--- title: "Attribution Is Not the Same as Lift" description: "An attributed order carried our discount code — that isn't proof it wouldn't have happened anyway. Here's what a holdout tells you that attribution can't." canonical: https://navona.ai/blog/attribution-vs-measured-lift generator: scripts/generate-agent-md.mjs published: 2026-08-26 category: analytics tags: ["attribution","incrementality","measurement"] --- # Attribution Is Not the Same as Lift Every vendor in this space — including us — reports an "attributed" revenue number. It's the easiest number to produce, and the easiest one to misread. Here's what it actually measures, and what it doesn't. ## What "attributed" means An order is attributed to Navona when it was placed using a discount code Navona issued during that shopping session. That's the entire definition. It says nothing about what the shopper would have done if the code had never appeared. ## The gap between attribution and lift Some share of shoppers who redeem an intervention offer were already going to buy. They add the item to cart, get an unprompted offer on the way out, use the code because it's there, and complete a purchase they would have completed anyway. That order still shows up as "attributed" — the code was used — but it added nothing. No incremental revenue was created by the intervention. The reverse also happens, less visibly: an offer occasionally nudges someone who would have bought anyway to buy something slightly larger, or sooner rather than in a later session they might not return for. That's also attribution, and it's also not proof of causation on its own. The only figure that answers "how many of these orders would not have existed without the intervention" is *lift* — the incremental effect of the intervention, isolated from what would have happened anyway. ## Why a holdout is the only way to see it The only way to isolate lift is to deliberately withhold the intervention from some share of otherwise-eligible shoppers — a holdout group that sees no offer, nothing different from browsing the store before any intervention existed at all. Their purchase rate is compared against the group that did see the intervention. The difference between the two groups, not the raw count of attributed orders, is the honest measure of what the intervention actually did. Running a holdout costs real money. Every shopper routed into it is a shopper who, statistically, converts less often without the intervention live — that's the whole premise for building the intervention in the first place. A merchant running a holdout is deliberately leaving some short-term revenue on the table in exchange for an honest read on long-term causation. ## Why the distinction matters A vendor who reports attribution and calls it lift usually isn't lying about the number — it's typically a real, correctly-computed figure. What's misleading is the label. "This tool drove $X" implies causation. "$X in orders carried this tool's discount code" is the same number, described accurately. The gap between the two claims is exactly the revenue that would have happened with or without the tool — and no one selling the tool has much incentive to go isolate that gap. ## Where our own numbers sit We report attribution, and we want to be explicit that it's attribution. Across 38 stores, Navona has logged 12,950 orders that carried one of our discount codes, adding up to $1,357,503 in attributed revenue against $129,415 in discounts given — a blended 10.5× return on every dollar discounted, with the monthly figure ranging from 6.7× to 13.7× (measured 2026-08-25). Those are real, verifiable numbers. On their own, they are not a measurement of incremental lift, and we're not going to describe them as one. There's a second distortion worth flagging alongside this: raw session and cart counts on most stores include a meaningful share of bot traffic, which inflates the pool an attribution rate gets divided by. [More on that separately](https://navona.ai/blog/how-much-of-your-cart-abandonment-is-bots). ## What to ask any vendor, including us - Is this number attribution or lift? - If it's attribution, do they run a holdout at all? - If they run a holdout, what share of traffic is in it, and for how long? If the answer to the second question is no, the vendor can't actually tell you what their tool is worth to your business — only what it touched. [See what a holdout-backed measurement costs](https://navona.ai/pricing) ## FAQ ### What does "attributed revenue" mean? An order is attributed when the shopper used a discount code issued during that session. It confirms the code was redeemed — it does not confirm the order would not have happened without it. ### What's the difference between attribution and incremental lift? Attribution counts every order that carried a discount code, including ones the shopper would have placed anyway. Lift is the difference in purchase rate between shoppers who saw an intervention and a holdout group who saw nothing — it isolates the effect the intervention actually caused. ### Why would a merchant deliberately show no offer to some shoppers? A holdout group that sees nothing is the only way to know what would have happened without the intervention. Comparing that group's purchase rate to the group that did see an offer is what turns a raw order count into a measurement of real incremental value. ### Is a 10x return on discount spend the same as a 10x lift? No. A return-on-discount figure divides attributed revenue by discounts given — both real, measured numbers — but it does not account for orders that would have happened without any discount at all. Only a holdout comparison answers that question. ### Should a merchant distrust every attribution number a vendor reports? Not distrust — just read it for what it is. An attribution figure can be accurate and still overstate incremental value if no holdout was run. Ask whether the number being quoted is attribution or lift, and whether a holdout backs it.