The CRO Metrics Worth Reporting to Leadership

Most conversion programmes are judged on the wrong number, and the programme chose it. Omniconvert has run CRO across 70,000+ experiments and 2,500+ Shopify stores over 13 years in eCommerce, and the pattern in programmes that lose their budget is rarely that the work was bad. It is that the reporting invited a conversation nobody could win: a headline conversion rate that moves with traffic mix, a percentage uplift the reader silently annualises, and a methodology debate in place of a decision. Three numbers fix it. This is what they are and why each one survives contact with a board.

Number one: realised money, with the assumption written down

So do the conversion yourself rather than leaving it to them. For each shipped change: the traffic it affects annually, the measured effect on the step it moved, the value of a conversion, and the margin. Multiply, then state the result alongside the single assumption it rests on. One line, visible, something like: assumes the observed effect holds on the same traffic for twelve months.

Then discount it, visibly, and say you have. A measured three-week effect does not hold perfectly for a year, novelty fades, and the person who found the win is the last person who should be optimistic about it. A report claiming a modest figure and explaining the haircut is believed. A report claiming a large figure with no haircut is discounted silently by the reader, usually by more than you would have.

Number two: one leading indicator

A leadership report carrying only realised money has a timing problem. The money arrives quarters after the work, so a good quarter of effort and a bad quarter of results are indistinguishable until much later. That gap is where programmes get cut.

Revenue per session, on the specific segment your tests are addressing, fills it. It moves when you improve things, it does not move simply because more people arrived, and it cannot be gamed by a change that lifts order count while dropping basket size. A conversion rate can, which is why a conversion-rate win sometimes coincides with flat revenue.

Name the segment explicitly and keep it fixed across reports. Revenue per session on mobile paid-social first-time visitors is a number that means something and can be trended. Revenue per session across all traffic is nearly as mix-sensitive as the conversion rate you were trying to escape. Before picking a target for it, it helps to know what the category looks like: that is covered in a good conversion rate and the related question of how those figures should order your work in how benchmarks should shape your test roadmap.

Number three: learning rate

A learning rate is defined as the number of tests completed and the share of them that produced a clear decision.

Learning rate is the least intuitive of the three and the one that most changes how a programme is perceived. It says: we ran eleven tests, nine reached a conclusion, two were inconclusive and here is why.

Its first job is to make failure legible. A test that disproved a planned redesign saved the cost of that redesign, and that is a real number you can put next to it. Without a learning-rate line, a quarter of honest negative results looks like a quarter of nothing, and the reporting pressure that follows is precisely what pushes teams into cherry-picking metrics.

Its second job is to expose the real constraint. Most programmes are limited by throughput rather than by ideas: not enough traffic per test, tests left running without a decision date, or a queue of hypotheses nobody will choose between. Reporting completion rather than backlog size makes that visible, and backlog size is a measure of intention that makes a team look busy instead of effective.

The bottom five rows are not useless numbers. They are working instruments, and the distinction this table draws is about audience rather than quality: a metric that needs a paragraph of context to interpret correctly will be misinterpreted in a room that has eleven minutes for your item.

How to present these CRO metrics without being ignored

1. Fix the format and never change it. Same three numbers, same order, same definitions, good quarter or bad. The moment the format moves with the result, the reader starts discounting everything.
2. Show a trend, not a point. Four quarters of each number. A single figure invites the question of whether it is good, which you cannot answer; a trend answers it.
3. One sentence of interpretation per number. What it did and why. Not three paragraphs, because the detail belongs in the appendix and will be read by one person.
4. State the assumption behind the money figure every time. In the same words. The first time it feels like a caveat, by the third it is a reason you are trusted.
5. Put one named decision at the end. What you want from the room: a budget, an engineering allocation, a choice between two directions. A report with no ask gets no decision and no memory.

If you do not yet have a defensible baseline to report against, start by getting one. A free CROBenchmark audit scores your store against the dataset above and gives you the funnel-level figures these three numbers are built from, which is a faster start than instrumenting from scratch. Where the ranked findings have to become a continuous programme rather than a quarterly document, Omniconvert Explore is the CRO platform for the testing side, and Omniconvert Commander (formerly Nexus) is the AI for eCommerce growth engine that unifies commerce data and prioritises experiments by True Profit, with you approving what goes live. That True Profit ordering is the same argument as number one above: a ranking by revenue and a ranking by profit disagree, and the board is asking about the second.

One external reference point is worth having to hand when the credibility question comes up, because it will. The Baymard Institute and its long-running usability research is the standard external citation for the kinds of funnel defects these programmes fix, which is useful when somebody asks whether the problems you are testing are real.

Teams often open a board pack with conversion rate and spend the meeting explaining a seasonal dip. The money from shipped tests, the number that would have answered the room, sits on slide nine.

AliveCor used Omniconvert to run a structured A/B testing programme and achieved +21% conversion rate, +5% revenue per visitor, and 94% statistical relevance across their experiments. [Omniconvert, AliveCor case study] Reporting realised results with their assumptions is how a programme like that earns its budget.

The adjacent case is now documented too: benchmark mobile vs desktop performance.

Frequently asked questions

Which CRO metrics should you report to leadership?

Three. The realised money from tests already shipped, stated as annualised profit rather than as a percentage uplift. One leading indicator that moves before the money does, usually revenue per session on the segment you are working. And a learning rate, meaning tests completed and the share that produced a decision either way. Those three answer what a board actually asks: did this pay, is it still working, and are we getting faster.

Why is conversion rate a poor headline metric for a board?

Because it moves for reasons that have nothing to do with your work. A shift in traffic mix, a discount period, a seasonal peak or a change in session counting will all move it more than a year of testing does. Reporting it as your headline means spending half of every meeting explaining why a number you are accountable for went the wrong way for reasons outside your control.

Should you report a test uplift percentage?

Not as the headline. A percentage uplift on a single step, measured on one segment over three weeks, becomes a wildly overstated annual figure when a reader multiplies it against total revenue, and they will. Report the money you believe the change is worth over a year, with the assumption stated, and keep the percentage in the appendix for the people who will check your arithmetic.

What should you never put in a leadership report?

Statistical significance, bounce rate, time on site and the number of hypotheses in the backlog. The first invites a debate about methodology instead of a decision. The middle two move without meaning anything and will be used against you the quarter they go the wrong way. The last is a measure of intention rather than of output, and it makes the programme look busy rather than effective.

How do you report a quarter where nothing won?

As the result it is, with the learning rate carrying the report. A quarter of well-run tests that all failed has told you which four expensive ideas not to build, and that has a value you can state. What destroys credibility is presenting a losing quarter as a win by switching to a metric that happened to rise. Report the same three numbers every time, including when they are bad, because consistency is the only thing that makes the good quarters believable.