An AI-source report shows new orders. That is useful evidence that shoppers are reaching the store through an identifiable path. It is also the beginning of the business question: what did those orders contribute, and is further investment justified?
Order count and revenue cannot answer that alone. Discounts, product costs, delivery, payment processing, returns and channel costs all change what remains. During Black Friday, a successful promotion can increase unit sales while reducing contribution per order. The same calculation matters throughout the year.
A sound evaluation needs three distinct views: the orders associated with AI, their contribution after relevant costs, and the additional profit the investment caused. Keeping those views separate makes a channel report more useful for a budget decision.
What remains from an $80 order?
Illustrative calculation only. The 10% service/channel fee below is a fictional modeling assumption. It is not DeepLumen's actual fee, a published or announced future fee, or an indication of what DeepLumen will charge. This is not merchant performance data.
Assume a product listed at $100 sells for $80 during a promotion. The $20 discount is already reflected in product revenue; do not deduct it again. Sales tax and other pass-through amounts are excluded from revenue here. Assume no shipping revenue is collected from the shopper, so the merchant bears the shipping cost shown below.
| Item | Amount per order (USD) |
|---|---|
| Net product revenue after discount | 80 |
| Product cost | -30 |
| Warehousing, packing and delivery | -10 |
| Payment processing | -3 |
| Expected net returns and support loss | -5 |
| Allocated cost of this channel test | -2 |
| Assumed channel fee: 10% of product revenue | -8 |
| Estimated contribution profit after test-cost allocation | 22 |
The arithmetic is $80 − $30 − $10 − $3 − $5 − $2 − $8 = $22. On this defined revenue base, the contribution margin after the test-cost allocation is 27.5%.
This preserves a complete view of the example's order-related costs. It also requires explicit accounting boundaries:
- Product cost ($30): the assumed cost of the item sold. If inbound freight or duties are included here, do not add them again elsewhere.
- Warehousing, packing and delivery ($10): the allocated fulfillment cost for this order, including outbound delivery. Exclude any warehouse cost already included in another allocation.
- Payment processing ($3): an assumed dollar cost, not a published processor rate. Replace it with the actual charges and any relevant refund treatment.
- Expected net returns and support loss ($5): a reserve for expected refunds, reverse logistics, unrecoverable product value and support, net of recoveries and costs recognized elsewhere. Reconcile it with actual outcomes rather than subtracting both the reserve and the same realized loss.
- Allocated test cost ($2): this order's share of defined content, tooling or evaluation work. It excludes the $8 fee and all other listed costs. Record the total test budget and allocation method so that changing order volume does not hide the investment.
- Assumed fee ($8): 10% of the $80 product revenue in this model only. The example excludes tax and shipping from that assumed fee base; an actual agreement may use different definitions.
The $22 is not net profit. It excludes unallocated fixed overhead, financing costs and applicable business taxes. Because the table includes an allocated test cost, it is a management view of contribution after that allocation; finance teams may present contribution before shared test spending and deduct the spending at channel level instead. Either presentation can work if definitions remain consistent.
If the shopper pays for shipping, include that revenue and its related costs consistently. If accounting revenue is already reduced for actual refunds, do not deduct those refunds again through the returns line.
Use sensitivity to set a commercial limit
Before the hypothetical fee, this order leaves $30 after all other listed costs. With the assumed $8 fee, it leaves $22. If fulfillment becomes $5 more expensive, the result falls to $17. If the promotional selling price falls to $70, the assumed percentage fee becomes $7 and contribution falls to $13, holding every other cost constant.
Those are planning scenarios, not predictions. In practice, payment charges, returns and product mix may change with price and volume.
A merchant that requires at least $20 per order after the other allocations could afford no more than $10 in channel fees under the original assumptions: $30 minus the $20 floor. That is an internal affordability limit for this scenario, not a suggested service price. The exercise helps teams decide whether to change the discount, shipping offer, product mix or investment level before adding volume.
Confirm performance-billing rules before modeling them
Cost per sale, often abbreviated CPS, generally links a fee to a qualifying sale. The word “qualifying” carries much of the commercial detail. Confirm:
- The fee base: discounted or list price, with or without shipping, tax and other charges.
- Attribution and eligibility: evidence required, attribution window, returning-customer treatment and handling of overlapping channels.
- Order adjustments: cancellations, partial and full refunds, chargebacks, fee reversals and reconciliation timing.
- Review rights: merchant-visible records and a process for resolving disputes.
As checked on October 21, 2026, DeepLumen's pricing page labels Agentic Page performance pricing as coming soon and publishes no universal commission rate. Applicable rates and validation rules are to be agreed before activation. Current installation terms should be checked through the app listing or onboarding. None of those terms can be inferred from the fictional 10% used here.
Attribution measures association; incrementality measures added value
Attribution assigns credit using a stated rule and available evidence. For example, a report might identify orders following a recorded AI referral within a defined window. It should disclose the rule, sample size and tracking gaps. Some journeys will remain unobserved or ambiguous.
Incrementality asks what would have happened without the intervention. A shopper might have purchased through search, email or a direct visit anyway. An AI-attributed order can be profitable without being wholly incremental, and a performance fee tied to that order does not establish causation.
Where feasible, evaluate deeper optimization through a randomized holdout or a carefully designed staged rollout across comparable product groups. Track total contribution as well as AI-attributed sales, since a channel may shift credit from an existing path. Record prices, stock, advertising and promotions; allow time for returns. A weak or small comparison supports a directional conclusion, not a precise incremental ROI claim.
Once the counterfactual is credible, a useful definition is:
Incremental ROI = (incremental contribution before the evaluated investment − investment cost) ÷ investment cost.
Define which fees, implementation work, subscriptions and ongoing labor belong to “investment cost.” If they have already been deducted in the contribution figure, add them back before using this formula, or use the equivalent formulation consistently. Summing the table's $22 across attributed orders and dividing by the allocated $2 would not establish incremental ROI.
Turn measurement into an operating decision
Agentic Page helps establish AI-readable merchant information and provides observed AI activity and available sales attribution. Profit evaluation still needs order, fulfillment and finance records; causal evaluation needs an appropriate test. DeepLumen's Proving AI-Driven Revenue framework offers further context for separating the measurement stages.
A useful review should show attributable orders and evidence, contribution after reconciled costs, the investment required and the confidence of any incremental estimate. That supports a concrete decision: expand a promising product group, repair an expensive fulfillment pattern, revise a promotion or continue testing until the evidence is stronger.
An AI channel creates durable commercial value when it brings worthwhile demand at a cost the business can sustain. Build the order economics first, review the applicable commercial terms, and let contribution and credible incremental evidence guide the next investment.