Attribution Tells a Story About Credit; Incrementality Tests Whether Marketing Added Value

October 24, 2026•3 min read

Attribution and incrementality answer different questions. Attribution assigns credit under a chosen rule. Incrementality asks whether an outcome would have happened without the marketing action. A campaign can receive attribution for a sale it did not cause, and an influential campaign can receive little credit under a narrow attribution rule.

For intent-data buyers, this distinction prevents a customer list, a matched record, or a late touchpoint from being mistaken for proof of added revenue.

Define the attribution rule explicitly

State which events are eligible for credit, the lookback window, the unit of analysis, and how multiple touches are handled. Keep the rule stable when comparing periods.

An account that appears on an intent list and later buys is not automatically an intent-generated customer. It may already have been in an active opportunity or arrived through a referral. Preserve the earlier relationship and campaign exposure history.

Attribution can still be useful for operational reporting. Its limitation is that the credit rule does not establish the counterfactual by itself.

Ask what would have happened otherwise

A test of incremental value needs a credible comparison. Where feasible, randomly assign eligible units to treatment and a suitable control, then measure outcomes under a pre-specified design.

The unit might be an account rather than an individual contact if several people share the buying decision. Avoid contamination in which treated and control contacts belong to the same account and influence one another.

Preserve essential service and existing obligations. A test should vary a discretionary marketing action, not deny customers required support.

A hypothetical prioritization test

Suppose the business wants to know whether an intent-prioritized research queue improves qualified opportunities. One group follows the proposed policy and another follows the current process, with comparable resources and documented rules.

The result would evaluate the policy as implemented. It would not necessarily isolate every component of the score, message, and sales behavior. Describe the treatment accurately.

If the groups are selected retrospectively rather than assigned under a controlled design, the comparison may still be informative but requires stronger caution about alternative explanations.

Choose the outcome before the campaign

Use a business outcome appropriate to the decision, such as qualified opportunities or realized contribution within a defined horizon. Track intermediate events separately.

Do not change the primary outcome after seeing which metric looks best. Exploratory findings can be reported as exploratory, with a plan for confirmation.

Allow enough follow-up for the sales cycle. A campaign can appear ineffective if revenue is evaluated before purchases normally occur. It can also appear stronger if unresolved records are selectively omitted.

Keep learning inputs separate from campaign credit

Cortex can learn from customers supplied to the process even when they were not on the original list. Those records can improve understanding without becoming evidence of campaign acquisition.

Maintain separate records of training or feedback inclusion and marketing exposure. This avoids a circular story in which every customer used by the model is later claimed as a customer produced by it.

The same principle applies to existing customers. Their expansion purchases may support an expansion study, but they should not become new-customer wins in an acquisition report.

Interpret a positive result carefully

A useful test result applies to its audience, offer, period, and implementation. It is not a guarantee of the same outcome for every future client. Report counts, uncertainty, and meaningful execution differences.

If the result is inconclusive, investigate whether the test had enough information to detect a commercially meaningful effect. Inconclusive does not mean proven ineffective, but it also does not justify a positive claim.

Do not convert association into causation because the headline is easier to sell. The distinction is central to a credible benchmark program.

Bring the right question to Cortex

Ask whether you are evaluating source quality, scoring, sales prioritization, or a specific marketing treatment. Each question can require a different design.

Plan an intent-data evaluation with Cortex. Bring the decision you want to justify and the current alternative. A clear comparison makes it possible to discuss value without relying on attribution alone.

Back to Blog