IndustryInvisible lost sales
Phantom inventory is the quiet tax
Stock the system believes is on shelf and is not suppresses replenishment silently, and the lost sale is never recorded because it never happened.
Program 08 · Domain specialisation
SCIKIQ Certified Data and AI EngineerRetail
Retail is the one consumer business that owns the shelf, the stock on it and the person standing next to it. That makes the data unusually direct — you see the actual purchase — and the constraints unusually physical. A recommendation the store cannot execute, because the stock is not there or nobody is rostered to do it, is not a recommendation.
- 80 hrstaught
- 40hands-on AI labs
- 8assessment tests
- +20 hrs2 E2E capstones
Phantom inventoryStock the system believes is on shelf and is not
AvailabilityThe lost sale nobody records, because it never happened
Markdown timingMargin decided weeks before anyone marks anything down
Store labourThe constraint that decides whether any plan is executable
01
Part 1 · The business
The value chain, the people, the trends
Part 1 follows a product from the buying decision to the shelf and, sometimes, back again. It names the people who make each of those calls.
The retail value chain
Six stages from range to returns. Select one.
01
Range & buy
Assortment planning, buying, supplier terms and range reviews.
Decisions made here
- What do we range?
- How much do we buy?
- What do we delist?
Where it breaks today
- Range decisions made on last year plus judgement
- Transferable demand ignored when delisting
- Supplier terms held in agreements nobody reads together
What is measured
- Sales density
- Gross margin
- Range productivity
Where AI landsTransferable demand modelling, supplier term extraction and range performance analysis.
Sales densityGross marginRange productivity
02
Price & promote
Everyday pricing, promotions, markdown strategy and competitor response.
Decisions made here
- What do we charge?
- What do we promote?
- When do we mark down?
Where it breaks today
- Markdown decided late, when the margin has already gone
- Promotions judged on uplift rather than on incrementality
- Competitor pricing collected manually
What is measured
- Gross margin
- Markdown rate
- Sell-through
Where AI landsMarkdown timing optimisation, incrementality measurement and competitor price monitoring.
Gross marginMarkdown rateSell-through
03
Allocate & replenish
Allocation to stores, replenishment, safety stock and inbound flow.
Decisions made here
- Where does stock go?
- How much do we hold?
- What replenishes tonight?
Where it breaks today
- Allocation on store grade rather than on local demand
- Replenishment driven by a stock figure that is wrong
- Safety stock set once and never revisited
What is measured
- Availability
- Stock turn
- Lost sales
Where AI landsCluster-aware allocation, phantom-inventory-adjusted replenishment and safety stock tuning.
AvailabilityStock turnLost sales
04
Run the store
Labour scheduling, task execution, shelf compliance and store standards.
Decisions made here
- Who works when?
- What gets done today?
- Is the plan actually executable?
Where it breaks today
- Tasks issued centrally with no view of available hours
- Compliance checked by visit rather than by evidence
- Store teams receiving more tasks than the roster supports
What is measured
- Task completion
- Labour cost to sales
- Compliance score
Where AI landsExecutable task generation within rostered hours and evidence-based compliance checking.
Task completionLabour cost to salesCompliance score
05
Serve the customer
Service, availability at the moment of purchase, loyalty and substitution.
Decisions made here
- Is it on shelf?
- What do we offer instead?
- What is this customer worth?
Where it breaks today
- Availability measured in the system rather than on the shelf
- Substitution left to whoever is on the aisle
- Loyalty value invisible at the point of service
What is measured
- On-shelf availability
- Basket size
- Customer retention
Where AI landsOn-shelf availability inference, substitution guidance and value-aware service.
On-shelf availabilityBasket sizeCustomer retention
06
Loss & returns
Shrink, waste, returns, refunds and the fraud around them.
Decisions made here
- Where is stock going?
- Is this return legitimate?
- What is waste costing us?
Where it breaks today
- Shrink measured at count, long after the cause
- Return fraud detected inconsistently store by store
- Waste treated as a cost of doing business
What is measured
- Shrink rate
- Waste value
- Return rate
Where AI landsShrink pattern detection, return anomaly review and waste driver analysis.
Shrink rateWaste valueReturn rate
Who you are building for
The store manager is the person every central decision eventually lands on. Select one to light the stages they own.
What the sector is seeing
A business with excellent transaction data and a persistent gap between system and shelf.
IndustryTiming over depth
Markdown timing beats markdown depth
Most margin is decided by when the markdown starts rather than by how deep it goes, which makes timing the higher-value modelling problem.
IndustryHours, not intentions
Labour is the execution constraint
Central task generation that ignores rostered hours produces plans stores cannot execute, and the tasks that get dropped are chosen by whoever is on shift.
IndustryDemand transfers
Transferable demand changes delisting
Removing an item does not remove its demand, and treating delisting as pure loss systematically under-ranges.
IndustryOn-shelf, not in-system
Availability moved from system to shelf
Measuring what the system holds is easy and increasingly seen as insufficient; what the shopper can actually pick up is the number that matters.
IndustryExplainable by design
Loss analytics carries a fairness obligation
Shrink and return analytics touch individual staff and customers, so explainability and proportionality are design requirements rather than afterthoughts.
02
Part 2 · Data and meaning
Domains, ontology, taxonomy and governed terms
The defining problem is the gap between the inventory record and the shelf. Almost every retail use case is either exploiting good transaction data or compensating for that gap.
The data landscape
Six families. The second one is the one that lies to you.
Transactions
Line-level sales, baskets, tenders, returns and refunds.
POS, transaction logInventory
Stock by item and location, movements, counts and adjustments.
Inventory systems, RFID, countsProduct & range
Items, hierarchy, attributes, assortment and store clustering.
Merchandising systems, PIMPrice & promotion
Prices, markdowns, promotional mechanics and competitor prices.
Pricing systems, competitor feedsStore & labour
Rosters, tasks, completion, compliance and store attributes.
Workforce management, task systemsCustomer & loyalty
Identified customers, loyalty, consent and interaction history.
Loyalty platform, CRMThe semantic layer
A retail agent that treats the stock figure as a measurement rather than as a record will recommend confidently and wrongly, every day, on exactly the items that are selling. These are the domains, the graph and the definitions it must be grounded on.
Data domains
Six subject areas, each with the entities it holds, its critical data elements and the function accountable for it.
Product & range
What is sold, and where it is ranged.
ItemVariantCategoryAssortment
- Critical data elements
- Item code; Barcode; Category node
- Owned by
- Merchandising
Location
Where it is sold, and which locations behave alike.
StoreClusterRegionZone
- Critical data elements
- Store identifier; Cluster assignment; Trading area
- Owned by
- Store operations
Inventory
What the system believes is where, and how it got there.
Stock positionMovementAdjustmentCount
- Critical data elements
- On-hand quantity; Movement type; Last count date
- Owned by
- Supply chain
Trade & price
What it costs and what it is being sold for.
PriceMarkdownPromotionSupplier term
- Critical data elements
- Retail price; Markdown depth; Promotion mechanic
- Owned by
- Trading
Transaction
What the shopper actually bought or brought back.
TransactionBasketLineReturn
- Critical data elements
- Units sold; Transaction timestamp; Tender type
- Owned by
- Retail operations
Labour & task
Who is available and what they were asked to do.
RosterShiftTaskCompletion
- Critical data elements
- Rostered hours; Task type; Completion status
- Owned by
- Store operations
Ontology & knowledge graph
The entities and the typed relationships between them — what the agent traverses instead of guessing joins. Select any entity.
Core entityEventReference / classification
Core entity
Item
The lowest-level sellable unit.
- Item classified in Category
- Stock position held for Item
- Transaction sells Item
- Price applies to Item
- Promotion discounts Item
Reference / classification
Category
The classification the item competes within.
- Item classified in Category
Core entity
Store
A selling location.
- Store assigned to Cluster
- Stock position held at Store
- Transaction occurs at Store
- Task issued to Store
Reference / classification
Cluster
A grouping of stores that behave alike.
- Store assigned to Cluster
Core entity
Stock position
What the system believes is on hand.
- Stock position held for Item
- Stock position held at Store
- Movement changes Stock position
Event
Movement
A recorded change to stock.
- Movement changes Stock position
Event
Transaction
A purchase at the till.
- Transaction sells Item
- Transaction occurs at Store
- Return reverses Transaction
- Promotion lifts Transaction
Event
Return
Goods brought back by a shopper.
- Return reverses Transaction
Reference / classification
Price
The retail price in force.
- Price applies to Item
Event
Promotion
A time-bound price or display mechanic.
- Promotion discounts Item
- Promotion lifts Transaction
Event
Task
Work issued to a store.
- Task issued to Store
- Roster bounds Task
Reference / classification
Roster
The hours actually available in a store.
- Roster bounds Task
Taxonomies
The classification hierarchies that make records comparable across systems.
Product hierarchy
- Division
- Category
- Sub-category
- Item
Grocery → Dairy → Yoghurt → 4-pack strawberry
Store clustering
- Region
- Format
- Cluster
South → Convenience → High-footfall commuter
Shrink cause
- Class
- Cause
- Addressable
Process → Mis-scan at till → Addressable by training
Governed business terms
The definitions an agent must use rather than invent. Most wrong answers in this industry are a term used loosely.
- Stock position vs on-shelf
- The stock position is a record; on-shelf is what a shopper can pick up. They diverge continuously and the divergence is the domain problem.
- Phantom inventory
- Stock the system believes is on shelf and which is not, suppressing replenishment silently.
- On-shelf availability
- The measure that reflects lost sales, as distinct from system availability which does not.
- Transferable demand
- Demand that moves to another item on delisting, rather than being lost.
- Sell-through
- The share of bought stock sold within a period; the input to markdown timing.
- Store cluster
- A behavioural grouping of stores, used instead of a static grade.
- Rostered hours
- The labour actually available, and therefore the bound on what any plan can ask for.
- Shrink
- Inventory lost to theft, damage or process error, usually quantified only at count.
03
Part 2 · The architecture
How it is actually built
The layers a deployment needs, what each one holds, and what the engineer owns there.
The reference architecture
Seven layers, with the inventory-truth problem addressed explicitly rather than assumed away.
01/07
POS, inventory & store sources
Transactions, the inventory record, range and pricing, plus rosters and task completion.
Takes in
Nothing. This is the upstream edge.
Hands on
Trustworthy transactions, stock figures carrying a confidence, and the roster capacity bound.
SystemWhat it doesKey business entities
Point of sale
Line-level transactions, baskets, tenders and returns. High volume, and the one retail dataset that is genuinely trustworthy.
TransactionBasketTenderReturn
Inventory management
Stock by item and location with movements, counts and adjustments. Drifts from physical reality continuously, which is the central problem of the domain.
Stock positionMovementAdjustmentCount
Merchandising systems
Range, assortment, product hierarchy, store attributes and clustering.
ItemAssortmentStoreCluster
Pricing and promotions
Everyday prices, markdowns, promotional mechanics and competitor price feeds.
PriceMarkdownPromotion
Workforce management
Rosters, available hours, task issue and completion. The constraint that decides whether any central plan is executable at all.
RosterTaskCompletionHours
Loyalty and customer
Identified customers, loyalty state, consent and interaction history.
CustomerLoyalty accountConsent
What the engineer owns here
Understanding how the inventory record is maintained, and therefore how it fails.
Where this layer fails
Treating the stock figure as truth. It is a record that drifts through theft, damage, mis-scans and misplacement, and a replenishment system that believes it will silently stop ordering the items that are selling fastest.
02/07
Ingestion & inventory truth
High-volume transaction ingestion plus inference of where the stock record is untrustworthy.
Takes in
Read-mostly access paths, change feeds and document streams from the systems of record.
Hands on
Replayable, resolved, quality-checked records with their lineage back to the source row.
SystemWhat it doesKey business entities
Change data capture
Reads the source’s own change log rather than polling it, so the platform sees every state a record passed through instead of only where it ended up.
Change eventOperation typeCommit timestamp
Landing and raw zone
Every extract kept as it arrived, immutable and timestamped. The thing you replay from when a downstream definition turns out to have been wrong.
Raw extractBatchWatermark
Document processing
Layout-aware extraction over the unstructured half of the estate: sectioning, tables, signatures, and the page reference every later citation depends on.
DocumentSectionPage reference
Entity resolution
Decides that two records are the same real-world thing, with a survivorship rule and a confidence, so downstream joins are a decision rather than an assumption.
Match candidateSurvivorship ruleCluster
Data quality and contracts
Schema, freshness and volume expectations asserted at the boundary, so a bad load fails loudly here rather than quietly two layers later.
ContractExpectationViolation
Phantom inventory inferencethis domain adds
Infers where the stock record has diverged from the shelf, and attaches a confidence to the figure rather than passing it through unchanged.
Stock positionConfidenceGap signal
What the engineer owns here
Producing a confidence on the stock figure, rather than passing it through unchanged.
Where this layer fails
Passing the stock figure downstream as though it were a measurement. It is a record, it drifts, and replenishment built on it stops ordering exactly what is selling.
03/07
Range model & semantics
Product, store, cluster and calendar as governed dimensions with agreed definitions.
Takes in
Replayable, resolved records with lineage from ingestion.
Hands on
Typed, classified, defined data an agent can be grounded on without inferring meaning.
SystemWhat it doesKey business entities
Business glossary
One agreed definition per term, owned by a named person, so the number an agent quotes means what the business means by it.
TermDefinitionOwnerSteward
Ontology and knowledge graph
The typed entities and relationships the domain actually has, so an agent can traverse "which customers are exposed to this" rather than guess from adjacent text.
EntityRelationshipNodeEdge
Semantic and metrics layer
Metrics defined once, in one place, with their filters and grain. Removes the class of error where the agent computed something plausible and wrong.
MetricDimensionGrainFilter
Classification and policy
Sensitivity labels and the purpose each classification permits, applied at the field level and inherited by everything downstream.
ClassificationPurposePolicy
Lineage and catalogue
Where a value came from and what it touched on the way. The layer that makes an answer defensible rather than merely correct.
Lineage edgeDatasetColumn
What the engineer owns here
One definition of availability, agreed between supply chain and stores.
Where this layer fails
Letting the model infer meaning from column names. It will, it will be plausible, and nobody will notice until the number reaches a regulator or a board pack.
04/07
Retrieval & grounding
Retrieval over supplier agreements, range review packs, policies and store procedures.
Takes in
Typed, classified, defined data from the governance layer.
Hands on
Ranked, permission-trimmed, citable evidence scoped to the caller and the moment.
SystemWhat it doesKey business entities
Hybrid index
Vector and lexical retrieval together, because exact identifiers, codes and clause numbers are the thing semantic search is worst at.
ChunkEmbeddingTermScore
Chunking and metadata
Splits on the document’s own structure and carries effective dates, version and source into every chunk, so a retrieved passage knows when it was true.
ChunkEffective dateVersionSource reference
Security trimming
Applies the caller’s permissions inside the query rather than filtering results afterwards, so the model never sees what the user may not.
PrincipalEntitlementFilter
Graph retrieval
Traverses the knowledge graph for questions that are joins rather than similarity — exposure, genealogy, ownership, causation.
PathHopTraversal
Reranking and citation
Reorders candidates on relevance and returns the passage identifier behind every sentence, so the answer can be checked rather than trusted.
CandidateRankCitation
What the engineer owns here
Grounding a recommendation in the supplier term or store policy that governs it.
Where this layer fails
Filtering results after retrieval instead of constraining the query. The model has already seen the rows you removed, and it will use them.
05/07
Agent & orchestration
Agents for phantom detection, markdown timing, task generation and loss review.
Takes in
Ranked, permission-trimmed, citable evidence from retrieval.
Hands on
Actions taken or proposed, each with the evidence, the identity and the trace behind it.
SystemWhat it doesKey business entities
Agent runtime
Plans, calls tools and holds the loop. Where the step budget, the timeout and the stopping condition are enforced rather than hoped for.
TaskPlanStepTrace
Tool and MCP servers
Typed, permissioned tools with declared schemas. An agent’s real capability surface is this list, which is why the list is a security artefact.
ToolSchemaScopeInvocation
State and memory
Durable task state with checkpoints, so a run that dies mid-way resumes instead of restarting and re-doing side effects.
CheckpointSessionMemory record
Human-in-the-loop gates
Approval steps on the actions that need one, carrying enough context for the approver to actually decide rather than rubber-stamp.
ApprovalThresholdDecision
Workflow integration
Writes back into the systems people already work in, under a service identity with its own audit trail.
Work itemWrite-backActor
Capacity-bounded task generationthis domain adds
Bounds generated store work by rostered hours, so the plan issued is the plan that can actually be executed.
TaskRoster hoursCapacity
What the engineer owns here
Generating tasks that fit the roster, because an unexecutable task is worse than none.
Where this layer fails
Issuing more tasks than the roster supports. The list is then triaged by whoever is on shift, and you will never know which half was dropped.
06/07
Fairness, privacy & guardrails
Loss analytics fairness, customer consent, staff monitoring limits and audit.
Takes in
Proposed actions and drafted responses from the agent layer.
Hands on
Permitted, grounded, logged output — or a refusal with a stated reason.
SystemWhat it doesKey business entities
Policy enforcement
The rules that decide whether an action is permitted at all, evaluated before the action and independently of the model that proposed it.
PolicyDecisionObligation
Input and output screening
Injection detection on the way in, and on the way out the checks for leakage, unsupported claims and content the domain forbids.
PromptResponseViolation
Groundedness checking
Verifies each assertion resolves to retrieved evidence, and fails the response rather than shipping the sentence that does not.
AssertionSupportVerdict
Model risk management
Model inventory, intended use, validation evidence and the sign-off that lets a model be used for a purpose. Not optional in a regulated estate.
ModelIntended useValidationApproval
Audit log
Immutable record of what was asked, retrieved, decided and done — the artefact a reviewer reads when they do not take your word for it.
Audit entryActorEvidence reference
Loss analytics fairnessthis domain adds
Tests shrink and return models for systematic disparity, and requires findings to point at process before they point at people.
CohortDisparityReview
What the engineer owns here
Ensuring shrink analytics point at process before people, and are explainable when they do not.
Where this layer fails
Shrink analytics that name individuals first. It is the fastest way to turn a data programme into an employee-relations incident.
07/07
Evaluation & observability
Availability lift, phantom precision, task completion and cost per store.
Takes in
Permitted, grounded, logged output from the guardrail layer.
Hands on
Measured quality, cost and latency — and the evidence to change any of the three.
SystemWhat it doesKey business entities
Evaluation harness
Held-out sets and graded runs on every change, so a prompt edit is a measured change rather than a hopeful one.
Eval setCaseScoreBaseline
Regression gate
Blocks a release when quality drops, in CI, on the same evidence for everyone. The difference between a system and a demo.
ThresholdRunGate decision
Tracing
End-to-end spans across retrieval, model and tool calls, so a bad answer can be opened and read rather than argued about.
TraceSpanLatencyToken count
Cost and capacity
Cost per task and per tenant, against throughput and quota. The number that decides whether the pilot can become the rollout.
Cost per taskQuotaThroughput
Drift and feedback
Watches quality against production traffic rather than the test set, and routes real corrections back into the eval suite.
Drift signalFeedbackIncident
What the engineer owns here
Proving on-shelf availability moved, not that the dashboard did.
Where this layer fails
Evaluating once, before launch. Quality moves with the data, the model and the traffic, and a system with no live measurement has no idea which of the three moved.
Typical tech stack
Models
- Frontier models for agreement and policy reasoning
- Small models at transaction volume
- Demand and markdown models
Grounding
- Supplier agreement index
- Store policy and procedure corpus
- Range review history
Agents
- Phantom detection agents
- MCP over inventory and task systems
- Markdown timing agents
Data
- Transaction ingestion at store volume
- Movement reconciliation
- Store clustering
Assurance
- Phantom precision measurement
- Fairness testing on loss analytics
- Availability measurement
Runtime
- Store-edge tolerance
- Overnight batch windows
- Cost per store budgeting
Constraints that shape the design
The inventory record is not the shelf
Stock figures drift from reality through theft, damage, mis-scans and misplacement. Any system that treats the record as truth will silently suppress replenishment on exactly the items that are selling.
Labour bounds executability
A store has a fixed number of rostered hours. Task generation that ignores them produces a plan that is silently triaged by whoever is on shift, and you will not know which parts survived.
Loss analytics touches people
Shrink and return models implicate staff and customers. Proportionality, explainability and pointing at process before individuals are design requirements.
Volume is high and margins are thin
Per-transaction inference is not affordable at retail scale. Reserve reasoning for decisions, not for records.
04
Part 3 · Getting it done
Where the business and the technology meet
The most valuable use cases here are the ones that stop trusting a number the whole business has been trusting for decades.
Use cases, mapped to the chain and the architecture
Filter by stage or by earned autonomy. Selecting a use case jumps the value chain and the architecture to the stage and layer it depends on.
Value chain stage
Autonomy earned
12 use cases
No use cases match that combination.
The delivery playbook
How a retail deployment actually runs.
01
Week 1 — find out how wrong the stock figure is
Measure inventory record accuracy against physical reality on a sample. Every replenishment and availability use case is capped by this number.
02
Week 2 — agree what availability means
Supply chain and stores must agree one definition, including whether it is measured in the system or on the shelf. They usually mean different things by it.
03
Weeks 3-6 — start with phantom detection
It needs only transaction data, recovers sales nobody knew were lost, and produces the inventory confidence every later use case depends on.
04
Weeks 7-9 — take on tasks only with roster data
Do not generate store tasks until you have rostered hours. Without them you are adding to a queue that is already being triaged without you.
05
Weeks 10-12 — fairness, availability proof and handover
Test loss analytics for disparity, prove on-shelf availability actually moved, then hand over runbook, thresholds and the definitions agreed in week two.
The curriculum
8 modules, 80 taught hours, 40 hands-on labs and 8 assessments — every lab provisioned and graded by the SCIKIQ Agentic AI Playground. Open a module to see its labs.
01The SCDAI delivery model in retailLanding in a business where the store, not the head office, decides what actually happens.10 hrs · 5 labs
Delivery mandateSystem vs shelfValue sizing on availability and marginStakeholder mapStore adoptionThin-slice scoping
01
Decision selectionScore four retail decisions on value, cadence and data readiness.
02
Value sizingSize a lost sale, a late markdown and a point of shrink.
03
Inventory accuracy auditMeasure stock record accuracy against a physical sample.
04
Stakeholder mapMap buying, pricing, replenishment, stores and loss prevention.
05
Thin sliceScope a one-category, one-cluster slice with a measurable outcome.
Assessment 01 — Operating model & scoping
02Retail data foundationsTransactions you can trust, inventory you cannot, and the labour data nobody shares.10 hrs · 5 labs
Transaction data at volumeInventory movements and adjustmentsRange and clusteringPrice and promotion dataRoster and task dataLoyalty and consent
01
Transaction ingestionIngest line-level transactions within a cost budget.
02
Movement reconciliationReconcile stock movements and find where the record breaks.
03
Store clusteringBuild store clusters from demand behaviour rather than from grade.
04
Roster ingestionIngest rostered hours and model available task capacity.
05
Consent handlingApply loyalty consent state at the point of retrieval.
Assessment 02 — Retail data landscape
03Context engineering for retail languageItem codes, store codes, units and the arithmetic of packs and cases.10 hrs · 5 labs
Item and store resolutionUnit and pack handlingStructured outputsConfidence on inferred stockRefusal on individual attributionContext budget
01
Item resolutionResolve items across supplier, internal and barcode identifiers.
02
Pack arithmeticConvert case, pack and unit without silent error.
03
Structured findingsEmit a phantom finding with confidence as a typed structure.
04
Attribution refusalMake the agent refuse to attribute shrink to an individual.
05
Context budgetCut context cost on a chain-wide nightly run.
Assessment 03 — Context engineering
04Grounding on agreements and store policyWhat the supplier owes, and what the store is actually allowed to do.10 hrs · 5 labs
Supplier agreement retrievalPolicy and procedure retrievalRange review historyProduct graph traversalCitationsRetrieval evals
01
Agreement retrievalRetrieve the operative supplier term for an item and period.
02
Policy retrievalAnswer a store-procedure question with the policy cited.
03
Range historyRetrieve what happened last time this item was delisted.
04
Citation contractCite agreement or policy and date on every recommendation.
05
Retrieval evalsBuild a buyer-labelled question set and measure precision.
Assessment 04 — Retrieval & grounding
05Retail agent designDetecting what the record hides, and generating work that can actually be done.10 hrs · 5 labs
Workflow vs agentPhantom detection patternsConfidence propagationCapacity-bounded task generationFairness in loss analyticsEscalation
01
Phantom agentDetect phantom inventory from transaction gap patterns.
02
Confidence propagationCarry stock confidence through into replenishment output.
03
Capacity-bounded tasksGenerate a task list that fits the rostered hours.
04
Fairness designDesign shrink analytics that point at process before people.
05
EscalationDefine when a loss finding requires human review and prove it does.
Assessment 05 — Agent design for retail
06Integrating with inventory, pricing and task systemsRead the chain, write to the task list, and never straight to the price.10 hrs · 5 labs
MCP over inventory and pricingTask system write-backBatch windowsStore-edge toleranceIdentityError handling
01
MCP over inventoryExpose stock, movement and range lookups as typed read tools.
02
Task write-backCreate store tasks under a traceable service identity.
03
Batch windowsFit a chain-wide run inside the overnight window.
04
Store edgeHandle a store that loses connectivity during the trading day.
05
IdentityBind agent access to banner, region and role entitlements.
Assessment 06 — Systems integration
07Fairness, privacy and securityLoss analytics and staff monitoring are where retail AI causes real harm.10 hrs · 5 labs
Fairness testingProportionalityStaff monitoring limitsCustomer consentOWASP LLM Top 10Incident response
01
Fairness testingTest shrink analytics for systematic disparity across stores and staff.
02
Proportionality reviewAssess whether a monitoring design is proportionate to the loss.
03
Consent enforcementApply loyalty consent inside retrieval, not at send time.
04
Injection defenceDefend against instructions embedded in supplier content.
05
Red teamRun an OWASP LLM Top 10 pass over the loss pipeline.
Assessment 07 — Fairness, privacy & security
08Production, availability and cost per storeOn-shelf availability is the number. A dashboard that moved without it is a failure.10 hrs · 5 labs
Eval suitesAvailability measurementDriftPhantom precisionCost per storeHandover
01
Eval gateGate deployment on phantom precision regression.
02
Availability studyMeasure on-shelf availability before and after, on the shelf.
03
Drift watchDetect drift after a range change or a store refit.
04
Cost per storeMeasure and reduce cost per store across the estate.
05
HandoverProduce runbook, thresholds and the agreed availability definition.
Assessment 08 — Production & outcomes
Two capstones — 20 hours
End-to-end agent design, development, deployment and testing, 10 hours each. Reviewed by a Senior SCDAI Engineer against a published rubric.
Capstone A — Phantom inventory agent
Build an agent that detects items the inventory record shows on shelf but which have stopped selling abnormally, distinguishes them from genuine slow sellers, propagates a confidence into the replenishment output, and is measured on precision against a physically counted set.
Capstone B — Executable task agent
Build an agent that generates a store’s daily task list bounded by actual rostered hours, prioritised by measured sales impact, citing the store policy governing each task — and demonstrably never issuing more work than the roster supports.
What you deliver, and how it is marked
- A running agent, deployed on the program's target stack, reachable by a reviewer.
- An architecture note: boundaries, tools, autonomy level and the trade-offs you took.
- A versioned eval set with a baseline and your final scores.
- A trace walkthrough of one success and one deliberate failure.
- Cost per task, measured — not estimated.
- A runbook and rollback path a client team could operate on Monday.
Problem fit15%Is this a decision worth automating, scoped to a defensible boundary?
Architecture20%Right pattern for the constraints; autonomy earned, not assumed.
Grounding & accuracy20%Measured groundedness against the eval set, with the failures named.
Safety & compliance20%Domain controls actually enforced, and demonstrably so.
Operability15%Traces, runbook, rollback, cost — could someone else run this?
Communication10%Can a client executive follow the reasoning in ten minutes?
One lab, start to finish
Every lab in this program follows the shape below. This is lab 05 in full — the brief you are given, the environment that is provisioned for you, the code you start from and the assertions that decide whether you passed.
Module 05 · Retail agent design
Finding stock that is not there
Detect phantom inventory from sales behaviour, without flagging every slow seller.
The brief
You are given 12 weeks of line-level sales, the stock record and the movement history for 40,000 item-store combinations, plus a physically counted set. Some items stopped selling because they are not physically on the shelf; most stopped because nobody wanted them. Return the phantoms, with a confidence, and propagate that confidence into the replenishment signal.
Provisioned for you
- 40,000 item-store combinations, 12 weeks of sales
- Stock record and movement history
- Physically counted validation set (1,200 combinations)
- Replenishment interface stub
You start from this
def detect_phantoms(sales, stock, movements):
"""Return [(item, store, confidence, evidence)] for suspected phantoms.
A slow seller and a phantom look identical in a stock figure. The
difference is in the SHAPE of the sales stop, against that item's own
prior rate of sale at that store.
"""
# 1. baseline rate of sale per item-store, not per item
# 2. a phantom is an abrupt stop while stock is still recorded positive
# 3. never return a bare flag -- return the evidence that produced it
raise NotImplementedErrorWhat the grader asserts
- Precision at or above 0.7 against the physically counted set
- Slow sellers with a declining but continuous curve are not flagged as phantoms
- Baseline is computed per item-store, not per item across the chain
- Every flag carries the evidence and a confidence, never a bare boolean
- The replenishment signal receives the confidence, not a corrected stock figure
05
Assessment
Assessed, not attended
Every module ends with a timed, randomised assessment delivered through the SCIKIQ Agentic AI Playground. The certificate requires a pass on all of them plus two reviewed capstones.
Assessment blueprint
What each test covers, how long it runs, and how many items are currently in the versioned bank behind it.
| # | Assessment & coverage | Items | Time | In bank |
|---|---|---|---|---|
| 01 | Operating model & scopingDelivery mandateSystem vs shelfValue sizingThin slicing | 25 | 35 min | 3 |
| 02 | Retail data landscapeTransactionsInventory movementsClusteringLabour data | 25 | 35 min | 2 |
| 03 | Context engineeringIdentifier resolutionPack arithmeticConfidenceRefusal | 25 | 35 min | 2 |
| 04 | Retrieval & groundingAgreement retrievalPolicy retrievalRange historyRetrieval metrics | 25 | 40 min | 2 |
| 05 | Agent design for retailPhantom detectionConfidenceCapacity boundingFairness | 25 | 40 min | 2 |
| 06 | Systems integrationMCP designWrite-back safetyBatch windowsIdentity | 25 | 35 min | 2 |
| 07 | Fairness, privacy & securityFairnessProportionalityConsentLLM security | 30 | 45 min | 2 |
| 08 | Production & outcomesEval gatingAvailabilityPhantom precisionEconomics | 30 | 45 min | 2 |
Single best answerOne defensible option among plausible distractors drawn from real field mistakes.
Scenario judgementA client situation with constraints; you pick the action a Senior SCDAI Engineer would take.
Architecture selectionGiven non-functional requirements, choose the pattern and justify the trade-off.
Failure diagnosisA trace, an eval report or a cost curve — identify the root cause.
- Each module ends with a timed assessment; the program certificate requires a pass on every one.
- Items are drawn at random from a versioned bank, so no two learners sit an identical paper.
- Pass mark is 70%. Two retakes are included, each with a fresh draw.
- Scenario items carry double weight — they are the ones that predict field performance.
- Capstones are reviewed against a published rubric by a Senior SCDAI Engineer, not auto-graded.
Try four questions from the bank
Real items, drawn from 17 in this program's bank — weighted toward the scenario and diagnosis types, because those are the ones that predict field performance. Instant feedback, nothing saved.
Four sample items — one attempt each, then the reasoning is shown.
Q1A retailer asks for AI to improve availability. What do you measure in week one?
Every availability and replenishment use case is capped by inventory record accuracy. Measuring it first sets an honest ceiling instead of a later surprise.
Q2Should a rate-of-sale baseline be computed per item or per item-store?
A chain-level baseline flags every slow store as anomalous and misses genuine stops in fast ones. The baseline has to be local to be meaningful.
Q3Why should an agent refuse to attribute shrink to a named individual?
Shrink patterns have many innocent explanations. Attribution is an HR process with its own standards, and an agent that pre-empts it causes real harm.
Q4A phantom model achieves high recall by flagging 15% of item-store combinations. Is this useful?
The constraint is store labour. A model tuned for recall against a workforce that can check twenty items a day produces nothing but noise.
06
Career & outcomes
What this makes you, and where it goes next
The same ladder whichever specialisation you enter through — what changes is the domain you go deep in. Below: how the program is delivered, the skills it moves, the roles it leads to, and the specialisations closest to this one.
Available two ways
The same labs, assessments and capstones, delivered to an enterprise cohort or to individual professionals.
B2B
Direct to enterprise
Cohorts of 20 to 2,000+ on your own tenancy, with your data patterns and your cloud. Skill-gap baselining up front, per-team mastery reporting throughout, and capstones scoped against your real backlog so the output is deployable work.
- Private cohort on your tenancy
- Baseline skills assessment before kick-off
- Capstones scoped to your backlog
- Per-team mastery dashboards for L&D
- Optional Senior SCDAI Engineer architecture reviews
B2C
Direct to individual professionals
The same labs, assessments and capstones for individual engineers and analysts, run on shared infrastructure with a fixed cohort calendar. You leave with a graded portfolio, not a certificate of attendance.
- Fixed-date cohorts, self-paced within them
- Same graded labs and item bank
- Portfolio-ready capstones you own
- Certificate on a pass in every module
- Community and office-hours access
Skill matrix
Find your row and aim one column right. The Playground scores you against this after every module.
| Skill | Beginner | Intermediate | Advanced |
|---|---|---|---|
| Retail domain fluency | Knows the functions. | Maps range, price and stock decisions to margin. | Sizes availability, markdown and shrink value credibly. |
| Inventory truth | Reads a stock figure. | Detects where the record and the shelf diverge. | Propagates stock confidence through every downstream decision. |
| Executable operations | Generates a task. | Bounds task generation by rostered capacity. | Proves plans were executed rather than silently triaged. |
| Context engineering | Writes clear prompts. | Structures retrieval, tools and state deliberately. | Designs context strategy for reliability and cost at scale. |
| Retrieval & grounding | Builds basic vector search. | Tunes chunking, hybrid search and reranking. | Designs graph + vector grounding with measured recall. |
| Agent orchestration | Runs a single tool-calling agent. | Builds supervised multi-step and multi-agent flows. | Designs autonomy boundaries and failure containment. |
| Tool & system integration | Calls a documented API. | Writes an MCP server over a system of record. | Designs a least-privilege tool estate across systems. |
| Evaluation | Eyeballs outputs. | Builds labelled eval sets and regression gates. | Runs online evals with drift and judge calibration. |
| Observability & cost | Reads logs. | Traces runs, tracks tokens and latency. | Owns cost per task and capacity planning in production. |
| Security & guardrails | Adds output filters. | Mitigates the OWASP LLM Top 10 in a build. | Threat-models an agent estate and proves controls. |
| Client delivery | Takes notes in a workshop. | Runs discovery and scopes a thin slice. | Owns the account technically, from scope to handover. |
Where it takes you
The SCDAI ladder is the same whichever specialisation you enter through — what changes is the domain you go deep in.
01
Data and AI EngineerEntry
Python, LLM APIs, first RAG build, prompt and context basics.
02
Associate Data and AI EngineerPractitioner
Client-facing discovery, tool integration, agent build under supervision.
03
SCIKIQ Certified Data and AI EngineerSpecialist
Owns a deployment end to end — scoping, build, evals, handover.
04
Senior Data and AI Engineer / Domain LeadSenior
Multi-agent architecture, regulated deployments, reference patterns for the practice.
05
AI Architect / Practice LeadLeadership
Portfolio of accounts, delivery standards, hiring and capability strategy.
Terms worth knowing
- Phantom inventory
- Stock the system believes is on shelf but which is not there.
- On-shelf availability
- What the shopper can actually pick up, as distinct from the system figure.
- Transferable demand
- Demand that moves to another item when one is delisted, rather than being lost.
- Sell-through
- The share of bought stock sold within a period; the input to markdown timing.
- Store cluster
- A grouping of stores that behave alike, used instead of a static grade.
- Shrink
- Inventory lost to theft, damage or error, usually measured only at count.
- Markdown
- A permanent price reduction to clear stock; timing decides most of its cost.
- Rate of sale
- Units sold per store per week, the comparable across different store counts.
- SCDAI
- SCIKIQ Certified Data and AI Engineer — the credential this program awards.
- SCDAI engineer
- An engineer embedded with the customer who turns their constraints into a working system.
- Thin slice
- The narrowest end-to-end path through a workflow that still produces business value.
- Agentic RAG
- Retrieval where the agent decides when, what and how often to retrieve.
- MCP
- Model Context Protocol — open standard connecting an agent to tools and data.
- A2A
- Agent2Agent — protocol for agent-to-agent delegation across vendors.
- Eval set
- A labelled, versioned set of cases an agent must pass before release.
- LLM-as-judge
- Using a model to score outputs, calibrated periodically against human ratings.
- Guardrail
- An enforced constraint on an input, output or action — not a prompt request.
- Human-in-the-loop
- A required human approval step before a consequential action executes.
- Cost per task
- Total token, tool and compute cost of one completed unit of work — the engineer's unit economic.
- Golden path
- The reference architecture a practice standardises on so deployments stay reviewable.
- Handover
- The point at which the client can operate, evaluate and change the system without the engineer.
Related programs
Skill a team, or join a cohort
B2B cohorts run on your tenancy with capstones scoped to your backlog. B2C cohorts run on a fixed calendar.
07
Enrolling
What you need, what you get, and what it costs
Stated plainly enough to rule yourself in or out without a sales call: the prerequisites, how the program runs, exactly what the credential is worth, and the questions everyone asks.
Before you start
Stated plainly so you can rule yourself in or out without a sales call. Nothing here is a formal qualification — it is what the first lab assumes you can already do.
Technical
You should already be able to do these
- Comfortable in Python: functions, typing, virtual environments, reading a stack trace
- Have called an LLM API and handled a structured response
- Git, the command line, and reading someone else’s code without panic
Domain
What we assume, and what we teach
- Basic commercial literacy: what range, markdown and replenishment mean
- Comfort with the idea that the inventory record is frequently wrong
Time and access
What the program asks of your week
- A machine that can run a browser and a terminal — labs run in the Playground, not locally
- Roughly 8 hours a week for the taught weeks, plus 20 hours for the two capstones
How it runs
Cohort dates and pricing are confirmed on enquiry rather than printed here, because both move with the intake.
- Format
- Self-paced within a fixed cohort window, delivered in the SCIKIQ Agentic AI Playground
- Taught hours
- 80 hours across 8 modules
- Capstones
- 20 hours — two end-to-end builds, reviewed not auto-graded
- Typical duration
- 10–12 weeks at around 8 hours a week
- Level
- Intermediate → Advanced
- Built for
- Engineers and analysts deploying AI across retail merchandising and store operations
- Next cohorts
- — confirmed on enquiry
- Pricing
- — confirmed on enquiry
The credential
The credential is awarded per specialisation, so it names the domain or stack you were assessed in rather than claiming general competence. On this program the badge reads SCDAI — Retail.
How it is earned
- A pass at 70% or above on every module assessment — not an average across them
- Both capstones reviewed against the published rubric by a Senior SCDAI Engineer
- All labs submitted and machine-graded green
- Awarded as
- A verifiable digital badge with a public credential record
- Valid for
- Two years from award
- Recertification
- One current-year assessment plus a refreshed capstone
- Retakes
- Two included per assessment, each with a fresh item draw
- Named specialisation
- The badge states the domain or stack, e.g. SCDAI — Healthcare
- Verification
- Third parties confirm a holder from the credential record, no login required
A certificate that cannot be checked is decoration. Every award resolves to a record showing the specialisation, the award date and the assessments passed.
The certificate you leave with
This is the credential the program awards, shown exactly as it is issued — with the specialisation named, the assessment record attached and a verification link anyone can check without an account.
SCIKIQ Data Academy
Certificate of Professional Certification
This is to certify that
Your name
has been assessed and certified as
SCIKIQ Certified Data and AI Engineer
Retail
- Specialisation
- Retail
- Assessments passed
- 8 of 8, each at 70% or above
- Labs graded
- 40 of 40 machine-graded green
- Capstones reviewed
- 2, against the published rubric
- Taught hours
- 80 across 8 modules, plus 20 capstone hours
AwardedOn completion
Valid untilTwo years from award
Credential IDSCDAI-RETAIL-<issued>
Senior SCDAI Engineer, Certification Board
- 01The specialisation is on the certificate
Not “Data and AI Engineer” but the domain or stack you were actually assessed in. A general claim would be a weaker one.
- 02The assessment record travels with it
Modules passed, labs graded and both capstones reviewed — so the credential states what was measured rather than that you attended.
- 03Anyone can verify it, without a login
The credential ID resolves to a public record showing the specialisation, the award date and the assessments passed.
Share it on LinkedIn
Add it to your LinkedIn profile in one step. The link pre-fills the certification fields from the credential record, so the entry on your profile matches the record a reader can check.
Add to LinkedIn profile The button is live on your real certificate; here it opens LinkedIn pre-filled with this specialisation so you can see exactly what the profile entry will say.A certificate that cannot be checked is decoration. Every award resolves to a record showing the specialisation, the award date and the assessments passed.
Questions we get asked
The objections that come up in every conversation about this program, answered without the brochure voice.
Is this a course or a certification?
Both, and the second is the point. Eight timed assessments and two reviewed capstones stand between you and the credential, so a pass means someone measured the skill rather than recorded your attendance.
How is it self-paced without falling apart?
Every lab is provisioned, graded and unblocked by the Playground rather than by an instructor. That is what lets a cohort of 2,000 cost the same faculty time as a cohort of 20 — and why you are never waiting on someone to mark your work.
What if I fail an assessment?
Two retakes are included per assessment, each drawing a fresh item set from the bank, so a retake is a genuinely new paper rather than the same questions again.
Do I need my own cloud account?
For the domain specialisations, no — labs run in provisioned sandboxes. For the four tech-stack programs you will want access to that platform, since deploying into a real subscription is much of the point.
Can we run this privately for our team?
Yes. B2B cohorts run on your own tenancy with your data patterns, a skills baseline before kick-off, per-team mastery reporting, and capstones scoped against your actual backlog so the output is deployable work rather than an exercise.
Which specialisation should I take?
Take the domain you deploy into. If you move across industries, take a tech-stack program instead and pick up domain context on the engagement. The chooser on the programs page will narrow it.
How current is the material?
The trends, platform capabilities and regulatory positions are reviewed each quarter, and every external claim on these pages links to its source so you can check the date yourself.
What do I actually leave with?
A graded portfolio: forty machine-graded labs, two reviewed end-to-end agent builds with measured evaluation and cost per task, and a verifiable credential naming your specialisation.
Still deciding?
Tell us the systems you deploy into and we will say plainly whether this specialisation is the right one — or which of the 21 is.