tasks + prints labels.
4–9 moRecommender learns which plans recover sales; one-tap multi-system execute.
10–18 moNear-autonomous: auto-approve low-risk plans within policy; humans on exceptions.
Main KPIs
Action Plans approvedTime-to-executeOn-shelf availabilitySales recoveredPrice / promo complianceForecast vs actual

Reporting into PULSE

Waste & Markdown Analyst HARVEST

Flags near-expiry stock, recommends markdown, transfer, return or removal before it becomes waste, and prints the ticket.

How it works
Signal
POS sell-through and ERP expiry dates per 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.
ProductStaff scan the barcode and OCR the expiry into a near-expiry register; the AI recommends markdown, transfer, return or removal and prints the ticket. The model 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.
Roadmap
0–3 moScan + expiry alerts on staff phones.
4–9 moAI markdown / transfer / return recommendations with printed labels.
10–18 moPredictive waste + automated order adjustments into ERP.
Main KPIs
Waste %Expiry write-offsSell-through before expiryMarkdown recoverySupplier returns
Price & Promo Compliance Officer TAG

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.
ProductWhen a price or promo changes centrally, the store app issues a job: print the label, place it, scan the product, photograph the shelf edge. Computer vision reads the photo, 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).
Roadmap
0–3 moDigital price/promo jobs + printer connection.
4–8 moException alerts + photo verification.
9–15 moFull CV shelf-tag check + promo sales-lift analysis.
Main KPIs
Price accuracyComplaintsRefundsPromotion compliancePromotional sales uplift
Floor & Queue Supervisor FLOW

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, unattended areas and staff clustering quietly erode service at peak — the trip killer at 500,000 customers a day.
ProductA small AI edge appliance (or cloud connector) on the store’s existing CCTV/NVR — no new cameras. Computer vision counts the queue, sees open vs closed tills and flags empty aisles; POS gives transaction rate per till. The manager’s phone gets “Queue > 6 at till 2 — open till 3.” Later the same feed drives shift staffing.
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.
Roadmap
0–3 moEdge box on CCTV; queue and till alerts to manager phone.
4–9 moZone coverage and inactivity monitoring.
10–18 moAI dynamic staffing linked to the store task app.
Main KPIs
Queue timeTill utilisationResponse timeUnattended minutesTransactions per labour hour
Facilities & Cold-Chain Engineer FROST

Sensors in fridges and cold rooms; raises an automatic work order on temperature drift and optimises energy 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
An automatic maintenance work order 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.
ProductWireless temperature and humidity sensors inside cold rooms, fridges and freezers, plus optional energy meters on compressors/HVAC. A store/DC gateway pushes readings to the cloud; on drift, maintenance gets an automatic work order with the exact asset and reading. The AI later predicts compressor failure and optimises cooling against footfall and tariffs.
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.
Roadmap
0–3 moSensor pilot + live temperature alerts.
4–9 moVendor SLA tracking and maintenance work-order system.
10–24 moPredictive maintenance + energy optimisation.
Main KPIs
Temperature incidentsSpoilageDowntimeMaintenance responseEnergy consumption
Dept 2 · Loss Prevention The Security Director SENTRY End-to-end shrink risk picture · 2 specialists
Director · loss prevention
The Security Director SENTRY

Owns shrink end to end. Synthesizes camera and POS signals into a single risk picture — where theft happens, which tills look wrong, which hours are exposed — and directs its specialists and the store’s guards to where loss actually occurs.

Day one: a shrink risk map across products, aisles and dayparts, built from the incidents its specialists surface.

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 layer over LENS (camera theft detection) and TRACE (POS anomaly detection). It fuses camera and transaction signals into one shrink risk picture — where theft happens, which tills look wrong, which hours are exposed — and directs specialists and store guards to where loss 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.
Roadmap
0–3 moCombine LENS + TRACE incidents into one risk map.
4–9 moDirected guard/camera deployment by risk; case management.
10–18 moPredictive shrink forecasting and prevention planning.
Main KPIs
Total shrink %Validated incidentsRecovery valueIntervention timeShrink per store cluster

Reporting into SENTRY

Shoplifting Detection Officer LENS

Computer vision on existing CCTV detects concealment and walkout patterns in real time and sends a discreet alert with a short clip.

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 usually discovered only at inventory count or after manual CCTV review — pure margin loss.
ProductThe same AI CCTV appliance as FLOW (or a second analytics channel on it) watches existing feeds for concealment, exit-without-payment patterns and sweethearting cues, sending a discreet alert with a short clip to the security phone or earpiece. A back-office console maps high-risk SKUs, aisles and hours to redeploy cameras and guards.
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.
Roadmap
0–3 moIncident reporting + one-click CCTV clip linking.
4–9 moReal-time AI alert pilot on live camera streams.
10–18 moPredictive theft-risk maps and guard deployment.
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.
ProductSoftware on the POS data stream. A rules layer catches known fraud patterns; an anomaly-detection model learns each till and cashier’s normal behaviour and flags deviations from the peer baseline. Cases are queued for loss-prevention with the evidence attached — and can be cross-referenced with LENS camera timestamps for the strongest ones.
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.
Roadmap
0–3 moIngest POS logs; rules-based flags + investigation queue.
4–9 moPer-till / per-cashier anomaly baselines; camera cross-reference.
10–18 moContinuous learning; ranked risk scoring across the estate.
Main KPIs
Cashier shrinkValidated fraud casesRecovery valueFalse positivesTime-to-investigate
Dept 3 · Fleet & Distribution The Logistics Director COMPASS On-time delivery, cost-per-drop, AI dispatch · 1 specialist
Director · fleet & distribution
The Logistics Director COMPASS

Owns on-time delivery and cost-per-drop across Kazyon’s 300-truck fleet. Live map, ETAs and a driver app feed an AI dispatch that re-optimises routes and loads from store demand — and links late trucks to the stockout risk they would cause.

Day one: every truck on a live map with ETAs and proof of delivery; nightly route and load optimisation.

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.
ProductEvery truck gets GPS (or connect existing telematics); drivers use a phone app for stop confirmation and proof of delivery; dispatch runs a live map with ETAs and door times. The AI re-optimises routes and loads nightly and mid-day using store stockouts and WMS order files. Reefers can carry temperature probes so a broken cold chain is flagged before the store opens.
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.
Roadmap
0–3 moGPS live + ETAs + delivery tracking on dispatch screen.
4–9 moAI dynamic routes and load optimisation.
10–18 moDemand-linked deliveries + predictive fleet maintenance.
Main KPIs
On-time deliveryCost per deliveryCapacity utilisationFuel useDelivery-related stockouts

Reporting into COMPASS

Receiving Dock Inspector GATE

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.
ProductFixed cameras and OCR at each DC dock door count cases, read expiry dates and match every pallet against the purchase order before it enters the warehouse — turning receiving into an automatic, evidenced 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.
Roadmap
0–3 moCamera + OCR on pilot dock doors; case count vs PO.
4–9 moExpiry-date reading + automatic GRN discrepancy flags.
10–18 moEstate-wide rollout; supplier scorecard feedback.
Main KPIs
Receiving accuracyShort/over deliveries caughtNear-dated stock rejectedDock throughput
Dept 4 · Pricing & Merchandising The Commercial Director MARGIN Weekly margin Action Plans · 4 specialists
Director · pricing & merchandising
The Commercial Director MARGIN

Synthesizes competitor prices, supplier economics, shelf productivity and media demand into weekly commercial Action Plans — where to hold the price line, where to push private label, which supplier to challenge, which shelf to rework.

Day one: a live commercial cockpit combining the four specialist feeds below into a ranked list of margin moves.

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 layer over RADAR (competitor prices), LEVERAGE (supplier economics), BLUEPRINT (shelf productivity) and SPOTLIGHT (media demand). It synthesizes the four feeds into weekly commercial Action Plans — where to hold the price line, where to push private label, which supplier to challenge, which shelf to rework — ranked by margin impact.
What gets installedCommercial cockpit joining the four specialist feeds; margin-impact ranking model; Action Plan workflow (approve to route into pricing/negotiation/space/media); private-label and price-gap views.
Roadmap
0–3 moCockpit combining RADAR + LEVERAGE into ranked moves.
4–9 moAdd BLUEPRINT + SPOTLIGHT; weekly margin Action Plans.
10–18 moElasticity-aware pricing and private-label optimisation.
Main KPIs
Gross marginPrivate-label sharePrice index vs competitorsTrade termsMedia revenue

Reporting into 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.
ProductA cloud scraper pulls competitor websites and apps every morning into a price board for the commercial team, matched to Kazyon’s basket. It alerts when a competitor moves on a watched SKU and arms buyers with live benchmarks for 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.
Roadmap
0–3 moDaily scrape + price board for core categories.
4–9 moGap alerts + negotiation benchmarks by supplier.
10–18 moElasticity-aware repricing suggestions into MARGIN.
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
ProblemBuyers negotiate on relationships and gut feel because the numbers — what space and terms each supplier gets versus the sales they earn — aren’t on the table.
ProductA buyer dashboard wired to sell-through, shelf-share versus sales, fill rate and delivery reliability by supplier. It quantifies which suppliers under-earn the space they hold and produces a one-page scorecard to open in the negotiation room.
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.
Roadmap
0–3 moSupplier scorecards for top categories.
4–9 moSpace-vs-sales and reliability scoring across the base.
10–18 moScenario tool: model term changes before the meeting.
Main KPIs
Trade terms improvementFill rateOn-time deliveryMargin per supplierShelf-space ROI
Planogram & Space Optimization Analyst BLUEPRINT

Predicts the sales impact of a shelf change before rollout and runs a virtual store simulator to test layouts risk-free.

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.
ProductFour modules in one loop — inputs to forecast to layout to audit. A Sales Forecast Engine predicts sales-per-facing; a Virtual Store Simulator lets a merchandiser pick a tier, load the approved planogram, move SKUs and see live forecast (revenue, compliance, OOS) for A vs B; a CV Compliance Scanner scores a phone photo of the real shelf against the approved planogram in under 60 seconds. Exports to DotActiv / 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.
Roadmap
0–3 moForecast engine + simulator on a pilot category and tier.
4–9 moCV compliance scanner in stores; A/B rollout decisions.
10–18 moEstate-wide, tier-aware planograms with automated audit.
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.
ProductDigital screens at entrance and checkout plus sponsored slots in Kazyon Plus, sold to suppliers as retail media. The AI matches each booking to before-and-after sales lift for the promoted SKU, so every placement is sold with proof — a new revenue line, not a favour.
What gets installedIn-store digital screens + CMS; Kazyon Plus sponsored-slot inventory; booking/billing tool for suppliers; sales-lift measurement tied to POS.
Roadmap
0–3 moScreens + CMS pilot in flagship stores; manual bookings.
4–9 moKazyon Plus slots + automated sales-lift reporting.
10–18 moSelf-serve supplier media marketplace with proof of lift.
Main KPIs
Media revenueSupplier-funded spendSales lift per campaignFill rate of slots
Dept 5 · Customer Care The Customer Experience Director ECHO Scores 100% of interactions; owns CSAT / FCR · 2 specialists
Director · customer experience
The Customer Experience Director ECHO

Scores 100% of hotline, WhatsApp and delivery interactions — not a sample — and owns CSAT, First Call Resolution and handle time. Surfaces recurring pain points to the rest of the workforce and, over time, grows into a service co-pilot that resolves routine requests itself.

Day one: every conversation transcribed, scored and turned into an agent scorecard with coaching, plus live flags on critical calls.

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 over TONE (quality scoring) and THREAD (WhatsApp commerce) that owns CSAT, FCR and handle time across 100% of interactions. It surfaces recurring pain points to the rest of the workforce and, over time, grows into a service co-pilot: wired to ERP/CRM/order systems it creates or changes delivery orders, suggests substitutes, quotes status, schedules drops and starts refunds — handing only hard cases to a human.
What gets installedInteraction intelligence layer over TONE + THREAD; CSAT/FCR/AHT ownership dashboard; pain-point feed to other agents; later an AI co-pilot wired to SAP, CRM and order management for end-to-end routine requests, 24/7.
Roadmap
0–3 moScore 100% of interactions; scorecards + live critical-call flags.
3–9 moFCR up, repeat complaints down; standardised quality across branches.
9–18 moERP/CRM co-pilot runs routine orders, status and refunds; humans on complex cases.
Main KPIs
CSAT / NPSFCRAHTRepeat complaintsManual QA hoursAutomated interactions

Reporting into ECHO

Call & Chat Quality Analyst TONE

Transcribes and scores every interaction (Arabic, dialect-aware), builds agent scorecards and coaching, flags policy breaches live.

How it works
Signal
Every hotline, WhatsApp and delivery-call recording — 100%, not a sample — transcribed in dialect-aware Arabic.
How it decides
Scores each interaction against a quality rubric and flags high-risk conversations as they happen.
Output
An agent 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, agent performance and recurring pain points stay invisible until customers leave.
ProductConnectors to the call recorder, WhatsApp Business API and delivery-call lines. Every conversation is auto-recorded, transcribed (Arabic + dialect-aware), scored by AI and written to an agent scorecard — no supervisor listens to a sample. Managers see rankings, CSAT/NPS sentiment, FCR, AHT and recurring themes; critical calls push a live flag to the supervisor.
What gets installedConnectors to hotline recorder, WhatsApp Business and delivery-call lines; speech-to-text + LLM scoring engine (100% of interactions); manager console with scorecards, coaching tips, pain-point dashboards and live critical-call flags.
Roadmap
0–3 moConnect channels; AI scorecards + coaching; 80–90% less manual QA.
3–9 moFCR up, repeat complaints down; process/delivery insights.
9–18 moFeeds the ECHO co-pilot for routine request handling.
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.
ProductA WhatsApp Business bot on the channel customers already use: search the catalogue, build a basket, place and track an order, check loyalty — all in Arabic conversation, grounded in live catalogue, stock and loyalty data. No app download, no new habit.
What gets installedWhatsApp Business API integration; Arabic conversational model; live connectors to catalogue, stock, pricing and loyalty; order + payment/handoff flow.
Roadmap
0–3 moSearch, basket and order on WhatsApp for a pilot area.
4–9 moLoyalty, order status and re-order; live stock/pricing.
10–18 moPersonalised offers; handoff to the ECHO co-pilot.
Main KPIs
Orders via WhatsAppBasket sizeRepeat order rateLoyalty engagementCost per order
Dept 6 · Growth The Expansion Director SCOUT Site scoring before the lease is signed · 1 specialist
Director · expansion & growth
The Expansion Director 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.

Day one: a site map the expansion team opens before committing, with a go / no-go score per 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 over MAPPER that 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 open — closing the loop on site decisions.
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.
Roadmap
0–3 moGo/no-go site scoring map for candidate locations.
4–9 moCannibalization simulation + day-one assortment.
10–18 moPost-open benchmarking feeds back into the model.
Main KPIs
New-store sales vs forecastNet-new vs cannibalizedPayback periodSite-decision hit rate

Reporting into 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.
ProductA map tool the expansion team opens before signing a lease: it scores each candidate on catchment wealth, simulates how much of the new store’s sales would be pulled from existing branches versus genuinely new, and proposes a day-one assortment for that catchment.
What gets installedCatchment wealth index (governorate/district data); existing-branch network model; cannibalization simulation; day-one assortment recommender; site scoring map UI.
Roadmap
0–3 moCatchment index + map for candidate sites.
4–9 moCannibalization simulation + go/no-go score.
10–18 moPost-open benchmarking vs predicted potential.
Main KPIs
New-store sales vs forecastNet-new vs cannibalized salesPayback periodSite-decision hit rate
The full roster

All roles at a glance

TierDepartmentTitleCodenameMandate
Day ZeroFoundationThe Systems IntegratorANCHOROne shared memory across SAP, POS, WMS, CCTV, GPS, calls
DirectorStore OperationsThe Operations DirectorPULSEAction Plans across the floor, executed on approval
SpecialistStore OperationsWaste & Markdown AnalystHARVESTNear-expiry markdown / transfer / return
SpecialistStore OperationsPrice & Promo Compliance OfficerTAGShelf tag matches POS, verified by photo
SpecialistStore OperationsFloor & Queue SupervisorFLOWQueue, till and zone alerts from existing CCTV
SpecialistStore OperationsFacilities & Cold-Chain EngineerFROSTFridge sensors, work orders, energy optimisation
DirectorLoss PreventionThe Security DirectorSENTRYEnd-to-end shrink risk picture
SpecialistLoss PreventionShoplifting Detection OfficerLENSReal-time theft detection on CCTV
SpecialistLoss PreventionTransaction Fraud InvestigatorTRACEPOS anomaly detection
DirectorFleet & DistributionThe Logistics DirectorCOMPASSOn-time delivery, cost-per-drop, AI dispatch
SpecialistFleet & DistributionReceiving Dock InspectorGATECamera + OCR PO verification at the DC
DirectorCommercialThe Commercial DirectorMARGINWeekly margin Action Plans
SpecialistCommercialCompetitive Intelligence AnalystRADARDaily competitor price / promo tracking
SpecialistCommercialSupplier Negotiation AnalystLEVERAGESupplier economics on screen in negotiation
SpecialistCommercialPlanogram & Space Optimization AnalystBLUEPRINTPredict shelf-change impact before rollout
SpecialistCommercialRetail Media ManagerSPOTLIGHTSupplier-funded media, proof of lift
DirectorCustomer CareThe Customer Experience DirectorECHOScores 100% of interactions; owns CSAT / FCR
SpecialistCustomer CareCall & Chat Quality AnalystTONETranscribe, score, coach on every conversation
SpecialistCustomer CareWhatsApp ConciergeTHREADConversational commerce in Arabic
DirectorGrowthThe Expansion DirectorSCOUTSite scoring before the lease is signed
SpecialistGrowthCatchment & Cannibalization AnalystMAPPERCatchment index and cannibalization simulation
Future extensions

Additional roles to consider as the business scales

The model is designed to extend. Once the core capabilities are established and delivering measurable results, these additional roles 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 routine orders, changes and refunds automatically, around the clock, with complex cases routed to human agents.

Checkout-Free Experience Lead EXPRESS

A pilot for queue-free, walk-out shopping, using the same camera intelligence that FLOW and LENS already operate.

Micro-Fulfilment Specialist 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.

The model can begin with a single Director to validate impact in one domain, add Specialists as results are confirmed, and extend to further domains over time. It is designed to start focused, demonstrate measurable value, and scale alongside the business.

Codenames are for reference only. Each role maps to a concrete, installable system — expand its full technical profile above for the detail. Roadmap windows and KPIs are indicative sequencing, not committed dates; investment sizing requires separate scoping per capability.

← Back to the Kazyon retail and technology research