How does roadmapping change when your dependency is a vendor's release schedule?

Turnstile Live sells ticketing and event software, and is building live, multilingual captioning for in-person events. Halden Ruiz leads the accessibility product line, and the roadmap in front of him depends on a speech-recognition vendor's own release schedule for a model that hasn't shipped yet.

How to answer this, stage by stage

Nobody is scoring whether you can guess a vendor's real ship date. They're scoring whether your roadmap survives them being late.

Let's learn

The tablet at the check-in desk is bolted to the wall, and for years it's only ever done one thing at Turnstile-powered events: print a badge.

Live captioning would add a second thing entirely: real-time, multilingual subtitles projected at conferences and concerts, read by anyone in the room on their own phone. Before this project, event organizers who wanted captioning at all hired a human stenographer, at real cost, booked weeks in advance, accurate but limited to whatever languages that one person spoke. Turnstile's baseline in-house model can now caption English live, cheaply, for any event with no advance booking. The bigger promise, live captioning in eleven languages at once, depends entirely on a vendor's still-unreleased multilingual speech model.

Two tiers, two very different dependencies. Only one of them is actually Turnstile's to schedule.

Here's the turn: Turnstile's roadmap treated the vendor's own preview timeline as if it were a date on Turnstile's own calendar. It committed a public launch to six marquee customer events, each one tied to the vendor's promised general-availability quarter. Every one of those six events happened. None of them got the multilingual tier, because the vendor's promised quarter slid by exactly one quarter, every single time anyone checked.

At its worst, a roadmap that treats a vendor's promise as a commitment can spend a full year publicly missing dates it never actually controlled in the first place.

What I would leave alone: I wouldn't touch the baseline English captioning tier's roadmap at all. It runs entirely on Turnstile's own model, ships on Turnstile's own schedule, and has no vendor dependency to hedge against in the first place.

The lesson: a vendor's release date is a number on someone else's roadmap, not yours. Roadmapping around it means ranking what you actually control first, not hoping their calendar behaves like a promise.

Now here is the same thing as a story

The short version above is what you'd say defending a missed multilingual launch to a customer. Read this one for how a conference host's own voice quietly started changing.

Perrin Okafor hosts the opening keynote at three or four Turnstile-powered conferences a year, and has done it long enough to read a room without looking at notes.

For Perrin's first event using Turnstile's baseline captioning, everything ran smoothly: natural, conversational keynote delivery, captions lagging only slightly, audience members glancing at their phones and following along easily.

One of these decisions can be quietly adjusted next sprint. The other one is already printed on a customer's event program.

For the second event, marketing had already promised the client multilingual captioning, based on the vendor's assurance that general availability was one quarter out. It wasn't ready. Turnstile fell back to the baseline English-only model for that event too, but this time Perrin had learned something from watching the captions at the first event: certain phrasings, quick asides, people's names said quickly, jokes with a pause built in, occasionally garbled or dropped a beat behind.

Turnstile could plan around its own model any day it chose. It could only ever hope around the vendor's.

So for the second event, without anyone asking him to, Perrin started rephrasing his own keynote as he spoke: simpler sentence structures, names repeated slowly, jokes flattened into plainer statements, pauses added before anything that felt likely to trip up the captions. He wasn't talking to his audience anymore. He was talking to the model.

Nobody had told Perrin to do this. He'd simply learned, the way anyone learns a tool's failure shape, and adjusted himself around it rather than around a product design that adjusted around him.

None of these four checks needed the vendor's cooperation. All four were sitting there, unwatched, the whole time.

The public date is the one item in the top right, exactly where a roadmap shouldn't put anything it can't take back.

What I'd tell myself, hearing how Perrin had quietly rebuilt his own keynote around a captioning model that wasn't even the one running that day: the real cost was never the missed date. It was a skilled person changing how he talked to a room, to protect a promise that was never his to keep in the first place.

ORDER, so the vendor's calendar doesn't become yoursNot a plea for patience. ORDER is what tells you which piece of the roadmap you actually control.

The recap, one line per letter: outcome is captioning customers can count on, reversibility is the public date being the one thing that can't be walked back, dependency is the baseline tier needing nothing from the vendor while the premium tier needs everything, evidence is a monthly check on the vendor's own slip signals, and rank is baseline first, vendor watch second, premium tier's public date last, only once the vendor actually ships.

And if you want to be sure it really works, try it somewhere elseSame five letters, a textile mill instead of a ticketing platform. Different flip family entirely, the same borrowed calendar.

Fendrick Textiles committed its fabric-inspection roadmap to a vendor's promised computer-vision model for detecting weave defects, reassigning two floor inspectors to other lines the same quarter the vendor demoed a working prototype. Mapped onto ORDER: outcome is catching defects before a bolt of fabric ships to a customer. Reversibility is the same shape, the reassignment of trained inspectors is far harder to undo than a wait-and-see check-in would have been. Dependency is the inspection line waiting entirely on a vendor model still mid-development, with no in-house fallback drafted. Evidence would have been the vendor's own release notes, which had already slipped twice before the mill's commitment. Rank should have kept the inspectors in place until the vendor reached general availability, not a working demo. The flip here is different: concealment, not input. Sylvie Marchetti, the mill's quality lead, stopped mentioning in her weekly reports that the new system was still vendor-dependent and unproven, worried it would look like she'd championed a bad bet, and by the time defects started slipping through, nobody above her even knew there was a dependency to question.

The same missing check that would have caught Turnstile's slipping vendor would have caught Fendrick's too.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "ship what you control now, watch the vendor cheaply, and keep any public date off their calendar until they've actually shipped," and stop.
Cost: no time to build a formal vendor-tracking process this quarter. Say so honestly, and start with the cheapest version, one person checking the vendor's public changelog once a month.
The vendor ships early, for real: if the model genuinely lands ahead of schedule, the honest move is to move the public date up and say so, not treat every vendor promise as permanently untrustworthy out of habit.

Where people run it wrong.
They let a vendor's demo or preview access feel like a shipped product, and plan a public date around it.
They reassign their own people away from a manual fallback before the vendor has actually reached general availability.
They keep no visible internal record of the vendor's own slip pattern, so each new promise gets evaluated fresh instead of against a track record.

How to use it live. The moment someone asks about a vendor dependency, ask yourself: what can we ship without ever needing their calendar to cooperate? Rank that first, and let everything vendor-dependent stay provisional until it's real.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0