How do you communicate a roadmap that explicitly contains unknowns?

A roadmap slide can be honest inside the building and dishonest the moment it leaves it, with nobody changing a single word. Haldric builds an AI tool that triages and auto-resolves customer support tickets. Yewande Coker is Head of Product, and the question in front of her is how to talk about a roadmap that is, on purpose, still full of things nobody can promise yet.

How to answer this, stage by stage

Nobody is scoring whether you can write a caveat. They're scoring whether you'll notice the caveat has to survive being copied into someone else's document.

Let's learn

Here is what happens when a roadmap's honesty depends on a font style nobody wrote down as a rule.

Before Haldric's AI triage tool, a support team read and routed every incoming ticket by hand, about three minutes each. With Haldric, a model reads the ticket, drafts a response, and for common issues, resolves it without a human touching it at all. The roadmap Yewande's team keeps promises this capability will keep expanding: more ticket types, higher resolution rates, quarter over quarter.

Everyone in the building can read the nuance in this slide. Nobody outside it can.

Here's the turn: the roadmap was never dishonest on paper. Every internal version correctly showed which lines were sure things and which were bets, using a simple visual convention, light grey italics for concept-stage items. The problem started the day that convention quietly disappeared in a design refresh nobody flagged as a risk, because it looked, to the designer who made it, like a cosmetic cleanup.

At its worst, a slide that reads as calibrated inside the building can read as a guarantee the moment it's copied into an email, and the customer with the least legal muscle ends up carrying the gap between the two.

What I would leave alone: the redesign's visual polish everywhere else. The old deck's inconsistent fonts and spacing genuinely looked unprofessional, and fixing that had nothing to do with the caveat that got lost alongside it.

The lesson: a caveat that lives in a font choice, a norm, or one person's memory isn't a safeguard. It's a habit wearing a safeguard's clothes, and habits don't survive a redesign, a new hire, or a slide getting copied into someone else's deck.

Now here is the same thing as a story

The short version above is what you'd say defending a contract dispute to legal. Read this one for how a design refresh, done for entirely good reasons, quietly took away the one signal a salesperson relied on.

Vikram Aggarwal has closed enterprise deals for Haldric for three years, and he's good at it honestly: he reads the technical roadmap closely enough to explain it better than most account execs, and customers trust him because he doesn't oversell.

For his first two years, every time Vikram walked a prospect through the roadmap, he'd pause at the light-grey italic lines and say some version of the same thing: "that one's still cooking, don't hold us to a date on it." Nobody told him to say it. He just always had, because the slide made it obvious which lines needed the caveat.

Same words, same numbers. Only one version of the slide ever leaves the building.

Then Haldric brought on a product designer to professionalize the roadmap deck ahead of a board meeting. She unified every line to the same clean sans-serif, the same weight, the same charcoal color, a genuine improvement by any normal design standard. Nobody in the room reviewing her work was thinking about italics as a safety feature. Nobody had ever called it that.

Vikram kept giving his caveat for a few more weeks out of habit. Then, gradually, without any single day he could point to, he stopped. The slide no longer prompted him. Every line looked equally confident, equally finished, equally real. He wasn't being careless. He was reading the document the way it was now designed to be read.

The gap in the middle used to be filled by one sentence Vikram said out loud, unprompted, for two years.

Six months after the redesign, during a renewal negotiation, Vikram quoted the roadmap's "ninety percent auto-resolution by Q3" line to Oskander Freight, a smaller logistics customer with no in-house legal team reviewing the deal. Their operations lead wrote it into the signed contract almost verbatim, as a delivery commitment tied to a support-staffing reduction on their end.

No single day explains this drift. That's exactly what makes it dangerous: nobody could see it happening in order to stop it.

Vantridge Logistics, a much larger customer negotiating the same renewal season, had lawyers who caught the vague ninety percent figure and pushed it down to a contractually safer seventy before signing. Oskander had nobody doing that work for them. When Q3 arrived and Haldric's model hit sixty-one percent auto-resolution, Vantridge's number held. Oskander's didn't, and their support costs, staffed down in anticipation of the promised automation, spiked for two quarters straight.

GUARD, the mark that has to survive leaving the buildingNot a communication tip. GUARD is what tells you exactly which design decision let a bet travel as a fact.

The recap, one line per letter: groups is the sales team who can ask for the real odds against customers who only see the slide, unequal is customers with legal review absorbing far less risk than those without, ability to contest is a customer having no way to know a number was ever a bet, reduce is a mark that survives outside the deck plus a hard rule against unmarked contract terms, and detect is a standing quarterly contract audit that catches drift before a renewal dispute does.

And if you want to be sure it really works, try it somewhere elseSame five letters, a farming cooperative's crop-disease alerts instead of a support-ticket tool. Different flip family entirely, the same unmarked bet.

Cressfield Cooperative runs a phone-photo disease-detection model for member farms, with a roadmap line promising expanded coverage for rare fungal infections "as data volume grows." Mapped onto GUARD: groups are Cressfield's agronomy staff, who know exactly how thin the rare-fungus training data still is, and member farmers, who only hear "the app will catch more diseases soon." Unequal lands hardest on farms exporting to markets with strict phytosanitary rules, where a missed rare-disease flag risks an entire shipment being rejected, versus farms selling only domestically, where the same miss costs far less. Ability to contest is the same gap: a farmer has no way to know the app's confidence on an unfamiliar fungus is still low. Reduce is the same fix, adapted: a visible low-confidence flag on any diagnosis outside the well-covered categories, with a clear "send to a human agronomist" path, not a silent pass. Detect is tracking false-confidence complaints by crop type and export destination. The flip here is substitution, not over-trust: Teodor Vasquez, a field agent, used to photograph every unusual leaf he found over the radio to his co-op's central line; once data costs rose and photo uploads got capped per visit, he started saving his uploads for cases he already suspected were serious, exactly the cases the model most needed new examples of to actually improve.

The same four parts fix the gap in a support contract and a crop shipment alike: know which quadrant you're actually in before you promise anything.

The same three requirements, whether the contract is a support SLA or a crop insurance claim.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "build the caveat into the artifact, not the presenter's memory, and bar unmarked bets from contracts," and stop.
Cost: no time to redesign the whole deck template this quarter. Say so honestly, and start with one non-negotiable rule, a plain-text tag on every unproven line, cheaper than a redesign and enforceable immediately.
The model gets better, for real: if auto-resolution genuinely clears ninety percent next quarter, that's the bet paying off, and the honest move is promoting the line to a real commitment with a real date, not quietly leaving the mark on out of habit.

Where people run it wrong.
They treat a caveat as a communication skill some people are better at than others, instead of a property the document itself needs to guarantee.
They assume a design refresh is purely cosmetic and skip reviewing it for anything load-bearing that might quietly disappear.
They wait for a contract dispute to reveal the gap instead of auditing signed terms against the roadmap on a schedule.

How to use it live. The moment someone asks how to communicate uncertainty on a roadmap, ask yourself: if this slide got copied into a stranger's inbox with no context, would the uncertainty survive the trip? If not, you're relying on a person, not a document.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0