How do you prioritize data investment against feature investment on a roadmap?

Say we build a tool that reads an X-ray and flags what a vet should look at twice. Alder Vale Veterinary Partners runs that tool across fourteen clinics, and Nadia Corbett, its VP of Product, is the one who has to decide, this quarter, whether the next dollar goes to a new imaging feature or to fixing a data gap nobody had filed a single complaint about.

How to answer this, stage by stage

Nobody is scoring whether you know that both data and features matter. They're scoring whether you can commit to a pick, then show the asymmetry that actually justifies it.

Let's learn

The four letters, held up as one page. Cost asymmetry is the step this question is really testing.

Alder Vale's imaging assistant flags likely conditions from X-rays across fourteen clinics and about ninety vets, handling roughly six hundred cases a week across forty five breed and species categories. Before it existed, a vet reviewing an unusual case took about twelve minutes cross-referencing textbooks or calling a specialist. With the tool, a flag appears in seconds, for any breed.

Two kinds of shortfall. One shows up on a list. The other shows up nowhere at all, until someone checks.

Here's the turn: the extra confidently-wrong flags on rare breeds were never the whole problem by themselves. Say plainly what mattered instead: vets handling those breeds didn't file a complaint, request a fix, or ask for anything. They simply stopped opening the tool for those specific cases and went back to doing it the old way.

What I would leave alone: common-breed usage, at 81 percent, was healthy and didn't need investment redirected toward it. The problem was never company-wide, it was concentrated in six specific breed combinations.

The lesson: a missing feature and a data gap don't cost you the same way. One sits on a list where you can see it. The other quietly narrows exactly who still trusts the product, in exactly the cases where trust matters most.

Now here is the same thing as a story

The short version above is what you'd say defending a budget line in a planning meeting. Read this one for how the gap actually widened for six months before anyone thought to look for it.

Nadia Corbett has run product at Alder Vale for five years, and can usually tell from a roadmap review which proposal is going to win funding before anyone votes.

Four questions the roadmap review never asked, because one blended usage number always seemed like enough.

Every quarter, the roadmap review compared a proposed feature's projected revenue against a data investment's projected cost, and the feature almost always won, since it came with a clean number attached, a bundled ultrasound package projected at roughly $340,000 a year, against an $85,000 ask to label two thousand more X-rays for six under-represented breed and condition combinations.

Nobody decided, on any single day, to stop trusting the tool for these six breeds. It happened one quiet case at a time.

Six months earlier, usage on those six breed combinations sat at 58 percent, not far below the common-breed average. Nothing dramatic happened. A handful of vets, independently, hit a confidently wrong flag on a rare-breed case, once each, and quietly stopped opening the tool for that specific breed afterward. No complaint got filed, because nothing about the workflow required one, a vet could simply go back to reading the X-ray the old way.

By month six, usage on those six combinations had fallen to 22 percent, and it took a public story about a competing veterinary AI vendor misdiagnosing a similarly rare condition, and the trade-press backlash that followed, before anyone at Alder Vale thought to check their own usage logs broken out by breed.

Both problems sat on the same roadmap. Only one of them was ever going to show up if nobody went looking for it.

When the shared budget line was first set up, someone said, "let's not carve out a separate data budget, we'll fund whatever earns its case each quarter," and it sounded reasonable, since forcing every investment to prove itself looked like good discipline at the time.

Rerun the same six months with a protected data budget line and usage tracked by breed from day one: the six under-represented combinations get their $85,000 in the first quarter, before usage ever falls below 50 percent. The ultrasound feature waits one more quarter, a visible, absorbed delay vets can see on a roadmap update. No competitor's scandal is needed to prompt an audit, because the audit was already the plan.

What I'd tell myself, seeing that 22 percent number for the first time: the tool didn't get less accurate company-wide, not once. It just quietly stopped being trusted in the six places we most needed it to be.

PICK, the pick that survived its own asymmetryNot a script for always choosing data over features. PICK is what tells you exactly when the asymmetry actually favors one over the other.

The recap, one line per letter: position is data investment first this quarter, impact is a visible wait against a silent trust collapse, cost asymmetry is the hidden, expensive gap living in the highest-value segment, and kill criteria is proven feature-driven churn or a rarity explanation that would flip the pick back.

And if you want to be sure it really works, try it somewhere elseSame four letters, a ticketing platform instead of a veterinary practice. Different flip family entirely, the same asymmetry doing the real work.

Dario Feskin leads trust and safety at Vantage Gate Ticketing, where a fraud-detection model flags suspicious purchases before tickets ship. Mapped onto PICK: position is investing in more labeled examples of a new scalper-bot obfuscation pattern this quarter, ahead of a customer-facing "verified fan" badge feature marketing has requested. Impact says the missing badge is felt by fans who ask for it and keep buying tickets anyway, while the data gap is felt by trust and safety analysts quietly absorbing more manual review time on a pattern the model handles worst. Cost asymmetry says the badge's absence is visible, on a support forum thread; the data gap is invisible, buried inside analyst workload nobody was tracking as a roadmap signal. Kill criteria says if the new pattern's fraud volume stays proven small relative to total fraud, the badge could still win. The flip here is substitution, not abandonment: as overall fraud volume grew and review time got scarcer, analysts rationed their manual attention toward exactly the rare pattern the model handled worst, since that's where they felt the stakes were highest, which meant the measured "catch rate" on that pattern actually looked fine on paper, propped up entirely by human effort the model was quietly failing to replace.

A different flip entirely: not vets quietly walking away, but analysts quietly propping up the model's weakest spot by hand.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "invest in data first wherever the gap is hidden and hitting your highest-stakes segment, features can wait, silent trust can't be won back the same way," and stop.
Cost: there's no time to build breed-level or pattern-level usage tracking before the next roadmap review. Say so honestly, and ask analysts or vets directly which specific cases they've quietly stopped trusting, as a cheap first signal.
The feature turns out to matter more than assumed: if real churn data ties directly to the missing feature, that's a legitimate reason to flip the pick, not a failure of the framework.

Where people run it wrong.
They let features win by default because they come with a cleaner, more attributable revenue number.
They track one blended usage or accuracy metric and never break it out by the segment most likely to be quietly struggling.
They assume "no complaints" means "no problem," when a silent workaround or a silent abandonment produces exactly that signal.

How to use it live. The moment an interviewer asks you to prioritize data against features, ask yourself which gap is visible and absorbed, and which one is invisible and already costing you trust somewhere specific. The pick follows from that asymmetry, not from which line item sounds more exciting on a roadmap slide.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0