Market Finder: how to pick a market when you don't know the basis

The problem behind it

When we planned the Market Finder on 23 October, I put the question like this: there are markets available, I do not know which to pick, how do I pick, what is the basis?

Most small businesses sell to whoever comes first. A good number of marketing mistakes start there. A message can be strong and fail because the market is weak. A market can be strong and fail because the message is wrong. If you test the wrong thing, you learn the wrong lesson.

How it works

Five steps, one screen each, with the business-cycle map across the top:

1. Generate. 20 to 40 candidate markets from the job the client does for buyers, non-customer tiers, competitor audiences, own leads and niche-downs.
2. Market cards. Each market is scored 1 to 3 on four things: Pain, Targetability, Purchasing power and Growth. Evidence is attached, labelled "assumed" or "measured".
3. Shortlist. A funnel shows how the list narrows. The ranking is by pain first, then reach, then purchasing power, then growth. Any score of 1 sets a market aside unless a niche-down carries it. A shrinking market is never sent to niche down, because niching cannot fix growth. A person ticks the ones to test.
4. Test plan. A free round first, then a paid round within a budget the person sets. The default budget in the first build was Rs 45,000 in total, Rs 27,500 for the test round and the rest in reserve. Three to five messages per market, the same angle letter meaning the same idea. Nothing is launched or spent. It is words for a person to copy.
5. Results. Chats, booked calls and deposits are entered by hand. The grid reads "market weak" against "message weak". Results rank what people did: payment, then booked call, then chat, then cost per result.

The staff version uses 28 markets and a shortlist of 5. A free public version, switched off by default, uses 16 and 3, takes about three minutes, and by construction reads no client data.

The decisions

Code decides, AI drafts. The ranking and the reading of results are plain code. The AI only proposes candidates, drafts scores and writes messages, and code checks every reply. Wording never says "winner" or "dropped".

Advisory, never a gate. Nothing in Market Finder blocks a task, a flow or an approval. A person always picks.

Works with no AI key. Every step has a hand path.

Messages are tested neutral. On 31 October I argued that testing a market does not need the brand voice, because several messages and voices are tested anyway. My collaborator agreed, with a guard. The five test messages stay in one plain, neutral style so that the market and the angle are what get tested, not random changes of style. The hard rules stay: no false claims, no banned promises.

It did not use the Pricing tool. The scoring used one simple "Price" line from the business description to judge purchasing power. Linking to the Pricing tool was listed for later.

Two later additions

Stickiness (4 November). I added a goal: customers who stay. Low churn, high lifetime value, not only a small share climbing the value ladder. For Mukika that means keeping the customer after the game session. For IIPTA it means students graduate and leave, so we need reasons for them to stay. So each market gets a stickiness check with four yes or no questions: repeat need, switching cost, community, ongoing outcome or renewal need. It produces an estimated churn, a lifetime value and a range, from one shared economics module. It sits outside the four filters, so it never removes a market. It is shown on every row. One caution recorded in the spec: the starting churn bands are placeholders we chose, editable, to be replaced by CRM cohorts later. They are not findings.

Story cards. You write one value proposition, choose markets, and get one card per market: the pain or gain line, the moment they feel it, the proof needed, which offer rung fits (kept as an id, never free text), easy or hard to reach, and stickiness. Every line edits in place. An article starts from a card, and the card is drawn from the tool's own scores, so the article cannot disagree with the tool. Each proof item shows "have N verified facts" or "no verified fact yet".

What went wrong

The very first seed was Mukika's own round: 37 cards, 6 ranked for testing, 16 messages. I have no recorded result from a paid test. The tool shows how to run one, but I cannot yet tell you which market converted.

What's next

Link to the Pricing tool once it has been reviewed. Replace placeholder churn with measured cohorts. Connect ad results by API; today they are typed in. The public version still needs a decision to switch it on.

Numbers at a glance