Real-time decision intelligence for the connected retail enterprise
Customer, product, inventory, order, store, digital, supplier and pricing state all move at once. Knowing all of it is not the same as deciding what to do about it.
At a glance
State moves under the decision
Inventory, price, promotion and capacity change between evaluation and execution.
Decisions reach customers without a person in the loop
What once went to a store manager now runs at consumer scale.
The cost arrives as a reversal
Refunds, goodwill credits and failed promises follow actions taken on state that had already moved.
Boundaries carry the risk
Substitution, refund and priority thresholds are where outcomes flip.
One live retail state
- Customer
- Product
- Inventory
- Order
- Store
- Digital
- Supplier
- Price & promotion
ANANTDECISION
What should we do next?
- RECOMMEND
- REPLENISH
- TRANSFER
- REFUND
- HOLD
- MARKDOWN
NEXT BEST ACTION
A recommendation, made against the state above.
ANANTVERIFY
Should this decision be allowed to become an action?
ALLOW · HOLD · ESCALATE · BLOCK
Action
- commerce platform
- inventory & allocation
- store & POS
- fulfillment
- payments & fraud
- customer service
- human associate
Nothing here reaches a customer, a store or a payment system until an independent verdict has released it.
“Retail happens one decision at a time.”
Retail State → Decision Intelligence → Assurance → Next Best Action → Outcome
Retail happens one decision at a time
A retail enterprise answers the same questions thousands of times a day, at a clock and a margin. Each one looks like its own problem.
The questions a retailer answers every day
Each decision looks independent of the others
Every one of these is a decision, and every one of them is made against state that is already different by the time the next one is asked.
What should we offer this customer?
- Which product should we recommend?
- Should we approve this return?
- Should this order be released?
- Where should inventory be allocated?
- Should we change the price?
- Which store needs replenishment?
- How should we respond to a supplier delay?
- Is this transaction suspicious?
- Should this AI agent be allowed to act?
Each decision may look independent. It is not.
The decisions are connected
One decision changes the state the next one is made against
How the coupling runs
- Customer behavior
- Demand
- Inventory
- Fulfillment
- Customer experience
Read left to right, it is a supply chain. Read as a decision problem, every step is a decision someone has to make.
The coupling is not incidental. It is the reason a retail decision cannot be made well by one function looking at one system.
Also coupled
- pricing affects margin
- promotions affect demand
- supplier performance affects availability
- fraud controls affect conversion
Anant connects these decisions through a retail state that never stops changing.
From omnichannel retail to intelligent retail
Connectivity is not intelligence
What retailers have already connected
- stores
- e-commerce
- mobile
- marketplaces
- social
- loyalty
- fulfillment
- customer service
Omnichannel connectivity alone does not create intelligence. It creates a complete record of what already happened.
The next step is continuous decision intelligence.
Every change in any of these is a decision opportunity
- customer behavior
- inventory
- demand
- price
- promotion
- order
- supplier
- operations
See the change. Understand the context. Determine the next best action.
The retail enterprise is a dynamic system
A customer and a product are each a network, not a record
A customer is connected to
- identity
- household
- products
- orders
- returns
- loyalty
- offers
- channels
- service interactions
- behavior
A product is connected to
- inventory
- stores
- suppliers
- price
- promotion
- margin
- demand
- substitutes
- fulfillment
Do not optimize the customer, the inventory and the supply chain independently. Optimize the decision.
From retail data to retail state
Platforms store events. The state is what the events mean together.
Nine dimensions of retail state
Customer state
- Who is the customer?
- What is changing?
Product state
- What matters now?
- What is available?
Inventory state
- How much is there?
- Where is it?
Order state
- Purchased, shipped, returned, delayed?
- What is promised?
Store state
- What is happening at each location?
Digital state
- What is happening on web and mobile?
Price and promotion state
- Which offers apply?
- What are the economics?
Supply state
- What can suppliers deliver?
- By when?
Operational state
- What constraints affect execution?
Nine dimensions, read as one state at the moment a decision is made.
Traditional retail platforms record what happened, one event and one table at a time. A decision needs all of it at once, in the form the decision is actually made against.
Together, these are the retail state.
The layer that decides and the layer that assures
One connected retail state, two capabilities: an enterprise decision layer and an independent layer that decides whether a consequential decision may execute.
AnantDecision + AnantVerify
One layer answers what to do. Another answers whether it may happen.
Together
- OBSERVE
- DECIDE
- VERIFY
- ACT
- LEARN
The decision layer proposes against retail state. The assurance layer controls what reaches a customer, a store or a payment system.
AnantDecision determines
- next best action
- next best offer
- product recommendation
- price decision
- inventory allocation
- replenishment
- fulfillment route
- return decision
- fraud intervention
- customer-service action
- supplier response
AnantVerify independently evaluates
- state sufficiency
- evidence
- business authority
- decision behavior
- policy
- assurance coverage
Then it determines: ALLOW | HOLD | ESCALATE | BLOCK
One decision layer across retail
The same architecture, twelve domains
Twelve retail decision domains
Customer
What should we do for this customer now?
Commerce
How should this transaction proceed?
Merchandising
Which products, assortments and promotions need attention?
Pricing
What pricing action is appropriate?
Inventory
Where should inventory go?
Fulfillment
How should this order be fulfilled?
Stores
What needs attention at this location?
Supply chain
How should we respond to disruption?
Fraud and loss
Which transaction or relationship requires intervention?
Customer service
What is the next best resolution?
Workforce
Where should people be deployed?
Finance
Which operational exceptions require action?
Every domain reads the same state.
These are not twelve products. They are twelve decision surfaces that share one connected state and one assurance path.
One connected retail state. Thousands of decisions.
Real-time retail
Batch analytics explain yesterday. Retail decisions happen now.
One event, several retail decisions
One event
Supplier shipment delayed seven days
Inventory
Reallocate across distribution centers and stores
Fulfillment
Reroute orders that can still meet the promise
Promotion
Modify the regional campaign that is about to launch
Pricing
Hold a planned price change on the item
Customer promise
Contact the customers whose delivery date moves
The delayed shipment is a single fact. What the network does about it is five decisions, and they are made in five different functions.
The loop
- Event
- State change
- Decision opportunity
- AnantDecision
- AnantVerify
- Action
Each of these is an event the operation already sees:
- a customer abandons a cart → next best action
- inventory drops below expected demand → replenish
- a supplier shipment is delayed → reallocate
- a fraud relationship is discovered → hold
- store refrigeration changes → intervene
The gap is not detection. The operation already detects all of it. The gap is that detection and the decision it implies live in different systems, different teams and different timescales.
Retail decision replay
Reconstruct the decision and the authority behind it
When an automated retail decision is questioned — by a customer, an auditor or a regulator — the explanation has to include more than the model output.
Reconstructed from
- customer state
- order state
- inventory
- price
- promotion
- fraud evidence
- AI proposal
- applicable policy
- assurance verdict
- human intervention
- execution
- outcome
Do not just know what the AI did. Know why the business allowed it.
Learn from every outcome
Retail gives feedback faster than most industries
The loop
- State
- Decision
- Action
- Outcome
- Learning
- Better next decision
Every decision produces an outcome the next decision can use.
- A customer accepted, ignored, purchased, returned, churned or stayed
- A product sold, stocked out, was transferred or was marked down
- A fraud case was cleared or confirmed
Retail therefore provides rapid feedback to the decision layer, provided the outcome is connected back to the decision that produced it.
Customer decision intelligence
Retailers have spent years building a complete view of the customer. Knowing the customer is where the work starts, not where it ends.
Beyond Customer 360
The complete view is the input, not the answer
Customer 360 solved a real problem: the profile, the orders, the browsing and the service history in one place. It did not answer the question the business actually has in front of a customer.
What should we do about it?
AnantDecision turns Customer 360 into customer decision intelligence: the same state, connected to the actions available right now.
Customer 360 to next best action
The profile becomes an action under business constraints
From customer inputs to a next best action
What the retailer already knows
- profile
- purchase history
- browsing
- cart
- returns
- loyalty
- service history
- offers
- preferences
- channel behavior
- product affinity
- inventory availability
- current context
ANANTDECISION
What may happen next
- OFFER
- RECOMMEND
- RETAIN
- ASSIST
- WAIT
- ESCALATE
- TAKE NO ACTION
The inputs describe the customer. The outcomes are decisions about them, and they are not all offers.
The inputs are the profile. The output is a decision about this customer, now, given what the business can actually deliver and what it is willing to give away.
A customer profile tells you who they are. Decision intelligence tells you what to do next.
Example: next best action
The same retailer, two different conclusions
A loyal customer, a generic promotion
Customer
High-value loyalty member
- Digital behaviorBrowsed premium running shoes three times
- Purchase historyRegular athletic apparel buyer
- CartNo current item
- InventoryPreferred size available at a nearby store
- PromotionGeneric 20% promotion available
- MarginPremium product already selling strongly
- Recent experienceDelivery issue resolved yesterday
- Personalization engineWould send 20% OFF
AnantDecision
RECOMMEND LOCAL AVAILABILITY — NO DISCOUNT
The goal is not maximizing response probability. It is making the best business decision, and the discount was neither necessary nor free.
Neither answer is wrong in the abstract. One of them is wrong for this customer, this week, at this margin.
Hyper-personalization with business context
The right offer can still be the wrong decision
Personalization should not happen independently of
- inventory
- margin
- promotion
- customer value
- returns
- channel
- fulfillment
- business objectives
A message that maximizes response and a decision that improves the business are not always the same thing, and the difference is visible in these conditions.
Personalize the decision, not only the message.
Customer journey intelligence
A journey is a trajectory, not a stage
One journey, nine stages
Anant observes where the customer is in that journey and identifies meaningful changes. The question becomes: what intervention is appropriate at this moment?
Most retail systems know which stage a customer is in. Fewer know how the customer arrived there, which is what makes the next intervention appropriate or intrusive.
Churn and retention decisions
A probability is not yet a decision about a customer
A churn score is useful and incomplete. It says who is likely to leave. It does not say whether intervening is worth it, or what the intervention should be.
AnantDecision determines
- whether to intervene
- when
- through which channel
- with what action
- under what economics
Possible actions
- OFFER
- SERVICE RECOVERY
- LOYALTY BENEFIT
- CONTACT
- NO ACTION
Move from churn prediction to retention decisions.
Omnichannel state
Customers do not think in channels
They
- browse online
- visit stores
- purchase through mobile
- pick up curbside
- contact support
- return in store
- reorder online
Each of those events lands in a different system. A decision that is made against only one of them is made against a fraction of the customer.
One customer. One state. Every channel.
Retail trajectories
Two customers can arrive at the same cart value
Same cart value, two journeys
Customer A
- Browse
- Cart
- Purchase
A direct path that ended in a purchase.
Customer B
- Browse
- Cart
- Abandon
- Return
- Cart
The same cart, arrived at twice, after leaving.
Same cart, same moment, different history.
Same cart, same moment, different history. The next best action is unlikely to be the same for both, and a system that only reads current state cannot tell them apart.
Current state tells you where the customer is. Trajectory tells you how they got there.
Pricing, promotion and margin
A price or a promotion is a decision about margin, inventory, demand and customer value at the same time. It is rarely decided that way.
Pricing intelligence
A price decision depends on more than demand
Demand rising does not automatically mean the price should rise. It depends on what the demand is made of, what is on the shelf and what the item is worth to the business.
A pricing decision may depend on
- demand
- inventory
- competitors
- margin
- promotion
- customer segment
- seasonality
- product lifecycle
- store
- channel
- supplier cost
- business rules
Example: markdown decision
Marking down everywhere is one option, not the only one
One product, one state, three actions
Product
Seasonal apparel
- InventoryHigh
- Sell-throughBelow target
- SeasonThree weeks remaining
- MarginHealthy
- Online demandIncreasing
- Store demandWeak in selected locations
AnantDecision
MARKDOWN SELECTED STORES · TRANSFER INVENTORY · MAINTAIN ONLINE PRICE
One product. Different state. Different decision. A blanket markdown would have given away margin in the stores that could sell it and online, where demand was already rising.
The same product, read against local demand and online demand, supports three different decisions at once.
Promotion intelligence
A promotion is an economic decision, not a targeting exercise
Evaluate against
- incremental demand
- margin
- inventory
- customer value
- product affinity
- substitution
- cannibalization
- fulfillment capacity
The usual question is:
“Who is likely to click?”
Which action creates the best expected business outcome under the current state?
Dynamic pricing with control
Pricing can be automated without making authority implicit
Pricing is the surface where an autonomous system can do the most damage fastest, because the action is a number and the number can be wrong in either direction.
Pricing may involve
- business rules
- margin requirements
- promotions
- contractual restrictions
- channel strategy
- inventory objectives
- applicable legal or organizational requirements
AnantDecision can recommend. AnantVerify independently evaluates whether the decision satisfies configured authority and assurance requirements before automated execution.
Inventory and availability
Inventory intelligence is not knowing how much is on hand. It is knowing what to do with what is on hand, across a network that is already moving.
Inventory intelligence
Know what you have, and what to do with it
Inventory exists across
- distribution centers
- stores
- dark stores
- in transit
- suppliers
- returns
- reserved stock
- marketplaces
AnantDecision can determine
- REPLENISH
- TRANSFER
- ALLOCATE
- RESERVE
- EXPEDITE
- MARKDOWN
- SUBSTITUTE
- DEFER
Example: inventory allocation
The decision considers the network, not one location
Five hundred units, one network
Product
High-demand electronics item
- Available inventory500 units
- Store AHigh demand, low inventory
- Store BModerate inventory, weak demand
- E-commerceDemand accelerating
- SupplierNext shipment delayed
- PromotionCampaign scheduled tomorrow
AnantDecision
REALLOCATE + RESERVE E-COMMERCE CAPACITY
The reallocation is made against the campaign that starts tomorrow and the shipment that is not arriving, not against last month's sell-through.
A single-store view would send stock where the last sale happened. The network view sends it where the next demand is, and reserves the part that cannot be replaced.
From forecasting to inventory decisions
A forecast is an input to a decision
“How much will we sell?”
That is the forecasting question, and it is well served by existing systems. The decision question is different:
What should we do because of that forecast?
Possible actions
- ORDER
- ALLOCATE
- TRANSFER
- MARKDOWN
- EXPEDITE
- SUBSTITUTE
- CHANGE PROMOTION
- CHANGE FULFILLMENT
Shelf availability
An empty shelf has several possible causes
A shelf may appear empty because
- store inventory is wrong
- the product is in the back room
- the shipment is delayed
- a replenishment task was missed
- demand spiked
- inventory was lost
Every one of those has a different remedy, and the remedy is the decision. A restock task is the wrong answer to three of them.
Do not just detect the empty shelf. Decide why it is empty and what should happen next.
Stores, IoT and operations
A store generates operational state continuously: traffic, stock, queues, equipment, staffing and loss. Connected stores make that state available to decisions.
Store intelligence
A store is a live operational system
Signals
- traffic
- inventory
- POS
- returns
- staffing
- queues
- shelf availability
- fulfillment
- equipment
- loss events
- customer interactions
- IoT
Most of this is already collected. Turning it into operational next best actions is the part that is usually missing.
IoT and connected stores
Collecting sensor data is the beginning
Retail IoT can include
- RFID
- smart shelves
- beacons
- cameras
- refrigeration sensors
- temperature sensors
- energy systems
- footfall sensors
- POS devices
- electronic shelf labels
- warehouse equipment
- fleet and delivery devices
The value comes from connecting the signal to the retail decision.
Example: refrigeration
A temperature alert is not yet an action
One sensor reading, five pieces of context
Sensor
Temperature rising
- EquipmentRefrigeration unit 14
- InventoryHigh-value perishable stock
- MaintenanceSimilar issue occurred last month
- TechnicianAvailable nearby
- StoreHigh sales volume
- Monitoring systemWould raise a temperature alert
AnantDecision
DISPATCH TECHNICIAN + MOVE AT-RISK INVENTORY
IoT signal to business context to next best action. An alert would have told the same person the same fact at the same time; it would not have started the work.
The sensor state is one input. Who is nearby, what is in the unit, what happened last month and how much the store sells all change what should happen now.
Store workforce decisions
Workforce planning against live operational state
Decisions can reflect
- traffic
- orders
- fulfillment demand
- service queues
- skills
- availability
- store events
- expected demand
Potential actions
- REALLOCATE
- SCHEDULE
- CALL IN
- ASSIGN
- PRIORITIZE
Retail operations
The same architecture reaches the back office
Decision intelligence can extend to
- store opening and closing
- cash management
- facilities
- equipment
- maintenance
- loss prevention
- energy
- labor
- delivery
- compliance
- operational exceptions
One decision architecture from customer experience to back-office operations.
Orders, returns and trust
An order is a decision journey, and every step of it can invalidate the step before. Returns and fraud are where that is most visible and most expensive.
Order intelligence
State can change at every step of an order
One order
- Order
- Payment
- Fraud
- Inventory
- Fulfillment
- Shipment
- Delivery
- Return / service
AnantDecision can determine whether the order should
- PROCEED
- HOLD
- REROUTE
- SUBSTITUTE
- EXPEDITE
- CANCEL
- ESCALATE
Intelligent order routing
The cheapest fulfillment is not always the best decision
A fulfillment decision can consider
- inventory
- distance
- delivery promise
- store capacity
- warehouse workload
- shipping cost
- margin
- split shipment
- customer value
- return probability
- inventory strategy
A routing engine that optimizes shipping cost alone will make the promise it cannot keep, and pay for it in the service contact that follows.
Example: fulfillment
Balancing the promise against the operation that has to keep it
A premium customer, a busy store
Customer
Premium loyalty member
- OrderTwo items
- Local storeBoth available
- Distribution centerOne item available
- Store workloadHigh
- Delivery promiseTomorrow
- Alternative storeBoth available, with capacity
AnantDecision
FULFILL FROM ALTERNATIVE STORE
The decision balances customer promise and operational state rather than optimizing picking cost inside a store that was already behind.
The customer is entitled to the promise. The store is already at capacity. Both facts belong in the decision.
Returns and refunds
Returns are a major retail decision surface
Possible decisions
- APPROVE
- REFUND
- EXCHANGE
- STORE CREDIT
- REQUEST RETURN
- INSPECT
- ESCALATE
- DECLINE according to applicable policy
The appropriate decision may depend on
- order
- product
- return window
- customer history
- condition
- fraud signals
- policy
- value
- channel
Example: AI agent refund
Agent confidence is not refund authority
A confident agent and an independent verdict
Proposed by the agent
REFUND $475
- CustomerLong-term high-value customer
- OrderDelivered
- Return requestWithin the policy window
- ProductHigh-value electronics
- EvidenceDelivery confirmed
- New signalMultiple high-value refund requests across linked accounts
- Agent confidence96% refund
AnantVerify evaluates
- state sufficiency
- evidence consistency
- business authority for this amount
- linked-account history
Agent confidence is not refund authority. The proposal was well supported by the case in front of the agent and unsupported by the state around it.
Every fact in this case supports the refund. One fact outside the case changes what the business should do with it.
Fraud and loss prevention
Retail fraud is a relationship problem
Fraud crosses
- payments
- accounts
- devices
- orders
- returns
- promotions
- gift cards
- loyalty
- stores
- identities
Anant reasons over the relationships between those systems rather than evaluating every transaction on its own merits.
Relationship-aware retail fraud
Risk can exist in the relationship
A return can look normal. A customer can look normal. A device can look normal. Multiple customers using one device, returning similar products to multiple stores, is a different state entirely.
AnantDecision can determine
- CLEAR
- MONITOR
- HOLD
- INVESTIGATE
- ESCALATE
Each transaction passed on its own. The pattern is what the decision is about.
Agents, digital commerce and change
Agents can already browse, recommend, refund and re-price. The open question is what they are allowed to do without a second, independent opinion.
E-commerce and digital commerce
The website responds to more than clicks
Digital commerce generates decision opportunities across
- search
- recommendations
- content
- pricing
- promotions
- cart
- checkout
- fraud
- fulfillment
- service
Customer + Product + Inventory + Business State.
AI agents for retail
Retail is becoming an agentic environment
Agents may support
- shopping
- customer service
- merchandising
- pricing
- inventory
- procurement
- supply chain
- store operations
- employee assistance
An agent can reason and orchestrate. AnantDecision supplies the enterprise decision intelligence, and AnantVerify supplies independent assurance before a consequential action executes.
Agentic commerce
The question is no longer API access
What the agent can already do
- Understand a complaint
- Retrieve the order
- Examine delivery state
- Determine available remedies
- Recommend a refund
- Issue store credit
- Arrange replacement
“Can the agent use the refund API?”
Should this particular refund execute under this customer, order, evidence and policy state?
That is decision assurance, and it is the question the agent itself cannot answer.
Autonomous merchandising
More autonomy without implicit authority
AI agents may increasingly
- adjust assortment
- recommend markdowns
- create promotions
- change product placement
- identify inventory actions
- coordinate suppliers
Let AI propose. Keep consequential authority explicit.
Test retail AI under change
Yesterday's examples are not today's conditions
A model can perform well on the examples it was built from and still respond to the wrong thing when a promotion starts, a price moves or a policy window closes.
“When retail conditions change, does the decision change in the right way?”
Example: refund testing
A refund decision, tested under change
Five changes to one decision
- BaselineCustomer eligible, order valid, inside the return windowApprove
- Irrelevant changeThe order of records in the system changesStill approve
- Material changeThe return window has expiredThe decision changes
- Combined conditionsHigh-value item plus a linked-account fraud signalNeither condition is decisive alone
- Sequence changeRefund then delivery reversal, versus delivery reversal then refundThe behavior has to be consistent either way
Sensitive to what matters, stable to what does not.
The test is not whether the system can approve a refund. It is whether it responds to the conditions that should change the answer, and ignores the ones that should not.
Supply chain and merchandising
A supplier delay is rarely only a procurement problem, and a markdown is rarely only a pricing problem. Both are network decisions.
Supply chain
One disruption, several connected decisions
Retail supply chains connect
- supplier
- product
- purchase order
- shipment
- distribution center
- store
- inventory
- customer order
- demand
Potential decisions
- ORDER
- EXPEDITE
- REALLOCATE
- SUBSTITUTE
- REROUTE
- CHANGE PROMOTION
- CHANGE PRICE
- ESCALATE
Example: supplier disruption
The delay is not only a procurement problem
One delay, four connected responses
Product
Top-selling seasonal product
- SupplierShipment delayed seven days
- Distribution centersUneven inventory
- StoresDemand differs by region
- DigitalOnline demand accelerating
- PromotionNational campaign begins in four days
AnantDecision
REALLOCATE + PRIORITIZE DIGITAL + MODIFY REGIONAL PROMOTION
One disruption. Multiple connected decisions. Treating it as a procurement problem alone would have left the campaign to find the shortage on its own.
The shipment is late and cannot be made early. What can still be decided is what the network does with the four days it has left before the campaign starts.
Merchandising decisions
Assortment is a decision surface, not a plan
Merchandising decisions can incorporate
- customer demand
- product affinity
- inventory
- margin
- seasonality
- store cluster
- digital behavior
- supplier
- promotion
- competitive state
Potential decisions
- ASSORT
- PROMOTE
- MARKDOWN
- TRANSFER
- FEATURE
- REPLENISH
- RETIRE
Product trajectories
Pricing depends on where the product is in its life
One product, six stages
Pricing, inventory and promotion decisions should reflect where the product is in that trajectory.
The same product supports different decisions at launch and at exit, and the state that distinguishes them is time.
The business case and what to measure
The value is built from the retailer's own economics, measured at the level of the decision, and disclosed as a model rather than a result.
The business case
Six economic levers
Illustrative business cases
Five worked assumptions
Five illustrative business cases
No figure on this panel is a measured result or a promised outcome.
Customer retention
Assumptions
- 5,000,000 active customers
- 5% enter retention programs annually = 250,000
- $150 average annual contribution per customer
- 1% incremental retention of the targeted population = 2,500 customers
2,500 customers × $150
$375,000 of modeled annual contribution retained
Also excludes
- the cost of the interventions themselves
- whether the retained customers would have left
- this is a scenario model, not a promised result
Inventory
Assumptions
- $500M average inventory
- allocation and decisioning improve enough to operate with inventory lower by 0.5%
- while maintaining required service levels
0.5% of $500M
$2.5M of modeled working capital difference
Also excludes
- carrying cost and the service impact of holding less
- this is an illustrative economic model
Returns
Assumptions
- 2,000,000 annual returns
- 20% currently reviewed manually = 400,000
- $8 average manual review cost
- current modeled review cost $3.2M
controlled decision automation moving manual review from 20% to 15%
100,000 reviews leave the manual queue, modeled at $800,000 a year
Also excludes
- fraud effects
- customer-experience effects
- this is an illustrative economic model
Store operations
Assumptions
- 1,000 stores
- 30 minutes per day per location reviewing operational exceptions that could be triaged automatically
1,000 × 0.5 × 365
182,500 hours of management time a year
Also excludes
- the retailer should value that capacity using its own labor economics
- even a partial reduction is the point of the model
Stockouts
Assumptions
- $2B annual revenue
- products representing 1% of potential demand experience avoidable availability loss
1% of $2B
$20M of potential revenue exposed
Also excludes
- only a fraction is recoverable, and the fraction is the question
- the case should use SKU-level demand, substitution and margin data
Each case is arithmetic on the inputs above, and the two that rest on a percentage change are presented as models rather than as results.
Five cases, each built from stated inputs. Three are arithmetic on volume and unit economics; two involve a percentage change and are labeled as models rather than expectations.
This is an illustrative economic model, not a promised outcome.
Retail ROI model
Build the case from the retailer's own economics
The business case should be measurable at the decision task level, so that a single decision journey can be proven or abandoned on its own evidence.
Executive retail KPIs
The measures an executive already owns
Track
- revenue
- gross margin
- conversion
- customer retention
- inventory turns
- stockout rate
- sell-through
- markdown
- return rate
- fulfillment cost
- on-time delivery
- decision automation
- value generated from decision programs
Retail domain KPIs
The same program read from five functions
Customer
- next-best-action acceptance
- conversion
- repeat purchase
- retention
- customer lifetime value
- offer acceptance
- service resolution time
- engagement
- intervention-to-purchase rate
Inventory and merchandising
- inventory turns
- stockout
- excess inventory
- sell-through
- markdown rate
- allocation effectiveness
- transfer rate
- forecast-to-action time
- availability
- gross margin return on inventory
Fulfillment
- order cycle time
- on-time delivery
- split shipments
- cost per order
- store fulfillment rate
- cancellation
- substitution
- promise accuracy
- routing exceptions
Returns and fraud
- return rate
- manual review rate
- refund cycle time
- fraud loss
- false-positive rate
- investigation yield
- cost per review
- customer friction
- decision override rate
Store
- sales per store
- labor productivity
- shelf availability
- queue time
- fulfillment productivity
- operational exceptions
- energy
- equipment availability
- decision-to-action time
Decision intelligence KPIs
Whether the decisions themselves are healthy
Track
- decision volume
- decision cycle time
- automation rate
- next-best-action acceptance
- confidence calibration
- abstention
- decision stability
- boundary behavior
- decision drift
- human override
- outcome after decision
Assurance KPIs
Independence, measured rather than asserted
Track
- state sufficiency
- evidence conflicts
- authority conflicts
- ALLOW / HOLD / ESCALATE / BLOCK
- assurance coverage
- interaction failures
- human overrides
- execution mismatches
- decision replay completeness
Measure customer and business value together
The highest-response decision is not always the best one
Three trades that a single-objective system gets wrong:
- A deep discount raises conversion and reduces margin
- An aggressive fraud rule reduces loss and rejects valuable customers
- Fast fulfillment improves experience and destroys order economics
Optimize the business decision, not a single model score.
One decision, then the next
No retailer has to transform everything at once. The path is one decision with a measurable economic outcome, proven, then expanded.
Start with high-value decision journeys
Volume, change and measurable economics
Start where
- decision volume is high
- conditions change quickly
- manual effort is significant
- economic outcomes are measurable
- the action can be clearly defined
Strong starting points
- next best action
- returns and refunds
- inventory allocation
- replenishment
- fulfillment routing
- fraud disposition
- markdown
- supplier disruption
- store operational exceptions
Prove one decision, then expand
Nine stages from one decision to an architecture
One journey, nine stages
One decision becomes an enterprise intelligence architecture.
The order matters. A baseline taken after the workflow changed is not a baseline, and a second decision journey is only funded if the first one survived review.
From omnichannel to adaptive retail
The next transition after omnichannel
Retail, in four steps
- Store
- E-commerce
- Omnichannel
- AI-powered commerce
- Adaptive retail
Each previous transition connected more of the business. This one changes what the business does with the connection.
Continuously
- observes
- understands
- predicts
- decides
- assures
- acts
- learns
The retail enterprise that knows what to do next
Know the state, then decide what it implies
Know
- the customer
- the product
- the inventory
- the order
- the store
- the supplier
- the operation
- the moment
Then determine: what should happen next?
And when the action is consequential:
Should we allow it?
From customer signal to trusted action
Know the customer, the product, the inventory, the order, the store, the supplier, the operation and the moment. Then determine what should happen next — and whether the business should allow it.
ANANTDECISIONReal-Time Retail Decision Intelligence
ANANTVERIFYIndependent Decision Assurance
- Observe
- Understand
- Decide
- Verify
- Act
- Learn
From customer signal to business state. From business state to next best action.
Worked at decision level: Customer Refund Authorization and From Policy Document to Runtime Control.
AnantVerify does not by itself establish regulatory compliance. It provides decision evidence, testing, control and replay that can support a retailer's governance, financial-control and customer-treatment programs.
Illustrative throughout. Every volume, rate, cost, hour and percentage on this page is a worked assumption for building a value model — not a measured result and not a promised outcome. Replace them with the retailer's own economics.