JumpStart

James handed me a prototype and a name on 9 June. His pitch was paid media reporting with a friendlier front door. The tools agencies use felt clunky and heavy on data, and he wanted something closer to Monzo or Slack.

My audit said the instinct was right and the product was still a concept. The AI summary had nobody behind it. The client’s question box went nowhere. The login was a role picker. A second brief, drafted with AI, described client onboarding instead of reporting, so I went with what James had said in the room and made reporting the first job.

I went through the tools agencies already pay for, including Funnel, Supermetrics, AgencyAnalytics, Whatagraph, DashThis, Databox and Looker Studio. They’re strong on connectors, templates and scheduled reports. Every one of them assumes the reader already knows what CPA means.

The person reading a JumpStart report doesn’t. She’s a marketing lead with a budget to defend and a finance director who asks every month whether the ads are working. The night before that meeting she feels a small, familiar dread, because she can’t tell.

The early versions came quickly. By 20 June there was a working build with real sign-in and separate client views, plus a line of static prototypes I could change in minutes. Each version added what the team asked for: sign-off, a cheat sheet, benchmarks, a funnel, a channel comparison.

Then I looked at V3 and felt it was good-looking but soulless. That happens when you build before discovery has finished. So I went back to the start and ran the process properly. I set up three AI agents with one job each: a researcher for the client’s real worries, a designer for tone and feeling, and an editor for what to cut. Then I pulled what they found into one direction.

Six principles came out of it, and they still hold.

V5 brought it together. The read sat at the top, a “needs you today” list held anything waiting for sign-off, client questions reached a real person, and the report exported to A4. James and Alex reviewed it on 25 June.

They also wanted help justifying budgets in first meetings. I built a planner that works both ways. Start from the spend, or start from the goal and work back to the spend it needs. V5 became the approved reference at jump-start.app/approved, and later the first version to run on a real client’s data.

July was for visual direction. I tried a white editorial version, warm paper, a navy and orange brand, a lime bolt, a museum painting behind the numbers and a pixel pattern. Some were beautiful. None of them fixed the real problem, which was the order things appear in.

On 11 July I stopped and went back to first principles. A report should make a claim, show the evidence, then let you inspect it. Most dashboards do it the other way round and leave the client to find the story.

“No action needed” became one of the most important states in the product. A short report is proof that someone looked.

On 21 July I connected the National Football Museum’s Meta account to V5. A real campaign in the approved design taught me more than any mockup had. It also showed what we couldn’t honestly say yet. I took out the demo targets, benchmarks, funnel claims and invented action plans, and kept only what Meta reported.

On 13 August the client page went behind a password at jump-start.app/nfm. It got its own release branch, so my design work couldn’t change what the client saw by accident.

In late August the report became a briefing. It opens with one sentence, “Every £1 you spent brought back £2.86”, says whether anything needs you, and the account lead checks it before it reaches you. The numbers are there to back up the sentence.

Alex said it still looked complicated next to the Monzo and Slack feel James wanted. My first move was to cut, and I took out the week on week section. That was wrong. The page got emptier and no clearer. I put it back and redesigned it with paired bars, a plain line under each and changes written out, like “Down 8%, getting cheaper”.

Then I worked on the tone. The briefing was correct but cold, so I built the same report at four temperatures, from plain to handwritten, and judged each one against the last.

The second one shipped, with the counter from the third. Most of the warmth lived in the words: a greeting, a verdict, a sign-off. These are the rules I kept.

The agency needed its own view. The desk opens with what needs a person: broken things first, then decisions, then reports waiting to be signed. “1 connection down, and 4 other things for a person.” Anything that’s fine stays out of the way.

Two details show how I work. I tried eight sounds for the desk and kept three. The three I kept had measurable things in common: quiet peaks, a low body under the click, nothing high for long. So I wrote those down as rules, and a test now checks every new sound against them.

The logo lockup got the same treatment. A mark next to the JumpStart wordmark is aligned from two numbers measured off its own file, instead of being nudged until it looks right.

Clients forward reports to their boss, so the export matters as much as the page. One dialog serves the client and the agency. Pick any two dates, see the deck before it goes, then send it as a PowerPoint deck, a PDF of the deck, a full A4 report or an email. It remembers each client’s style: House, Night or Dawn. Until a person signs a report off, JumpStart stamps it as a draft.

First use got the same care. The agency sets a client up in seven steps. The client gets a short welcome that explains the briefing before they see their numbers.

Most of the feedback came from James and Alex, in reviews and on Slack. I’ve paraphrased it from my notes.