How do you run competitive analysis in a market where the landscape changes monthly?

No company comes attached to this one, so pick one and commit: Corvid Vision, a maker of SentryEye, an AI vision system that inspects machined metal parts for cracks and burrs on a factory line. Dorotea Kessling runs competitive intelligence there, in a category where a rival ships something new most weeks.

How to answer this, stage by stage

Nobody is scoring whether you can list five sources to check. They're scoring whether your system survives the month it actually matters.

Let's learn

Every month, Dorotea Kessling used to open forty-six links by hand. Blog posts, job listings, patent filings, app store reviews, one browser tab at a time, across six rival companies selling AI vision systems like Corvid Vision's own SentryEye. It ate most of her first week of the month, and she was good at it. She could tell a real shift in a rival's technology from a rewritten homepage in about ten seconds.

Four different kinds of source, read one tab at a time, every single month.

Then the team built RivalScan, an internal tool that reads those same forty-six sources and writes one plain page a month: what changed, in sentences anyone could follow. For the first five months, Dorotea still opened about one link in five to check the digest against the real thing. It was right every time.

Here's the turn: being right every time is exactly what taught her to stop checking. The habit didn't drop all at once. By month three she'd stopped checking a random fifth and started checking only the lines that sounded, in the digest's own words, like they mattered. By month six she'd stopped opening source links at all.

A new hire named Kwame Osafo, sitting in on his first competitor review, asked why a line from last month's digest, a rival called FlawSight "adjusting its onboarding requirements," hadn't come up again since. Dorotea pulled the source for the first time in two months. FlawSight had cut the labeled defect images a new customer needed to provide from five hundred down to fifty, using a training method built to get useful results from far less labeled data. RivalScan's summary of it was accurate, word for word. It sat under the same plain header as a line three items above it about a homepage redesign.

At its worst, this doesn't cost a missed headline. It costs a real quarter: Corvid Vision lost three competitive evaluations in the eleven weeks nobody knew, each one citing "faster to get started" as the reason, before anyone connected it back to one unflagged line.

What I would leave alone: I wouldn't make RivalScan run more often. Once a month is the right cadence for reading. The fix isn't a faster digest, it's a better signal on the digest that already exists.

The lesson: a competitive-intelligence tool that never gets a fact wrong can still fail completely, because the failure was never in what it said. It was in what it never said about how much any of it mattered.

Now here is the same thing as a story

The short version above is what you'd say defending this design under interview pressure. Read this one for how the miss actually felt from the inside, before anyone at Corvid Vision had a name for what had gone wrong.

Dorotea had run competitive research at Corvid Vision for four years. She was the person other product managers asked before a big deal review, not because she read fastest, but because she could tell in ten seconds whether a rival's announcement was a rewritten homepage or a real move.

The first months with RivalScan were good ones. She'd get the one-page digest on the first of the month, coffee still hot, and spend an hour cross-checking it against a fifth of the raw sources, picked at random. Every time, the digest held up. It felt, she told a colleague once, like having a smart intern who never got tired and never missed a source.

The anchor: not a faster summary, a tag that says what the summary itself never could.

The habit thinned in three beats that nobody marked on any calendar. By month three, she'd stopped checking a random fifth and started checking only the items that sounded, in the digest's own words, like they mattered. By month six, she'd stopped checking sources at all, because the digest had simply never once been wrong. By month eight, when Kwame Osafo joined the product team and sat in on his first competitor review, she couldn't remember the last time she'd opened a source link.

Kwame asked one small question. He'd read last month's line about FlawSight, a rival two years younger than Corvid Vision, "adjusting its onboarding requirements," and wanted to know what that meant in practice, since it hadn't come up again. Dorotea didn't know. She pulled the source for the first time in two months.

Same true sentence. One version buries it. One version says why it matters.

FlawSight had cut the labeled defect images a new customer needed to provide from five hundred down to fifty, using a training method built for exactly this: useful results from far less labeled data. For a factory switching vision systems, that's the difference between a two-week pilot and a two-month one. RivalScan's summary of it was, word for word, accurate. It sat under the same header, in the same font, as a line three items above it about a homepage redesign.

The old decision went back to RivalScan's very first design meeting, eight months earlier. Someone had asked whether the format needed a way to mark which updates actually mattered. The answer at the time was no: the team was still reading every source themselves, so a flag would have been one more thing to maintain for information they were already checking by hand. That was a fair call in month one. It quietly stopped being one the moment people started trusting the page instead of the sources under it.

The two items worth escalating sat in the quiet corner the whole time.

The replay: same new hire, same question, but the anchor already exists. FlawSight's post gets tagged strategic the day RivalScan reads it, because the structured check behind the tag, not the prose, notices the labeling number dropped ninety percent. It reaches Dorotea's inbox that afternoon. She raises it in the next roadmap review, four days later, not eleven weeks in. Corvid Vision ships a comparable low-label onboarding flow four months before the next big evaluation cycle, instead of losing three of them first.

One version reads every line the same way and finds out what mattered by accident, eleven weeks late, from a question nobody was required to ask. The other version reads every line the same way and finds out what mattered by design, the same afternoon.

What Dorotea took from it wasn't "read more." It was that she'd built a filter that was honest about its facts and silent about its priorities, and in a market that moves monthly, silence is its own kind of wrong answer.

SPARK, the one decision the whole system leans onNot a longer reading list. SPARK is what decides which line gets a person's attention and which one waits.

The recap, one line per letter: situation is a four-year analyst who could once tell a real shift from a rewrite in ten seconds, anchor is a severity tag driven by what changed rather than how it reads, payoff is trading a monthly ritual for a standing watch, risk is auto-escalating anything touching pricing, architecture, or data requirements no matter the tag's own confidence, and keep out is no real-time alert firing on every single update.

And if you want to be sure it really works, try it somewhere elseSame five letters, a county planning office instead of a factory floor. A different old decision breaks this one.

Bellwether County's planning department runs CodeScreen, a tool that pre-screens architectural drawings against local building code before a human reviewer sees them. Mapped onto SPARK: situation is that reviewers used to redline every drawing against a paper checklist, one page at a time, alone. Payoff is trading "every override is a personal judgment call to defend" for "every override is a routine, tracked signal." Anchor is a required one-tap reason code every time a reviewer overrides a flag, built into the workflow itself instead of left as an optional comment. Risk is that the first time an energy-code amendment shifts what counts as compliant, CodeScreen's flags on solar setbacks keep reading confident and stale unless the override log is actually watched, so the fix routes any drawing type with a rising override rate to a human code review within the month. Keep out is no automatic approval for high-confidence drawings on day one, since the category that just changed underneath the tool is exactly the one that needs a person looking, not fewer of them. The old decision here isn't a missing severity tag, it's an attribution choice: whether an override is visible to anyone but the reviewer who made it. Making it invisible was cheap when the code rarely changed. It got expensive the moment the code did, because nobody could see the tool quietly drifting out of date.

The third box is the one the first design left out entirely.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "tag every competitor signal by what it would change for the customer, not by how loud it is, and only interrupt a person for the ones that would," and stop.
Cost: leadership cuts the competitive-intelligence role to part-time. Keep the severity tag, drop the weekly batch review to monthly, since the tag, not the reading cadence, was doing the actual catching.
The model gets better, for real: if the tagging runs a full year with zero missed strategic shifts, that's the moment to check it hardest, not relax it, since a filter's best year is exactly when nobody's watching it anymore.

Four months, four states, and nobody chose the last one on purpose.

Where people run it wrong.
They build the faster digest and skip the tag, mistaking speed for the actual fix.
They let every update interrupt someone, and train the whole team to stop reading any of them within a month.
They copy a severity rule from a market that used to move slower, when what counts as "strategic" in a category changing monthly isn't what counted a year before it started moving that fast.

How to use it live. The moment an interviewer asks how you'd track a fast-moving market, ask yourself: which of these updates would actually change what a customer decides? Build the system to answer only that question, and let everything else wait.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0