Riskified’s growth needs to be read after risk costs

Riskified’s November 11, 2026 Q3 release: revenue $115.2 million, up 41%; GAAP gross margin 43% versus 50% a year earlier; gross profit up 20%. Those measures describe one business, but not one simple growth story. Results

The question is how to judge incremental revenue when a technology supplier also accepts some transaction risk. This is original business analysis, not an assertion about the cause of a particular quarter’s losses or an investment recommendation.

Transferred risk changes what revenue means

The annual report describes risk-adjusted fees on approved transaction value for Chargeback Guarantee; eligible chargeback reimbursements enter cost of revenue. 2025 annual report Each dollar therefore cannot be interpreted as though it belonged to a software service that bears no transaction losses.

A supplier selling analytical tools can leave the consequences of a decision with its customer. A supplier accepting specified reimbursement obligations must price the consequences of being wrong. The important issue is not an industry label, but the responsibility attached to the fee. A more expensive order is not necessarily a more profitable one: a higher charge may simply compensate for higher expected costs.

An increase in average fees could reflect greater customer value, a different customer mix or additional risk. Aggregate revenue alone cannot separate these possibilities. Calling all such growth stronger pricing power would award credit for an improvement not yet demonstrated.

A hypothetical example separates sales from contribution

Imagine two transaction groups, each producing 100 units of fee revenue. One requires 30 units of fulfilment and loss costs; the other requires 55. Equal sales leave 70 and 45 units respectively for development, selling and administration. These are illustrative units, not Riskified’s actual fees or loss experience.

The second group might still be worth accepting if it is less costly to acquire or supports enduring customer relationships. The test is whether incremental gross profit covers incremental operating resources, with sufficiently reliable risk estimates. A lower margin is not an automatic rejection signal, just as faster revenue growth is not an automatic endorsement. The comparison must include resources committed, obligations accepted and contribution retained.

This is why gross profit dollars and gross margin belong together. The former asks whether resources available to support the company increased; the latter asks how much remains from each sales dollar. They can move in opposite directions without an accounting contradiction. Declaring deteriorating business quality from a falling percentage, or a widening competitive advantage from rising sales, skips the same intermediate steps.

Scale must survive both losses and expenses

The product page presents approval improvement alongside chargeback protection. Product description For an analyst, the constraint between those aims is more informative than an isolated accuracy statistic.

Rejecting more genuine customers might reduce losses while costing a merchant sales. Relaxing decisions to approve more orders might increase reimbursements. A valuable system improves that trade-off rather than merely optimizing one side. Evaluation should likewise compare approvals, losses and service costs under comparable transaction conditions.

Scale can spread fixed development expense; it does not automatically eliminate risk costs. Additional observations may improve models, but their benefit is neither immediate nor uniform nor assured. Different products, markets and customer behavior can weaken historical patterns. Conversely, a temporarily less favorable mix need not prevent lasting customer relationships and greater absolute gross profit from funding investment. Consolidated figures do not reveal which force dominates. That requires comparable evidence, not a retrospective story.

Agent shopping is an opportunity awaiting proof

The October 8 Agent Identity Risk Intelligence announcement schedules a limited enterprise beta for December 2026 and anticipates general availability in 2027. It is a product plan, not realized revenue. Original announcement

A new shopping interface should not be confused with a measured new market. Agents may change how orders originate, but willingness to pay still depends on reducing net losses, preserving genuine transactions and fitting existing workflows. A recognition signal becomes an enduring paid capability only if it improves decisions and outcomes.

For example, automation could reduce checkout effort yet increase requests after a purchase. A risk service cannot evaluate the economics of orders independently of refunds. This is a possible mechanism, not a claimed company finding. A meaningful test would compare complete transaction cycles before and after adoption while controlling for customer mix. Agent-order counts alone would not establish success.

The evidence that would make the next report more useful

Comparable cohort gross profit, estimated versus realized losses and the operating burden of new customers would clarify the economics. Expense control also needs interpretation: repeatable efficiency differs from postponing necessary investment. An adjusted profit measure alone can miss the economic cost of equity compensation, ongoing maintenance or acquiring customers.

The results matter precisely because growth and cost signals are not moving in perfect alignment. Evaluating business quality requires neither dismissing expansion nor treating a technology narrative as profit protection. It requires asking under what conditions revenue was earned, what obligations accompanied it and how much resource remains available for sustained investment after a sufficiently long observation period.