How to Prioritise Fixes When You Are Below Benchmark
Fix exposure is defined as the number of sessions or orders that passed through the surface a proposed change would alter, and it is the first thing to sort by when you prioritise fixes below benchmark. A benchmark comparison hands you a list and no order, and the order is where most of the value sits: rank on exposure, then reversibility, then confidence, then effort, in that sequence. Around 70% of carts are abandoned across retail [Baymard Institute, 2025], which is why checkout surfaces usually top an exposure-ranked list. This guide sets out the four criteria, the two traps that invert them, and how to work the list once it is sorted.
- Sort by exposure first: a correct fix on a quiet surface returns nothing.
- Let reversible changes go before irreversible ones at equal exposure.
- Record evidence as measured, observed or believed, and never as a score.
- Estimate effort last, because estimating it first reorders everything.
- Finish one item per surface before starting the next on that surface.
- Re-rank monthly, since exposure moves with season, traffic mix and range.
Omniconvert has measured conversion behaviour across the CROBenchmark dataset of 7,000+ websites in 15+ industries, against 248+ audit criteria, over 13 years in eCommerce [CROBenchmark Report 2026, Omniconvert]. The common failure after a benchmark comparison is not a shortage of findings. It is a backlog of forty items with no agreed order, which quietly becomes a backlog sorted by whoever argued hardest.
How to prioritise fixes below benchmark: order beats list
The list itself is usually not controversial. Most teams looking at the same store agree on roughly what is wrong. What they disagree about is sequence, and sequence is what decides whether the quarter produces a measurable improvement or a set of changes nobody can attribute.
There is also a compounding effect. Early wins on high-exposure surfaces fund the credibility for the slower structural work later, while a quarter spent on a beautifully executed fix to a page three hundred people saw produces nothing to point at.
So treat ranking as the deliverable rather than as admin. The diagnostic that produced the list is covered in the companion piece on what to do when your conversion rate is below benchmark; this is what happens next.
The four criteria, in order
Exposure comes first because it is the only criterion that can zero out the others. A change that is correct, cheap and certain is still worth nothing on a surface nobody reaches. Count the real sessions and orders through the surface over a full recent month, not a percentage.
Reversibility comes second because it prices the risk. Most of what makes a change frightening is the cost of being wrong, and a change you can undo in an hour deserves to go before an equally promising one that needs a migration to reverse.
Confidence comes third and is recorded rather than scored. Evidence grade is defined as whether a finding rests on a measurement, on a direct observation of real users, or on somebody's belief, and keeping those three labels apart stops a strong opinion being averaged into a strong case.
Effort comes last, and this is the one teams get wrong most often. Estimating effort first feels responsible and silently reorders the list around what is convenient for the people estimating.
What an ordered backlog looks like
| Criterion | What it answers | How to record it |
|---|---|---|
| Exposure | How many people meet this surface | Sessions and orders, last full month |
| Reversibility | What undoing it costs | Hours, days or a migration |
| Confidence | What the finding rests on | Measured, observed or believed |
| Effort | What building it costs | A range, estimated last |
Resist the urge to multiply these into a single priority score. A score of 7.4 cannot be argued with, which sounds like an advantage and is the opposite: the argument is where the thinking happens, and a composite moves it somewhere nobody can see it.
The specific way this goes wrong: a team builds a scoring spreadsheet, weights the four criteria, and discovers that the weighting itself becomes the argument. Three weeks later the backlog is ordered by a formula nobody fully agrees with and nobody wants to reopen, and the first item is a navigation change that scored well because it was cheap. Keep the four columns visible and sort by hand.
The two traps that invert the order
The first trap is the annoyance list. Everybody who works on a site has a private register of things that irritate them, and those things are real, small and usually on surfaces that barely matter. An unranked backlog defaults to this list because the irritations are the easiest items to describe.
The second is effort-first sorting, which shows up as a sprint of quick wins. Quick wins are genuinely useful when they happen to sit on high-exposure surfaces, and the word "quick" is doing all the work in that sentence. Sorting by effort means the surfaces with the most traffic, which are usually the most complicated, go last.
Unexpected cost and unclear information remain among the most persistent reasons people abandon a purchase, at around 70% average cart abandonment in the research Baymard Institute has run for years [Baymard Institute, 2025]. That is an exposure argument as much as a usability one: the checkout is the surface every order passes through, so it tops a correctly ordered list in most stores and sits in the middle of an effort-ordered one.
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] The relevant detail is the last figure rather than the first: that programme reached statistical relevance because it ran one change at a time on surfaces with enough traffic to resolve, which is the same discipline this ranking is for.
Working the list once it is sorted
Take the top item and complete it end to end, including the measurement. Partial completion is the main way an ordered backlog reverts to an unordered one: three items half-built across two surfaces leaves nothing measurable and no clear next step.
Where the change is big enough and the surface busy enough, prove it rather than assume it. Omniconvert Explore is the testing platform for that, and Explore is also where a checkout change should be validated, because the checkout is the one surface where the whole store's traffic pools.
Re-rank monthly. Exposure is not a fixed property: a seasonal range, a channel mix change or a new landing campaign moves which surfaces people actually meet, and a backlog ordered in January is describing January. The roadmap view of the same question is in how benchmarks should shape your test roadmap, and the level question is in a good conversion rate.
Shopify's merchant guidance has pushed consistent product and checkout data across sales channels for years, and that consistency work is a good example of an item that ranks badly on effort and extremely well on exposure. It is exactly the kind of thing an effort-sorted list never reaches.
What to do this week
- Add an exposure column. Real sessions and orders through each affected surface, last full month. Sort by it before discussing anything else.
- Label every item as measured, observed or believed. No scores. The labels will change the order on their own.
- Pick the top item and finish it. Including the measurement. Then re-rank before starting the next. If you want the audit methodology that generated the list, the step-by-step method is at audit it.
Beyond the ranking itself, the longer-term move is to stop producing unranked lists in the first place. Omniconvert Commander (formerly Nexus) is an AI for eCommerce growth engine that unifies commerce data, prioritises experiments by True Profit, and generates campaigns and creative you approve before they go live, which is the same ordering problem solved upstream of the spreadsheet.
For a different angle on the same problem, see read a conversion report without being misled.
Frequently asked questions
How should you prioritise fixes when you are below benchmark?
Rank on exposure first, then reversibility, then confidence, then effort. Exposure is how many sessions or orders passed through the surface the fix would change, and it dominates because a correct fix on a page nobody reaches returns nothing. Effort comes last deliberately, since estimating it first reorders the whole list around convenience.
Why not start with the biggest gap against the benchmark?
Because the size of a gap says nothing about how many people it affects or how easily it can be closed. A large gap on a metric driven by one small segment is worth less than a small gap on a surface every order passes through, and benchmark tables do not carry that information.
What counts as exposure exactly?
Sessions through the page or step the change would alter, over a full recent month, and orders through it where the change affects a purchase decision. Use the real figure rather than a share, because a percentage hides the difference between a busy store and a quiet one.
Should irreversible changes ever go first?
Only when the evidence is a measurement rather than an opinion and the exposure is high enough to justify the risk. Most things that look irreversible, such as a navigation change or a pricing presentation, can be made reversible with a little more build effort, and that is usually the better trade.
How do we keep opinion out of the ranking?
Record the evidence type beside every item: measured, observed or believed. Do not convert it into a number. The moment confidence becomes a score it gets averaged with effort and exposure, and a strongly held opinion starts outranking a weak measurement.
How many items should be in progress at once?
One, on any given surface. Two changes to the same page in the same fortnight cannot be told apart afterwards, so the second one costs you the ability to learn from the first. Parallel work is fine when the surfaces do not overlap and the measurement can be kept separate.
Sort the list before you argue about it
Open the findings list and add one column: the real number of sessions and orders through the surface each item would change, last full month. Sort by that column. The conversation about what to do first usually ends there, because most of the disagreement was about things nobody had counted.
Then label each item measured, observed or believed, estimate effort last, and take the top one all the way through to a measurement before starting another on the same surface. One finished item beats four in progress, every quarter.