Knak surveyed 333 enterprise marketing decision-makers and found the least surprising and most expensive thing in the industry: marketers are 68% more likely to judge an email or landing page by click-through rate than by revenue or pipeline it influenced. Sixty-nine percent measure clicks. Forty-one percent track revenue or pipeline.
Meanwhile the production line got dramatically faster. AI now drafts the copy, builds the module, checks the brand rules, runs the localization. The output side of the shop was rebuilt. The measurement side is running the same instrument it ran in 2014.
We build this exact category of system for a living, so here is the blunt version. Speed without measurement is not efficiency. It is just volume with a shorter feedback loop, and a shorter loop around a metric that does not represent money will get you to the wrong place faster than you have ever gotten anywhere.
Read the sample before you read the number
Standard disclosure first, because this piece is about measurement rigor and it would be absurd to skip it. Knak sells marketing production software. Knak commissioned the survey, the sample includes Knak's own customers, and it is restricted to enterprises above $50 million in revenue already running an enterprise marketing automation platform. MarTech, which reported it, is owned by Semrush.
None of that makes the finding false. It does tell you which direction to discount. If anything, the bias runs against the headline: this is the most tooled-up, best-resourced end of the market, the respondents most likely to already have revenue reporting wired in, and they are still counting clicks. Whatever the number is at a 40-person company with a marketing coordinator and a spreadsheet, it is not better than 41%.
The gap is not data access, it is nobody's job
The most useful sentence in the report is the one that removes your favorite excuse. Most enterprise marketing automation and CRM platforms can already connect email engagement to opportunity records. The plumbing exists. It shipped. You are paying for it.
What is missing is the process and the reporting built on top: the definition of an influenced opportunity, the agreement with sales about what counts, the job of maintaining it when a field gets renamed, and a named human whose performance review contains the words "this report is correct."
That is not a technology gap. That is an operations gap, and it stays open because it is nobody's Q3 objective. Clicks arrive free in a dashboard nobody has to defend. Revenue influence requires a cross-functional argument, a data model, and someone willing to publish a number their own campaign might fail against. One of those is comfortable. Guess which one has 69% adoption.
Every organization measures what is cheap to measure and then, over time, starts believing that is what it values.
What a click actually tells you
To be fair to the click, it is a real diagnostic. It is the right instrument for a subject line, a preheader, a hero module, a call-to-action placement. If clicks collapse, something broke upstream, and you want to know that within the hour.
The failure is one of promotion. A diagnostic metric got promoted to a performance metric, because it was the one always in the room. And a diagnostic cannot answer the only question the company actually cares about: did this influence an opportunity, and did that opportunity turn into revenue.
The consequence compounds quietly. Optimize a year of production against click rate and you systematically breed a content library that is excellent at getting itself opened and unexamined at getting anything sold. Then AI arrives, multiplies your output volume, and you scale that library faster. Knak's respondents are already living in the result: only one in three say they consistently meet or exceed performance targets on email and landing page campaigns. Two thirds of a heavily-tooled enterprise cohort, missing targets, with a fast production line and a scoreboard that cannot tell them why.
Adoption stopped being the differentiator
Only 29% of respondents rated their organization an advanced AI adopter, and the profile of that group is the actionable part of the whole study. They were not distinguished by having better models. They were distinguished by operating discipline.
They put AI agents on the structural work / building and coding emails and landing pages, running brand and compliance checks, handling translation and localization. Everyone else used AI mostly for first-draft copy and image generation. Note the asymmetry: the advanced group aimed the machine at the repeatable, verifiable, rule-governed stages, which are precisely the stages where a machine's output can be checked automatically. The rest aimed it at taste.
They also ran structured project management and native approval workflows rather than improvised ones, and they were more willing to buy a dedicated tool when they found a capability gap instead of forcing an existing platform to fake it. That last habit is the least glamorous and the most predictive. Stretching a platform two feet past its design is how organizations acquire the invisible manual step that eventually becomes the constraint on everything.
Meanwhile 88% of all respondents said AI-generated content still needs moderate or substantial editing before use. That figure will not surprise anyone who read The Second Shift: the cleanup is real, it is universal, and it is still mostly unbudgeted. But it is the second story here. The first story is that the shop got faster at making things and no faster at knowing which things worked.
The playbook
DMAIC at the strategic layer, disciplined delivery underneath. Six moves, in the order we would run them.
Define the outcome metric before you touch the tooling. One sentence, agreed with sales and finance, on what an influenced opportunity is: which touch types count, over what window, with what credit rule. Boring, arguable, unavoidable. Every measurement program that never launched died in this sentence, and every dashboard built without it is decoration.
Connect engagement to opportunity records, since you already own the connector. This is a configuration and data-hygiene project, not a purchase. Campaign IDs that survive the round trip, consistent naming, one system of record, one owner. Expect the ugly part to be historical data, and do not let cleaning the past block instrumenting the future.
Keep clicks, demote them. Split the reporting explicitly into diagnostics and outcomes. Clicks, opens and form completions on the diagnostic panel where they belong. Pipeline influenced, opportunities touched, revenue influenced, and cost per usable asset on the outcome panel. Same data, radically different behavior, purely because of what the executive readout leads with.
Point the machine at the verifiable stages first. Follow the advanced cohort. Build and code, brand and compliance checks, localization, QA. These have deterministic right answers, which means AI failures surface immediately instead of six months later in a brand audit. Use it for first-draft copy too if you like, but understand that stage carries a supervision cost the structural stages do not.
Instrument the production line itself. If AI is meant to speed up production, measure production: assets shipped, cycle time from brief to live, rework rate, edit hours per asset, cost per usable asset. Right now most teams cannot state their rework rate, which means they cannot prove the AI investment paid off in either direction. Vendors love that. You should not.
Buy the gap, do not stretch the platform. When a capability is genuinely missing, the honest fix is a tool that does it. The expensive fix is a manual workaround nobody documents, performed by one person who eventually goes on vacation. The advanced adopters figured this out and it is most of what makes them advanced.
Our position
There is a version of the AI story where the technology is the hard part. This is not that story. The hard part is that a marketing organization is a factory, and the industry spent three years upgrading exactly one station in it while leaving the quality control, the routing and the ledger untouched.
A faster factory with an unchanged scoreboard does not produce better results. It produces more of whatever the scoreboard rewards, sooner. If the scoreboard rewards clicks, congratulations, you now generate industry-leading quantities of clicked-on material and still cannot tell your CFO what it bought.
The next competitive advantage in marketing is not a better model, because everyone has the same models. It is being the company that can say, with a straight face and a defensible number, what its production actually earned. That is a governance and instrumentation problem. It is unglamorous, it is completely buildable, and it is the job.
Produce at machine speed if you like. Just make sure something in the building is still counting.