Prioritize a conversion backlog by separating confirmed repairs from uncertain opportunities, making each proposal specific, and weighing the evidence against effort and the value of the decision. A score can summarize a discussion, but it should not replace the explanation. The team should be able to say why one item comes first without appealing to a mysterious number.
Start with a buyer problem and an outcome, not a list of page elements to change. “Add social proof,” “make the button brighter,” and “shorten the page” are possible interventions. They do not yet describe a problem or explain what the intervention is expected to accomplish.
Turn ideas into proposals
For each idea, identify the affected audience, the observed obstacle, the proposed mechanism, the commercial outcome, the evidence, and the next decision. A concise proposal might say: “Buyers choosing the large appliance cannot find installation requirements; add verified requirements near variant selection and test whether they can determine eligibility.”
Separate what you observed from what you expect. The observed problem may be supported by task sessions. The expected effect on purchases remains a hypothesis. Avoid a label such as “high confidence” that merges confidence in the problem with confidence in an uplift.
GOV.UK’s planning guidance is useful for turning assumptions into questions. Apply that discipline before assigning priority: an idea with no supporting problem may need research rather than immediate implementation.
Classify the next action
Use different decision paths for different kinds of work. A valid-address rejection, an unsupported promise, and a competing layout hypothesis should not all wait in the same experiment queue.
| Work type | Trigger | Appropriate next action |
|---|---|---|
| Functional repair | Reproduced task failure | Fix, verify, and monitor |
| Accuracy correction | Unsupported or misleading information | Correct with the responsible owner |
| Research question | Important uncertainty with weak evidence | Conduct focused research |
| Presentation improvement | Supported problem and credible remedy | Prototype or release with checks |
| Experiment candidate | Important uncertain effect and feasible design | Run a controlled comparison |
| Commercial policy question | Cost, eligibility, or terms limit fit | Review the underlying policy |
This classification prevents a common bottleneck: insisting that every useful change prove an incremental conversion lift before it can be made. Correctness and task completion have their own acceptance criteria. Equally, a strong design opinion should not silently become a confirmed repair.
Assess reach without inventing affected users
Use the number of people who can encounter the issue, not all site visitors. If an item concerns one variant on one product page, estimate that audience. If an error occurs after selecting a particular delivery region, use that eligible cohort.
Distinguish exposure from harm. Every viewer may see an unclear phrase, but only some need the information it obscures. Report the count you actually have and the gap you cannot estimate. For example: “2,000 monthly eligible visitors; frequency of confusion unknown.” That is more honest than multiplying the traffic by an invented frustration percentage.
A serious problem can deserve priority despite limited reach. A wrong commitment or inaccessible path may block a small audience entirely. Do not let a reach score erase the consequence for those people.
Estimate effort as a range
Ask implementers to identify the work, dependencies, review needs, integration checks, and ongoing maintenance. Copy changes can involve legal or commercial review. A short form adjustment can affect routing and reporting. A new calculator may need product data and long-term ownership.
Use a range when uncertainty is material: one to three days after pricing approval is more informative than “effort: 2.” Include the dependency, since a small implementation can remain blocked for weeks by a unresolved business decision.
Count the cost of learning as well as building. A experiment may consume engineering time, traffic, and analysis capacity. A targeted task study may answer the immediate question faster. Compare the next useful step rather than pricing a full solution prematurely.
Make value and evidence discussable
Create a short assessment with four fields: consequence, reach, evidence, and effort. For consequence, describe what fails and why it matters. For evidence, list the direct observations and what they do not establish. For effort, name dependencies and reversibility.
If you use scoring, keep it coarse and expose the inputs. Do not multiply several subjective estimates to create an apparently precise revenue forecast. Scores are private planning aids. A score of 82 does not mean a idea has an 82% chance of working.
For experimentation candidates, include decision value and feasibility. Microsoft’s pre-experiment guidance supports specifying hypotheses and metrics before running a comparison. A backlog item without a meaningful success measure is not ready simply because it ranks highly.
Worked example: choosing between three items
Synthetic example. A retailer considers a hero-image test, a postcode-validation repair, and a pricing calculator. The postcode problem is reproducible for a supported region. The image idea comes from internal preference. The calculator could clarify a real cost question, but depends on unconfirmed installation prices.
The team repairs postcode validation first and verifies successful completion. It turns the hero proposal into a research question about product understanding rather than immediately testing aesthetics. It asks the pricing owner to resolve installation rules and runs a prototype task to see which estimate buyers need.
The decision is defensible without a combined score: one item restores a legitimate task, one lacks a demonstrated problem, and one depends on a commercial input. If the image proposal later gains evidence and feasible traffic, it can become a valid experiment candidate.
Keep a small active queue
Choose only the work the team can advance. Give each item a owner, a next action, and a date to review what was learned. Maintain a parking area for ideas that lack evidence or depend on a future decision, with a clear condition for reconsideration.
After release or testing, record the outcome. Did the task become possible? Did understanding improve? Did the comparison resolve the commercial decision? A change that shipped is not automatically a conversion success. Returning the result to the backlog prevents the same weak idea from repeatedly reappearing with a new title.
When two items remain close, compare the reversibility and learning value of their next steps. A small prototype can expose a mistaken assumption before expensive implementation. A repair with clear acceptance criteria can remove a known obstacle immediately. Record the tie-breaking reason. This makes prioritization reviewable and prevents a temporary resource constraint from masquerading as evidence that another problem is unimportant.
Worksheet: CRO proposal and priority
| Field | Complete before scheduling |
|---|---|
| Audience and observed problem | ______________________________ |
| Source evidence and limitations | ______________________________ |
| Consequence of leaving it unresolved | ______________________________ |
| Eligible reach and unknown frequency | ______________________________ |
| Proposed mechanism | ______________________________ |
| Work type and next action | ______________________________ |
| Effort range, owner, and dependency | ______________________________ |
| Success or acceptance criterion | ______________________________ |
| Reason this comes before another item | ______________________________ |
| Review date and decision after learning | ______________________________ |
Revisit priorities when evidence changes
A backlog is a current decision aid, not a permanent ranking. Traffic mix, product availability, pricing policy, and new observations can change what matters. Reassess when a new fact changes consequence, reach, feasibility, or effort. Keep the explanation attached to the item so the team can update its judgment instead of merely rearranging a spreadsheet.