In this note
- Why a loop, not a project
- Week one: measure
- Week two: vary
- Week three: select
- Week four: inherit
- What a Generation doesn't do
- Do this this week
- Questions people ask
Over the past few weeks we've written about the pieces: the seven stages, the six buyer types, the Orchid Test, the predictions. This note puts them together. It walks through one Generation, which is what we call a single turn of our improvement loop, from the first day to the last.
A Generation usually takes a month. It has four steps, always in the same order: measure, vary, select, inherit. The rhythm bends a little depending on what's being worked on, but the order never changes, because each step depends on the one before it.
Why a loop, not a project
Most marketing is organised as projects. A website, a campaign, a rebrand. Each one has a launch date, and the launch is treated as the finish line. Then things go quiet until the next project, often years later.
The trouble is that buyers keep changing, rivals keep changing, and the way people search and ask questions keeps changing. A brand that's fit on launch day slowly becomes less fit, and nobody notices until a big project is needed again.
A loop works differently. Instead of one large change every few years, it makes one focused change every month, measures what it did, and keeps the winners. Each month starts from what the last month proved. Over a year, that's twelve small, checked improvements that build on each other.
Each Generation starts from what the last one proved. That's where the compounding comes from.
Week one: measure
Every Generation starts by reading the scoreboard. We look at the moments we're working on, the specific buyer types at specific stages, and check where they stand today.
Two questions matter. First, what did last month's change move, and how does that compare with what we predicted? Second, which moment is now the weakest? The answer to the second question is usually the next thing to work on.
The measurement uses the same buyer questions each time, asked of AI assistants and search, so the comparison is fair. We use RocMode to keep the baseline and track movement. Without a consistent baseline, you're comparing this month's impressions with last month's memories.
Week two: vary
This is the creative step, and it's the one most brands skip. Having found the weakest moment, we don't make one answer to it. We make several, and they have to be genuinely different.
If the weak moment is the sceptic at Believed, the variations might be a page of named results, a plain-English guarantee, a short video walking through the process, and a comparison that admits where a rival is stronger. Different ideas, not four shades of the same one.
Why several? Because if you only make one, you can't tell whether it was the best available answer or just the first one somebody thought of. Difference is the raw material. The bolder the variations, the more you learn.
Week three: select
Now the buyers decide. The variations go in front of real buyers, in whatever way suits the moment: live on the site, in sales conversations, in emails, or in front of a panel from the market.
Before anyone looks at results, we write down which variation we expect to win and what it should move. Then we look. We keep the winner, drop the rest, and write down why. The losers go in the report alongside the winner, because knowing what didn't work is how you trust what did.
This step is where opinions stop deciding. The most senior person's favourite doesn't win by default. The buyer does.
Week four: inherit
The winner doesn't just stay where it was tested. It becomes the new normal everywhere it fits. If a new way of describing your guarantee won on the pricing page, it moves into the sales deck, the proposal template, the frequently asked questions and the answers you want AI assistants to find.
Then the next prediction gets written, the next weakest moment is lined up, and the following month begins from here rather than from scratch.
What a Generation doesn't do
It doesn't fix everything. One Generation usually moves one or two moments. That can feel slow if you're used to big launches, but big launches change so much at once that nobody can say what worked.
It also doesn't replace big pieces of work. Sometimes the weakest moment needs a new website or a new position, and that's a larger adaptation. The loop just makes sure that after the big piece launches, it keeps improving rather than freezing.
Do this this week
You can run a small Generation of your own without any special tools.
- Measure one moment. Pick a buyer and a stage. Write five questions that buyer would ask and record what they'd find today.
- Make three variations. For the weakest answer, draft three genuinely different improvements. Different ideas, not different wording.
- Write your prediction. Before testing, write which one you expect to win and what it should move, with a date.
- Test, keep, spread. Try them in front of real buyers, keep the winner, and list every other place it should appear.
Questions people ask
What is a Generation?
A Generation is one turn of the improvement loop: measure, vary, select, inherit. It usually takes a month. Each one measures where you stand, makes several different answers to the weakest moment, lets buyers choose between them, and spreads the winner everywhere it fits.
How many moments does one Generation improve?
Usually one or two. The value comes from compounding: each Generation starts from what the last one proved, so a year of monthly Generations adds up to many small, checked improvements that build on each other.
Do we need special software to run Generations?
You can run a simple version with a spreadsheet and discipline. Consistent measurement is what makes it reliable, which is why we use RocMode to keep the baseline and track movement across all 42 moments.