We had built a conversational lead-capture widget around a simple practical job. In three taps, a visitor could identify the service they needed, their likely timeline and their budget range. At the end, the visitor could continue on WhatsApp, send an email or ask for an exact quote.
When an AI assistant was asked to critique the flow, it responded with familiar conversion-rate optimisation advice. The recommendation sounded confident: reduce the available actions, simplify the decision and focus attention on one preferred route.
Perhaps that advice was right. But the confidence of the answer was not evidence that it was right for this business, this audience or this moment in the customer journey.
How useful advice turns into doctrine
Useful design principles exist. Removing unnecessary effort can help. Clear visual hierarchy makes the next action easier to understand. Redundant choices deserve scrutiny. A principle can be useful without being universally correct.
The problem begins when a sensible observation is packaged, repeated and detached from the conditions that made it useful.
Tools need defaults. Agencies need repeatable processes. Templates need standard structures. AI assistants learn from the accumulated language and reproduce its consensus fluently. That does not make the advice dishonest. It means the advice reflects the visibility and constraints of whoever—or whatever—delivers it.
A system naturally optimises around the outcomes it can observe. The issue is not necessarily bad intent. It is visibility, incentives and what each system was designed to measure.
Context changes the recommendation
Three ways to continue may be unnecessary on a low-consideration purchase page. They may be appropriate after qualification for a service business. WhatsApp may suit one buyer, email another and an exact quote request a third. If each route serves a distinct need, preserves context and reaches a team able to respond, choice may support the journey rather than obstruct it.
- What stage is the customer at?
- What is each option for?
- Do different buyers need different routes?
- Does each route preserve context?
- Can the business respond well through each route?
- What commercial outcome are we trying to improve?
- It duplicates another action
- It creates unnecessary effort
- It serves an internal preference rather than a customer need
- The business cannot support the route properly
- It serves a genuinely different customer need
- It represents a meaningful level of intent
- It offers an appropriate way to continue
- The business can preserve context and respond properly
These are decision principles, not universal rules.
is not the objective.
Start with the decision the business needs to make
A more useful CRO question is: What business or customer outcome are we actually trying to improve? If the aim is enquiry quality, define quality before changing the interface. If it is lower sales effort, understand which enquiries consume time without progressing. If it is revenue, connect the test to later outcomes instead of ending at form completion.
What business or customer outcome are we trying to improve?
What do we believe should change—and why?
What would support or challenge the hypothesis?
What would make us keep, revise or reverse the change?
Test the recommendation, not the reputation
Good advice deserves a fair test, not blind obedience or reflexive rejection. Its source does not remove the need to validate it. Validation may be qualitative, operational or quantitative. An A/B test is useful when appropriate and feasible; it is not the only legitimate way to learn.
Before changing the interface
- Which route do people choose?
- What are they trying to accomplish?
- Do routes represent different intent?
- What questions appear after the click?
- What does each conversion represent?
- Are unlike actions combined?
- Does source context survive?
- Can later outcomes be connected?
- Which routes can the team support?
- Where does context disappear?
- Which route creates avoidable work?
- Is interface preference an operational constraint?
Sometimes the apparent CRO problem is a workflow problem.
What the AI critique was actually useful for
The critique surfaced an assumption that had not been stated: perhaps the final choice asked the visitor to make an unnecessary decision. Once visible, that assumption could be examined against the purpose of the flow, the follow-up routes and later commercial outcomes.
This is a productive role for AI in optimisation. It can surface assumptions, generate alternatives and identify familiar risks quickly. The output becomes stronger when treated as material for investigation rather than a verdict.
Optimisation should serve the business
CRO becomes more useful when it stops treating the interface as the whole system. The page, form or widget is one part of a journey that includes traffic quality, customer intent, follow-up, sales capacity and the eventual commercial outcome.
Before accepting a CRO recommendation, ask:
- What outcome are we trying to improve?
- What evidence supports this recommendation?
- What business context could change the answer?
- What would we measure after making the change?
- What evidence would make us reverse it?
References & further reading
- Journal of Consumer Research · 2010Scheibehenne, Greifeneder & Todd — Can There Ever Be Too Many Options?Choice-overload effects varied substantially across studies.
- Journal of Consumer Psychology · 2015Chernev, Böckenholt & Goodman — Choice OverloadContextual moderators help explain when larger choice sets create difficulty.