Real-time decision intelligence for every digital commerce moment
A customer searches, a product trends, a cart is abandoned, a payment is attempted, a promotion expires, a shipment slips and a return is requested. Every one of those events changes the business.
At a glance
Decisions cluster around the journey
Discovery, cart, checkout and post-purchase each commit the business to something.
Promises are made in milliseconds
A delivery date or an honored price is a commitment the enterprise can be held to.
Growth and fraud pull in opposite directions
Two individually correct decisions can conflict when one wants to allow and the other to hold.
Binding the promise to the state
Execution verifies the state that was assured before the commitment is kept.
The customer journey
- Search
- Product
- Cart (current step)
- Checkout
- Order
- Delivery
- Service
- Return
The marked step is where the session is now. What happened before it is part of the state.
Enterprise state, changing with every step
- inventory
- price
- margin
- fraud
- fulfillment
- promotion
- customer value
ANANTDECISION
A changing next best action, not a fixed rule response.
- RECOMMEND
- OFFER
- ASSIST
- CHANGE FULFILLMENT
- HOLD
- NO ACTION
ANANTVERIFY
Appears only at consequential execution boundaries.
ALLOW · HOLD · ESCALATE · BLOCK
Executed by
- commerce platform
- payment gateway
- order management
- fulfillment
- service desk
- AI agents
Most steps do not need assurance. The consequential ones — releasing an order, changing a price, refunding money — do not execute without it.
“Every click changes the state.”
Every state change creates a new decision opportunity.
Click → State → Decide → Verify → Act → Learn
E-commerce happens in real time
Nothing in digital commerce sits still long enough to be analyzed later. The events arrive continuously and each one changes what the right answer is.
Every event changes the business
The events are continuous, and so is the state they produce
Any of these can arrive in the next second:
- a customer searches
- a product trends
- inventory changes
- a cart is abandoned
- a payment is attempted
- a promotion expires
- a shipment is delayed
- a return is requested
- a fraud signal appears
- an AI agent proposes an action
The question is: what should happen next?
Anant turns continuously changing customer, product, inventory, order, pricing, fulfillment and risk signals into real-time decisions, next best actions and controlled execution.
The journey is a chain of decisions
A storefront looks simple. Behind every session are hundreds of decisions.
Thirteen decisions in one session
Discovery
What should the customer see?
Search
Which products should rank?
Recommendation
What should we recommend next?
Offer
Should we provide an incentive?
Pricing
What price or promotion applies?
Cart
Should we intervene?
Checkout
How should the transaction proceed?
Fraud
Should the order be released?
Fulfillment
Where should the order ship from?
Delivery
How should an exception be handled?
Service
What resolution should we offer?
Return
Should the return be approved automatically?
Retention
What should happen next?
Every step is a decision about what to do, not only about what to show.
Thirteen of them are unavoidable, and every one is a question about what to do rather than what to display.
The digital customer journey is a chain of decisions.
From digital commerce to adaptive commerce
The difference is what happens after the customer acts
The loop
- Customer acts
- Enterprise state changes
- Context understood
- AnantDecision
- AnantVerify
- Experience adapts
- Outcome becomes state
A conventional storefront responds to the action. An adaptive one responds to the state the action produced, which is a different and more useful thing.
The commerce experience continuously adapts to the customer and the business.
One real-time commerce state
The signals are already there. They are in fifteen systems.
Commerce data is usually fragmented across
- commerce platform
- CDP
- CRM
- search
- recommendation engine
- OMS
- WMS
- payments
- fraud
- customer service
- marketing
- loyalty
- inventory
- shipping
- marketplaces
Every one of those systems holds part of the answer, and a decision made inside any one of them is made against that part alone.
The commerce state
Eight dimensions, one connected decision context
Eight dimensions of commerce state
Customer
- identity
- history
- value
- loyalty
- preferences
- returns
- service
Session
- search
- browse
- clicks
- sequence
- cart
- intent
Product
- category
- affinity
- attributes
- lifecycle
- margin
Inventory
- availability
- location
- reserved
- in transit
Commercial
- price
- promotion
- margin
- offer
Order
- cart
- payment
- fraud
- fulfillment
- delivery
- return
Supply
- supplier
- lead time
- availability
- disruption
Operations
- warehouse
- store
- carrier
- capacity
- service
Eight dimensions, changing at different speeds, read as one decision context.
These are not eight views of a customer. They are the state a commerce decision is actually made against, and they change at different speeds.
Together they are one connected decision context.
The decision layer and the assurance layer
One commerce state, two capabilities: a layer that determines the next best action, and an independent layer that decides whether a consequential one may execute.
AnantDecision + AnantVerify
One layer determines the action. Another decides whether it executes.
Together
- STATE
- DECIDE
- VERIFY
- ACT
- LEARN
AnantDecision determines
- NEXT BEST PRODUCT
- NEXT BEST OFFER
- NEXT BEST ACTION
- PRICE
- PROMOTION
- FULFILLMENT ROUTE
- FRAUD DISPOSITION
- REFUND
- RETENTION ACTION
- SERVICE RESOLUTION
- OPERATIONAL RESPONSE
AnantVerify independently evaluates
- state sufficiency
- evidence
- business policy
- authority
- decision behavior
- assurance coverage
Then it determines: ALLOW | HOLD | ESCALATE | BLOCK
Beyond personalization
Personalization chooses content. Decision intelligence chooses action.
“What product should we recommend?”
That is one decision out of hundreds, and treating it as the whole problem is how a commerce AI program ends up with better banners and the same margin.
What should we do next for this customer and this business?
The answer may be one of these, or none of them:
- recommend a product
- offer assistance
- show local availability
- change fulfillment
- provide an incentive
- resolve a service issue
- do nothing
Real-time next best action
Every digital event can update the next best action
One session
- Search
- Product view
- Second product view
- Price check
- Cart
- Shipping check
- Abandonment
The events below arrive in seconds, and each one changes the state the previous action was decided against.
What is the best action now?
Event, state, decision
The decision comes from the state, not the event alone
The model
- Event
- Update connected state
- Understand context
- AnantDecision
- Next best action
The alternative to a state model is a rule for every event, and rules cannot see each other.
| The rule | What it cannot see |
|---|---|
| IF cart abandoned THEN coupon | Why the cart was abandoned |
| IF inventory low THEN alert | What the demand is doing |
| IF shipment delayed THEN email | Which customers the delay actually matters to |
Anant takes the broader route: an event updates the connected state, the state supplies the context, and the decision layer determines what to do.
Commerce decision replay
Reconstruct the commerce decision, not the API call
The record
- Customer state
- Order
- Payment
- Inventory
- Evidence
- Policy and authority
- AI proposal
- Assurance
- Execution
- Outcome
Two people ask this question with the same need for an answer: a customer asking why an order was held, and an executive asking why an agent issued a refund.
Reconstructed from
- customer state
- order
- payment
- inventory
- evidence
- policy and authority
- AI proposal
- assurance
- execution
- outcome
Learn from every commerce outcome
Digital commerce produces feedback faster than any other channel
The loop
- State
- Decision
- Action
- Outcome
- Learn
- Better next decision
Every decision produces an outcome the next decision can use.
- The customer clicked, ignored, purchased, abandoned, returned, complained, renewed or churned
- The order was delivered, failed, was returned or became fraud
Those outcomes become new evidence, provided they are connected back to the decision that produced them rather than to the campaign that happened to be running.
The customer journey as a decision surface
Discovery, intent, search, recommendation, cart and abandonment are not four systems with four models. They are one journey with one state.
Customer journey intelligence
A journey is a trajectory, not a collection of clicks
One product in cart, two paths
Customer A
- Search
- Product
- Cart
- Purchase
A direct path that is already converting.
Customer B
- Search
- Product
- Cart
- Abandon
- Return
- Product
- Cart
A path that has already survived one abandonment.
One cart, read without the path that produced it.
Both customers below have one product in the cart right now. A system that reads only current state sees one customer.
Same state. Different journey. Different decision.
Intent intelligence
What is this customer trying to accomplish now?
The usual question is what this customer resembles. It is answerable and it is not the question that decides anything.
What intent can be inferred from
- search behavior
- navigation
- product interactions
- cart activity
- purchase history
- channel
- time
- customer relationship
- business context
Search to decision intelligence
A relevance score is one input to a commercial decision
“Which products match this query?”
A search decision can also consider
- customer
- intent
- inventory
- margin
- availability
- delivery promise
- product relationship
- business objectives
- current journey
The result can influence ranking, recommendation, merchandising and the next best action without reducing the decision to one relevance score.
Recommendations with enterprise context
Recommend what makes sense for the customer and the business
Recommendation engines optimize relevance or conversion. A commercially intelligent recommendation can account for more than the click it is trying to earn.
Also considered
- availability
- margin
- return probability
- delivery
- customer value
- promotion
- substitution
- inventory strategy
- product lifecycle
Cart intelligence
A cart is a high-value decision state
AnantDecision can consider
- customer
- products
- cart value
- inventory
- shipping
- margin
- promotion
- historical behavior
- abandonment behavior
- fraud
- service state
Potential actions
- NO ACTION
- RECOMMEND
- BUNDLE
- ASSIST
- REMIND
- OFFER
- CHANGE FULFILLMENT
- ESCALATE
Abandonment intelligence
The cause of the abandonment is the decision
The conventional response is a coupon. It is the right answer to one of the eight causes below and the wrong answer to the other seven.
A cart may be abandoned because of
- shipping cost
- delivery time
- inventory
- price sensitivity
- payment failure
- comparison behavior
- service issue
- low purchase intent
Do not treat every abandoned cart as a discount opportunity.
Customer service intelligence
A service agent needs the next best resolution
Customer history is the starting point. What the agent — human or AI — needs next is what to do for this customer, on this order, under this policy.
AnantDecision can combine
- customer
- order
- delivery
- product
- payment
- return
- service history
- loyalty
- policy
- available remedies
Possible decisions
- REPLACE
- REFUND
- CREDIT
- EXPEDITE
- ESCALATE
- NO ACTION
Price, promotion and retention
Discounting is the most expensive default in digital commerce. Every section here is about spending it deliberately.
Pricing intelligence
Digital pricing decisions happen continuously
Continuous pricing is a capability. Whether each change is a good decision is a separate question, and it depends on more than demand.
Relevant context
- demand
- inventory
- margin
- competition
- customer segment
- promotion
- product lifecycle
- channel
- supplier cost
- business rules
Promotion intelligence
The promotion most likely to produce a click is not necessarily the best one
A promotion decision can consider
- incrementality
- customer value
- margin
- inventory
- product affinity
- promotion history
- cannibalization
- fulfillment capacity
- retention objective
Optimize the commercial decision, not simply campaign response.
Example: the customer who does not need a discount
Better personalization is not always more promotion
An offer that should not be sent
Customer
High-value repeat customer
- SessionThird visit to the same premium product
- ProductStrong demand
- InventoryLimited
- Customer historyFrequently purchases without promotions
- Current campaign15% discount available
- Campaign engineWould send 15% OFF
AnantDecision
SHOW AVAILABILITY + DELIVERY PROMISE
Sometimes the best offer is no offer. The discount would have reduced margin on a purchase that was already going to happen.
The campaign is available, the customer is eligible and the offer is on brand. None of that makes it the right decision.
Subscription commerce
Subscription businesses decide continuously
The decisions
- renew
- retain
- upgrade
- downgrade
- pause
- recover payment
- offer
- contact
- cancel
AnantDecision can incorporate
- usage
- payment
- engagement
- service
- plan
- customer value
- support history
- trajectory
Example: subscription retention
Prediction tells you who may leave. Decision intelligence tells you what to do.
A churn score and a retention decision
Model output
Churn probability 76%
- Historic valueHigh
- UsageDeclining
- Recent eventSupport issue
- Price sensitivityNo history of it
- Product affinityPremium features
- Retention playbookWould send 20% discount
AnantDecision
SERVICE RECOVERY + PRODUCT EDUCATION
The customer was not leaving over price. A discount would have paid for a problem the customer did not have.
The churn score identifies the customer. It does not choose between a discount and a service recovery, and on this account the two are not equivalent.
Checkout, payment and fraud
Checkout is where customer experience, economics and risk meet in the same second, and where the cost of both over-control and under-control is immediate.
Checkout intelligence
The transaction decision is made against everything at once
Relevant state
- customer
- cart
- payment
- device
- address
- inventory
- promotion
- fraud
- shipping
- loyalty
- order history
Potential decisions
- PROCEED
- REQUEST VERIFICATION
- CHANGE PAYMENT METHOD
- HOLD
- ESCALATE
- BLOCK according to configured policy
Payment and fraud
Too little control is loss. Too much is lost customers.
AnantDecision can reason across
- customer identity
- account
- device
- payment instrument
- address
- order
- product
- return history
- relationships
- behavior
The decision is not the score. It is what to do with this order.
Relationship-aware fraud
Risk can exist in the network, not the individual transaction
Read as one state
- Multiple accounts
- One device
- Related addresses
- Similar high-value products
- Repeated refund requests
One order can look entirely normal. The pattern it belongs to is what the decision has to be made against.
The pattern
- multiple accounts
- one device
- related addresses
- similar high-value products
- repeated refund requests
Example: high-value order
Every fact about the order is clean
A clean order with an unclean relationship
Order
$3,200 electronics purchase
- CustomerEstablished account
- PaymentAuthorized
- DevicePreviously observed
- ShippingNew address
- Relationship stateAddress linked to several recently created accounts
- Traditional decisionPayment approved
AnantDecision
HOLD FOR ADDITIONAL REVIEW
The decision reflects the connected state rather than one fraud score, and it is a hold rather than a decline.
The payment is authorized, the device has been seen before and the account is established. The state around the order is what changes the decision.
Assure agentic refunds
An agent can recommend a refund without becoming the refund authority
A confident agent and an independent evaluation
Proposed by the service agent
REFUND $1,250
- Agent confidence97%
- ContextConversational history available
- What the agent cannot seeThe customer, order, evidence and authority state together
AnantVerify evaluates
- customer state
- order state
- return policy
- evidence
- fraud relationships
- refund authority
- assurance coverage
The evaluation decides between ALLOW, HOLD, ESCALATE and BLOCK. Agent confidence is an input to the decision, not the decision.
The agent has conversational context and a high confidence. Neither is the same as authority over $1,250 of the merchant's money.
Should this particular refund execute under this customer, order, evidence and policy state?
Order, fulfillment and returns
After checkout the decision journey continues, and the state keeps moving underneath it: inventory, capacity, carriers, addresses and delivery promises all change mid-order.
Order decision intelligence
An order is a decision journey, not a record
One order
- Checkout
- Payment
- Fraud
- Inventory
- Fulfillment
- Shipment
- Delivery
- Return or service
An order can experience
- inventory changes
- payment changes
- fraud signals
- warehouse constraints
- carrier disruption
- customer changes
- delivery exceptions
What should happen to this order now?
Intelligent fulfillment
Where it ships from is a decision, not a lookup
Fulfillment decisions may consider
- inventory location
- customer location
- delivery promise
- warehouse capacity
- store capacity
- carrier
- shipping cost
- split shipment
- margin
- customer value
- return probability
Potential actions
- SHIP
- SPLIT
- ROUTE
- REROUTE
- SUBSTITUTE
- EXPEDITE
- HOLD
- CANCEL
Example: fulfillment decision
The cheapest route is not the one that keeps the promise
A constrained center and a nearby store
Customer
Premium subscriber
- PromiseNext-day delivery
- Distribution centerItem available, capacity constrained
- Nearby storeInventory available
- Alternative centerInventory available, higher shipping cost
AnantDecision
FULFILL FROM STORE
The current state best supports the customer promise and the business objective at the same time, which the cheapest route did not.
The distribution center is cheaper and it is constrained. The nearby store is not, and the promise is next-day.
Delivery exception intelligence
A delay should not produce the same response for every customer
Relevant context
- customer value
- order value
- delay severity
- product type
- delivery commitment
- history
- available alternatives
- service policy
Possible actions
- NOTIFY
- EXPEDITE
- REPLACE
- REFUND SHIPPING
- OFFER CREDIT
- ESCALATE
Turn delivery events into customer decisions.
Returns and refunds
A return decision is an economic decision
Returns combine
- customer experience
- cost
- inventory
- fraud
- policy
- operational workload
Potential decisions
- REFUND WITHOUT RETURN
- RETURN + REFUND
- EXCHANGE
- STORE CREDIT
- INSPECT
- ESCALATE
- DECLINE according to applicable policy
Example: intelligent return
The decision considers total economics, not just policy
A return that costs more than it recovers
Customer
Five-year relationship
- ProductLow-value household item
- Return reasonDamaged
- Shipping costExceeds the recovery value
- Customer historyLow return frequency
- FraudNo meaningful signal
AnantDecision
REFUND WITHOUT RETURN
The customer is made whole, the reverse-shipping cost is avoided and no policy was stretched to do it.
Policy would accept the return. The shipping cost of accepting it exceeds what the return recovers, and the customer has given no reason to doubt them.
Agents, autonomy and assurance
Agents can already converse, plan, retrieve and call tools. The open question is which of their decisions an enterprise is willing to let execute unattended.
AI agents and agentic commerce
Commerce is moving toward a world where agents transact
Agents increasingly
- shop
- recommend
- negotiate
- purchase
- service
- refund
- merchandise
- price
- procure
- operate workflows
That creates a new commerce architecture, and the architecture needs a decision layer that is not inside the agent.
Your agent is not your decision policy
Enterprise decisions should not depend on what an agent concludes inside its own reasoning loop
Agents are excellent at
- conversation
- reasoning
- planning
- retrieval
- tool use
- workflow orchestration
Decisions that should not rest on the agent alone
- refund $1,000
- change price
- release a suspicious order
- issue account credit
- approve seller payout
Anant creates a separate enterprise decision layer, so the agent's reasoning and the business's authority are different things with different owners.
Agentic commerce architecture
Agents orchestrate. Decision intelligence decides. Assurance controls.
Customer agents, merchant agents and the decision layer
Customer agentshops, compares, negotiates
Merchant AI agentanswers, proposes, orchestrates
ANANTDECISION
Determines the enterprise next best action, returned as structure rather than conversation.
- QUOTE
- OFFER
- ALLOCATE
- APPROVE
- REFUND
- FULFILL
ANANTVERIFY
Evaluates whether this consequential action is authorized and assured.
ALLOW · HOLD · ESCALATE · BLOCK
Commerce systems
- pricing service
- inventory and allocation
- checkout and payments
- order management
- service desk
- marketplace APIs
The agent proposes in conversation. The decision layer answers in structure, and a verdict decides whether the answer may execute.
The agent stays where it is strong — understanding intent and orchestrating work — and the decision and its permission come from layers with an enterprise-wide view.
Shopping agents
Commerce where the customer is represented by an agent
The first generation of digital commerce put a human in front of a website. Increasingly the interaction is agent to agent, which changes what has to be decided and how fast.
What changes
- discovery
- product comparison
- pricing
- offers
- inventory negotiation
- checkout
- service
Merchants will need decision infrastructure that can respond programmatically and safely, which is a different requirement from a faster website.
Machine-to-machine commerce
Commerce becomes programmable
The merchant has to determine
- inventory
- price
- discount authority
- customer terms
- fulfillment
- credit
- margin
- contract conditions
AnantDecision returns a structured commercial decision rather than a conversation, and AnantVerify assures the commitments that carry consequences.
Decision transformation testing for commerce
Test the behavior of the decision, not merely one answer
Six changes to one refund decision
- BaselineAn AI agent decides to approve a refundApprove
- Irrelevant changeSomething immaterial to the decision changesThe decision is unchanged
- Return windowThe return window has expiredIt should change
- Customer relationshipA long relationship replaces a new accountIt may legitimately matter
- Linked-account signalA linked-account fraud signal appearsThe decision has to respond
- Combined conditionsHigh value, a fraud relationship and an expired windowNot the same as any one of them
Sensitive to what matters, stable to what does not.
Start from a decision the system gets right. Then change one condition at a time and ask what should move and what should not.
Protect against interaction failures
Every condition can pass while the combination fails
| Condition | On its own |
|---|---|
| Promotion valid | Pass |
| Customer eligible | Pass |
| Inventory available | Pass |
Promotion plus loyalty credit plus bundle discount can still violate a margin requirement, and testing the conditions one at a time is exactly how that is missed.
AnantVerify can evaluate interaction behavior around consequential decisions.
Decision boundaries
Test where decisions change, not only where they are comfortable
Commerce is full of boundaries
- refund amount
- discount authority
- free-shipping threshold
- fraud risk
- inventory level
- credit limit
- promotion eligibility
- seller payout
- customer tier
AI behavior should change appropriately when one of them is crossed, and the boundary is where a decision is most likely to be wrong in a way nobody notices until it is expensive.
Marketplaces, B2B and cross-border
Not every digital commerce decision has two parties and one jurisdiction. Marketplaces, B2B accounts and cross-border orders each add parties to the decision.
Marketplaces
A marketplace decision has more parties than a checkout
The network involves
- buyer
- seller
- product
- listing
- transaction
- payment
- payout
- delivery
- return
- reputation
- fraud
- platform policy
Potential decisions
- approve seller
- rank listing
- release payout
- hold transaction
- resolve dispute
- refund buyer
- investigate seller
- remove listing
Seller intelligence
A seller score is not a seller decision
A marketplace can continuously evaluate
- seller performance
- fulfillment
- returns
- customer complaints
- product quality
- pricing
- fraud signals
- relationships
- policy state
Instead of a static seller score: determine the appropriate next action.
B2B e-commerce
Contracts add state that consumer checkout does not have
Relevant state may include
- account
- contract
- negotiated pricing
- credit
- inventory
- purchase history
- approval
- delivery
- payment terms
- sales relationship
- service level
Potential decisions
- QUOTE
- PRICE
- APPROVE
- ALLOCATE
- EXTEND TERMS
- EXPEDITE
- ESCALATE
Example: B2B order
Order arrival time is not the contract
A contracted account and scarce inventory
Customer
Strategic enterprise account
- OrderLarge volume
- InventoryLimited
- ContractPriority allocation
- CreditWithin approved exposure
- Other demandSeveral lower-priority orders pending
AnantDecision
PRIORITIZE INVENTORY FOR CONTRACTED CUSTOMER
The decision understands the commercial relationship rather than the order arrival time, and the commitment was already made.
Three conditions would let a first-come system allocate the stock elsewhere: limited inventory, other orders pending and a large volume. The contract is the fourth condition.
Social commerce
Commerce increasingly begins outside the storefront
Discovery and purchase happen through
- social platforms
- creators
- messaging
- live commerce
- embedded experiences
One decision layer across every digital touchpoint.
Cross-border commerce
The right action can change with the geography
Cross-border decisions may involve
- currency
- tax
- duties
- shipping
- inventory
- fraud
- payment methods
- local policies
- customer promise
- regional operations
A decision that is correct in one market can be unavailable, unprofitable or non-compliant in another, which makes the operating state part of the decision rather than context around it.
Inventory, merchandising and demand
Digital commerce makes every inventory promise visible the moment it is made, and every demand change visible the moment it happens.
Inventory intelligence
Inventory becomes part of customer decision intelligence
AnantDecision can continuously determine
- ALLOCATE
- RESERVE
- TRANSFER
- REPLENISH
- EXPEDITE
- SUBSTITUTE
- CHANGE PROMISE
Availability is shown to the customer continuously, so the decision and the promise have to be made from the same state.
Digital merchandising
Ranking and featuring are commercial decisions
Merchandising decisions can incorporate
- search demand
- customer behavior
- product affinity
- inventory
- margin
- seasonality
- promotion
- supplier
- product lifecycle
Potential actions
- FEATURE
- RANK
- BUNDLE
- PROMOTE
- MARKDOWN
- SUBSTITUTE
- REMOVE
Supply chain intelligence
Supply events arrive as customer-facing promises
E-commerce supply chains must respond to
- demand spikes
- supplier delays
- inventory imbalance
- carrier disruptions
- warehouse capacity
- weather
- returns
- customer promises
AnantDecision turns these events into operational actions.
Example: viral demand
React to the trajectory, not just the stockout
A product trending faster than its supply
Event
Product trending on social media
- Digital stateProduct views increasing rapidly
- OrdersAcceleration detected
- InventoryAdequate today
- SupplierLead time 14 days
- MarketingPaid campaign scheduled tomorrow
- If nothing changesOut of stock
AnantDecision
ADJUST CAMPAIGN + RESERVE INVENTORY + EXPEDITE SUPPLY
The decision was available four days before the stockout, and only to a system reading the trajectory rather than the stock level.
The stock is adequate today. The supplier needs fourteen days and the campaign starts tomorrow, which is the whole of the problem.
Demand is a trajectory
Direction of change matters as much as current state
Same sales today, different direction
- Product AStable for six weeks. Reorder at the usual cadence.
- Product BAccelerating over six hours. The stock decision cannot wait for the trend to finish.
Two products, identical revenue today, opposite implications for tomorrow’s inventory.
Two products with identical sales today support different inventory decisions, because one of them is accelerating.
Current state matters. Direction of change matters too.
The business case and what to measure
Commerce value is built from the merchant's own economics, measured at the level of the decision, and disclosed as a model rather than a result.
The business case
Seven economic levers
Illustrative business cases
Six worked assumptions
Six illustrative business cases
No figure on this panel is a measured result or a promised outcome.
Conversion
Assumptions
- 100M annual sessions
- 5M checkout sessions
- 60% current conversion from checkout = 3M orders
- $20 average contribution per order
- 0.5 percentage points of additional checkout conversion
25,000 incremental orders × $20
$500,000 of modeled annual contribution
Also excludes
- the rest of the funnel, which is not modeled here
- this is an illustrative scenario, not a performance claim
Discount optimization
Assumptions
- 2,000,000 discounted orders a year
- $12 average discount
- annual discounts of $24M
- 5% of discounted orders where an incentive is unnecessary
100,000 orders × $12
$1.2M of modeled avoided discount
Also excludes
- customer-response and incremental-demand effects
- this is an illustrative economic model
Returns
Assumptions
- 1,000,000 annual returns
- 20% reviewed manually = 200,000
- $7 average review cost
- current modeled review cost $1.4M
controlled automation moving manual review from 20% to 12%
80,000 reviews leave the manual queue, modeled at $560,000 a year
Also excludes
- fraud effects
- customer-experience effects
- this is an illustrative economic model
Customer service
Assumptions
- 3,000,000 annual commerce service cases
- 8 minutes average resolution time
- 60 seconds saved per case through better state and next-best-resolution intelligence
3,000,000 × 1 minute
50,000 service hours a year
Also excludes
- the merchant should value those hours using its own labor economics
Fraud review
Assumptions
- 500,000 orders requiring manual fraud review annually
- $10 average review cost
- $5M annual review cost
- a 10% reduction in unnecessary review
10% of $5M
$500,000 of modeled annual review capacity
Also excludes
- redirected toward higher-risk transactions rather than removed
- this is an illustrative economic model
Fulfillment
Assumptions
- 10M annual orders
- improved routing reduces average fulfillment cost by $0.10 per order
10M × $0.10
$1M of modeled annual difference
Also excludes
- the promise-accuracy effects that motivated the routing decision
- this is an illustrative economic model
Each case is arithmetic on the inputs above. Three cases apply a rate to a volume instead of a measured uplift, and they are presented as models rather than as results.
Six cases, each built from stated inputs. Three of them rest on a share of a population rather than on a measured uplift, and all six are labeled as models rather than expectations.
This is an illustrative economic model, not a promised outcome.
Build ROI from commerce economics
Measure at the decision task, then aggregate
A decision task with no baseline cannot be defended, and a program made of those cannot be either.
Executive e-commerce KPIs
The measures a commerce executive already owns
Track
- GMV and revenue
- gross margin
- conversion
- average order value
- customer retention
- repeat purchase
- cart abandonment
- return rate
- fraud loss
- fulfillment cost
- customer acquisition economics
- decision automation
- value generated by decision programs
Commerce domain KPIs
The same program read from four functions
Customer decisions
- next-best-action acceptance
- recommendation conversion
- offer incrementality
- discount efficiency
- retention
- repeat purchase
- customer lifetime value
- service resolution rate
- decision-to-purchase time
Commerce
- search-to-product conversion
- product-to-cart
- cart-to-checkout
- checkout conversion
- payment success
- order completion
- abandonment
- promotion effectiveness
- margin per order
Fulfillment
- cost per shipment
- split shipment
- order cycle time
- on-time delivery
- promise accuracy
- inventory allocation
- cancellation
- substitution
- rerouting
Fraud and returns
- fraud loss
- false-positive rate
- manual review
- review cost
- return rate
- refund cycle time
- refund abuse
- investigation yield
- customer friction
Decision intelligence KPIs
Whether the decisions themselves are healthy
Track
- decision volume
- decision latency
- automation rate
- confidence calibration
- abstention
- decision stability
- boundary behavior
- model disagreement
- human override
- outcome after decision
Assurance KPIs
Independence, measured rather than asserted
Track
- state sufficiency
- stale state
- evidence conflicts
- authority conflicts
- ALLOW / HOLD / ESCALATE / BLOCK
- assurance coverage
- interaction failures
- execution mismatch
- decision replay completeness
Optimize the whole commerce decision
The best model score is not necessarily the best business decision
Commerce organizations optimize separate models for
- clicks
- conversion
- fraud
- shipping cost
- margin
- retention
Those objectives interact:
- A discount can improve conversion while reducing margin
- A fraud rule can reduce losses while rejecting good customers
- A fulfillment choice can reduce cost while breaking the delivery promise
AnantDecision operates around the decision and its complete business context rather than around whichever model happens to own the surface.
One decision, then the next
The path is one decision with measurable economics, proven, then expanded — and the destination is a commerce architecture that can be more autonomous without becoming less accountable.
Start with high-value decision journeys
Where volume is high and conditions move
These are the journeys where a decision repeats often enough to be measured and where the state is already available in the systems.
Strong starting points
- next best action
- cart intervention
- checkout decisions
- fraud disposition
- returns and refunds
- customer-service resolution
- fulfillment routing
- inventory allocation
- promotion optimization
- subscription retention
Prove one decision, then expand
Nine stages from one decision to a commerce architecture
One journey, nine stages
One decision becomes an enterprise commerce intelligence architecture.
The order matters. A baseline taken after the workflow changed is not a baseline, and a decision task with no baseline cannot be defended.
From e-commerce to agentic commerce
The next generation is agentic and adaptive
Four generations
- Digitized storefront
- Personalized experience
- Connected channels
- Agentic and adaptive
The first generation digitized the storefront. The second personalized the experience. The third connected every channel. In the next one, customers increasingly interact through AI, merchants operate AI agents, and transactions increasingly become machine-to-machine.
The infrastructure underneath that future needs a decision layer that can answer two questions rather than one.
What should happen?
Should it be allowed to happen?
Build the commerce engine that knows what to do next
Every click changes the state
Every one of these changes the state
- click
- search
- cart
- order
- payment
- inventory change
- delivery
- return
- service interaction
- AI action
Every state change creates a decision opportunity.
And every consequential action can be independently assured before it executes.
Decide the moment, assure the consequence
Every click, search, cart, order, payment, inventory change, delivery, return, service interaction and AI action changes the state. The engine turns those changes into decisions — and assures the consequential ones before they execute.
ANANTDECISIONReal-Time Commerce Decision Intelligence
ANANTVERIFYIndependent Decision Assurance
- Click
- State
- Decide
- Verify
- Act
- Learn
From click to state. From state to decision. From decision to trusted action.
Worked at decision level: Fraud Response and Customer Refund Authorization.
AnantVerify does not by itself establish regulatory compliance. It provides decision evidence, testing, control and replay that can support a retailer's governance, payment-control and customer-treatment programs.
Illustrative throughout. Every session count, order volume, rate, cost 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 merchant's own economics.