CYCLOMEDIA · COMPETITOR INTELLIGENCE · 2026

The information was already there.Nobody had time to read it.

I designed a system that reads competitor news for a geospatial company, decides what actually matters, and only then interrupts someone.

248 things happened. 7 were worth someone's morning. Everything I designed lives in the gap between those two numbers.

01 — THE PROBLEM

Cyclomedia drives 100 cars around the world photographing streets. Their competitors don't send a memo.

Cyclomedia maps roads and infrastructure — over a million kilometres a year, sold to city councils, utilities and insurers in more than 30 countries. Big company, offices in the Netherlands and the US.

Meanwhile their competitors are busy too. Someone launches an AI feature. Someone quietly drops their price. Someone wins a contract in a city Cyclomedia was hoping to win. All of this is public. All of it is findable. And nobody has three hours a day to go find it.

My brief asked for an automated tool that checks competitors and sends a short weekly update to Microsoft Teams. It also contained one sentence in capital letters that I ended up taping to my brain:

Fair enough. So I did. And then I immediately made a completely different mistake.

02 — VERSION 1

I built a beautiful homepage for a product nobody needs to be sold.

My first version opened with a proper marketing page. Big headline. Hero image. A menu with Product, How It Works, Company. A Log In button in the corner. It looked like a real startup.

V1 — THE LANDING PAGE

'Stay ahead of your competitors with automated insights.' Nice line. Written for a customer who does not exist.

Then I thought about who would actually type this URL. Cyclomedia employees. People who already know what the tool is, why it exists, and who asked for it. I had built a shop window for a room that only staff can enter.

So I deleted it and folded it into the login screen instead. One page: you arrive, you log in, you're in.

V1 — LOGIN

Two separate pages doing one job.

FINAL — LOGIN

One page. The message lives beside the form.

The dashboard underneath it was closer to right. It already had the two ideas that survived all the way to the end: a category tag (Software, Hardware, Data) and an impact level on every update — and a "Why it matters" line explaining the news instead of just reporting it.

V1 — DASHBOARD

Cards in a grid. Which quietly fights the whole idea: if the system has already put the most important thing first, why am I laying them out side by side like they're equal?

KEPT

'Why it matters.' This one idea became the heart of the product.

CUT

'Time Saved: 2h 30m.' I invented that number. I had no way to measure it, so it had to go.

04 — THE DISAGREEMENT

Her idea: put thumbs up, thumbs down and comments on every update. Popular items float to the top, and the reactions help the system learn what's important.

It's a reasonable idea. I still thought it was wrong, for three reasons.

First, the whole product exists because the AI already does the filtering, carefully, with a score. If I then let votes reorder the feed, I've built a popularity contest on top of a judgement I just spent a month designing.

Second, votes and comments are work — and my brief specifically said the system must not depend on people maintaining it.

Third, a loud team can out-vote a quiet one. The signal that matters enormously to two people in Sales gets buried by the one everyone finds interesting.

So I proposed three smaller things instead: mark it Relevant (a quiet nod to colleagues, no ranking power), bookmark it for yourself, and share it with someone who needs it.

FINAL

Three actions. None of them changes the order of the feed.

05 — WHAT I BUILT

The final version starts with a subtraction.

Seven signals came in. Four are worth your time. That's the first sentence of the product, before any data appears, because the filtering is the feature.

And every update answers two questions, not one. What happened — and then So what: what this means for Cyclomedia and what to do about it. "Ecopia dropped their price 18%" is news. "Expect pressure on the Texas and Florida renewals where they're shortlisted — lead with the 5-year cost calculator" is something you can use before lunch.

INSIGHTS

'7 signals scanned. 4 updates worth your time.' One column, most important first, and it ends — no infinite scroll.

A SIGNAL

Who, what type, which layer of the market, how big, the fact, the 'so what', how sure the AI is, and the source. Enough to act or to ignore without clicking anything.

07 — WHO'S READING

A price cut is an emergency for Sales and background noise for Engineering.

This was the idea I liked most and it's embarrassingly simple: the system already knows who just logged in. So it should know what that person cares about.

Sales weights pricing changes and tender results high, product launches low. Product does the exact opposite. Same signal, one pipeline, different priority depending on who it reaches — and it lands in a different Teams channel for each team. Leadership gets summary only, because seniority means less detail, not more.

VIEWING AS

Five audiences, one pipeline. In the prototype you can switch teams to see the weighting actually change.

08 — USING WHAT THEY ALREADY HAD

Cyclomedia had already written the answers down.

They keep battle cards — documents on each competitor with their pricing, their weaknesses, and prepared counter-arguments. Good documents. Also frozen in time from the day they were saved.

My brief said use them, so I did — as the system's background knowledge. That's what turns a summary into an argument. Here's the actual trace:

Nothing there is invented. It's Cyclomedia's own positioning, pulled out at the moment it became useful.

And it works the other way too. If the card says a competitor is weak at AI and they then ship three AI features, that contradiction is itself worth knowing. So the profile is built as a living page: the stable knowledge in tabs — including Battle Notes — and the live signals running down the side.

COMPETITOR PROFILE

The battle card's own sections became the tabs. Nobody has to learn a new document.

One more thing about competitors: in this industry the dangerous one usually isn't the obvious one. A camera manufacturer is harmless right up until the day it adds AI analysis and lands in your business. So competitors are organised by where they sit in the value chain — hardware, positioning, data, software — and one company can sit in several at once.

COMPETITORS

The tabs are the market's structure, not a filing system I invented.

ADDING ONE

Checkboxes, not a dropdown — so a company that moves into your layer can be recorded as doing exactly that.

09 — BORING BUT IMPORTANT

A robot that breaks silently is worse than no robot.

Here's the nightmare: a scraper stops working, the feed gets thinner, and a quiet week looks exactly like a broken system. Everyone keeps trusting it while it watches nothing.

So the machinery is visible where it needs to be. Ten agents, each with a success rate. One is rate-limited. One has been offline for two days and says so in red, with a restart button. I left the broken one in the prototype on purpose — a screen full of green ticks would prove less.

Same for money. Every service is listed with what it costs: about $169 a month, all in. If someone has to justify this tool to a finance team, that's a design problem too.

PIPELINE

How much got thrown away.

AGENT HEALTH

One offline, 0%, two days.

COST

$169/month, itemised.