Skip to content
Provenly
Challenge · AI Product Manager

Turn 'Make Search Better' Into Something Buildable

One vague request, one Slack thread of contradictions. Produce a spec a team can build without you. It takes 60 minutes, the full rubric is published below before you start, and AI tools are allowed.

60 min
Time cap
L3
Target level
3
Skills assessed
5
Scored criteria

The brief

The request, verbatim, from the VP of Sales: "Search is bad. Customers keep complaining they can't find things. Can we make it smarter — maybe use AI? Would be great to have by end of month."

What else you know. Support tickets mentioning search: 31 in the last quarter. Reading them, 18 are about search returning nothing for product codes with hyphens, 7 are about results being in a confusing order, 4 are about search not covering archived items, and 2 are unrelated.

From the Slack thread. Engineering: "our search doesn't tokenise hyphens, that's a two-day fix." Design: "the results page has no sorting, that's the real problem." The VP: "I still think we need AI, competitors have it." A customer success manager: "the biggest account just wants archived orders searchable."

Your task. Write the spec. Decide what is actually in scope, be explicit about what is not, handle the edge cases, and define how anyone will know it worked.

You will be scored hardest on the parts specs usually omit: non-goals, edge cases, and testable acceptance criteria.

What you hand in

A spec containing: problem statement, in-scope, explicit non-goals, edge cases, acceptance criteria, and a short note on how you would handle the AI request.

On AI assistance

Any tools. An AI assistant will happily produce a well-formatted spec that includes everything and excludes nothing — the non-goals section is where that fails, and it is weighted heavily.

How is this challenge scored?

Every criterion, with its weight and the skill it produces evidence for. Nothing else is measured — not writing quality, not grammar, not length. You see this before you start, and there are no hidden criteria.

Explicit non-goals

States what is deliberately not being built and why. A spec with no non-goals section scores 0 here regardless of quality elsewhere.

Edge cases are handled up front

Addresses cases the request did not mention — empty results, permissions on archived items, partial matches, tokenisation of other special characters beyond hyphens.

Acceptance criteria are testable

Someone else could verify each criterion without asking you what you meant. 'Search should feel fast' scores 0; a stated threshold scores full.

Follows the evidence, not the request
Product Discovery · weight 1.3

Notices that 18 of 31 tickets point at a two-day tokenisation fix, and that nothing in the evidence supports the AI request. Building what was asked for without checking scores 1.

Handles the AI request without dismissing it

Addresses the VP's ask directly — deferred with a reason, scoped as a later phase, or included with justification. Silently omitting it scores 0.

Which skills does this prove?

Questions people ask

How long should the spec be?
Most strong submissions are 600–1,200 words. Length is not scored; completeness on edge cases and non-goals is.
Can I say no to the AI request?
Yes, if you justify it against the evidence. What scores 0 is quietly leaving it out of the document and hoping nobody notices.
Does the template matter?
No. Use any format. Reviewers are instructed to ignore structure and formatting entirely.
Other challenges