cottam.design

How the work happens

Great products aren't guessed at. They get worked out.

Every project runs through the same four phases. Skipping one is how teams end up six months in, building the wrong thing beautifully.

01Define

First, we get everyone solving the same problem.

One brief, one set of goals, one definition of done, agreed before anything gets drawn.

A product roadmap board: columns of colour-coded project cards, each with a description, a list of next steps and a set of links.
  1. 01Meet with stakeholdersWhat has to be true in six months for this to have been worth doing?
  2. 02Align on goals & purposeWho is this for, specifically?
  3. 03Identify key featuresWhat are we willing to cut?
  4. 04Set initial prioritiesWhat happens if we do nothing?

Project Alpha

A personal command center for the whole process: briefs, priorities, open work and every agent run in one place, so nothing gets tracked in someone's head.

Shown with sample data

Users are already in the room. What the business believes about them gets checked against what they've actually said.

OutputA brief everyone has read and nobody disagrees with

AI agents on this phase

02Discovery

Then we go looking, widely and on purpose.

The first idea is the fastest one, not the best one. So we go and find the alternatives.

A wall of design references: grids of collected interface shots grouped into sets: tickets, AI chat, FAQs, onboarding, toast messages.

Where we look

Competitors
Adjacent industries
Support tickets
Session recordings
Prior research
Pattern libraries
Failed attempts
The obvious answer
  1. 01Gather inspiration
  2. 02Research patterns
  3. 03Explore design directions

Findings go back to the team immediately, while directions are still cheap to argue about.

OutputTwo or three genuinely different directions, not one polished guess

AI agents on this phase

03Design

Rough before real.

Fidelity is earned one review at a time, which is why sketches stay ugly on purpose.

Two hand-drawn board sketches of a page, annotated in the margins with what each block has to say.

01Sketch

Grey-box wireframes of the same page at several widths, beside a list of the sections it needs.

02Wireframe

A high-fidelity exploration board: two competing chart directions for a revenue card, each drawn at four states: empty, partly filled, long-tenured, and with its tooltip open.

03High-fidelity

An A/B test result for a dashboard redesign: primary, secondary and guardrail metrics, each with its change and its chance to win.

04Review & iterate

Every stage goes in front of someone who doesn't already know the answer.

OutputScreens resolved enough to build, and then we build them

AI agents on this phase

04Validate & Launch

Then we prove it, and keep watching.

Launch is a checkpoint, not a finish line. Instrumentation goes in before it ships.

A results dashboard: a finished A/B test on a three-step signup with confidence intervals and per-metric change, a traffic and reach trend, a five-stage signup funnel, and progress against a team goal.
Anything uninstrumented gets judged on opinion instead.

What we run, and what it answers

  1. 01Lyssna testsCan someone who's never seen this figure it out?
  2. 02A/B testing (GrowthBook)Is it measurably better, or just newer?
  3. 03LaunchBehind a flag, if the risk earns one.
  4. 04Analytics & trackingWhat are people actually doing once it's live?
  5. 05Monitor & iterateDoes the result still hold a month later?

What we learn here rewrites the next Define. The four phases are a circuit, not a line.

OutputA shipped change, and the evidence to say whether it helped

AI agents on this phase

That's the process. Every month, for one rate.

See the priceBook a call
· drag to pan