Where the first clients come from when you have no case

The question assumes a market: somewhere there is a list of buyers and the work is to reach it. For someone with no case study, no company and no references, it is the first mistake. Nobody in that position buys reach. They buy a reason to believe a stranger, and it has to be tangible before the invoice exists.

What you actually have to trade

Capability you have not demonstrated is not an asset; it is a claim. The only thing available before the first contract is a small, finished, checkable artifact: a daily report that turns three exported spreadsheets into one number the store manager can verify against her own count. It is unimpressive, which is the point: unimpressive things get checked in ten minutes by the person who will use them, and a check is the start of a reference.

So the exchange is specific. You give a working thing plus the cost of checking it. They give twenty minutes and, if it holds, a sentence to someone else. The first client is not the person who pays. The first client is the person who verifies.

That is why one finished delivery outweighs any number of certificates, and why the ways people enter the role run through a delivery, not a course.

Four channels and what each one costs

Every channel works, and each is paid for in something other than money; the currency decides whether it fits.

Public output. Build in the open: a demo clip, a write-up of one solved case, a tool people can try. The cost is unpaid time — often months — and an audience that reads rather than buys. What it does produce is a stranger who arrives already half-convinced.

A small paid job inside someone else's project. The reliable version is a partnership with people already inside the account — the ERP vendor, the website shop, the office-systems supplier — taking one small need inside a project they already own. The cost is that you are now priced as a subcontractor, and the second job will be negotiated down from the first. Settle in writing who talks to the client, who collects, and who handles support. Partnerships die on those three questions far more often than on the work.

The room. A meeting where a specific person can point at a specific pain. The cost is travel, and a vague ask gets a vague answer. "Do you know anyone who needs AI?" produces nothing. "I built this for a store's daily report; do you know anyone who still merges store spreadsheets by hand?" produces an introduction, because it is a question the listener can actually answer.

Local and policy-adjacent resources. Outside the largest cities, the practical route runs through whoever already has the local relationships — an industrial park, an association, a service provider with government accounts. The way in is a project small enough to skip a tender. The cost is needing the partner's trust first, and a procurement rhythm that is not yours.

Referral is not a fifth channel. It is the receipt from the other four, and it arrives roughly one delivery late.

What makes a first project survivable

The safe first project has three things definable in advance: the input, the output, and the person who checks it. The manager supplies sales sheets in an agreed format; you produce the totals and the refund figures; the manager reconciles them against her own count. Every part of it can be argued about on day one and settled on paper. The seven questions are the checklist for that conversation, and they are worth running before a single line is written.

Size is not the number of features. It is the surface area: how many systems the thing touches, how many people have to change a habit, and whether something has to be watched at night. A report over an exported file is small even when it looks elaborate; a one-screen feature writing back into a legacy order system is large even when it looks simple.

Where the free demo stops

A demo is legitimate when it shows one key step on simulated data, or on a few sheets the customer has already desensitised: it is cheap, and the customer sees input and output in the same frame.

The signal to start charging is not the customer's enthusiasm. It is the shape of the next step. If moving forward requires organising months of history, negotiating rules that keep changing, or touching the company's systems, the work has become an engagement, and engagements are paid. The smallest honest version is a small fee for one validation round.

Before that conversation, three facts have to be settled: who will use it, who supplies the sample and verifies the result, and who decides and signs. If the answers are "someone in IT", "we will see", and "the boss", there is no engagement yet. Stop at the demo, ask for the material, and leave the door open. A demo that continues without a verifier is not marketing; it is unpaid work with a friendly audience.

The first invoice is arithmetic, not confidence

A worked example, deliberately boring. Two days of conversations and sorting, four of building, two of testing and handover, two held for revisions: ten person-days. At 1,000 a day that is 10,000; add 1,000 for the pilot's model and server costs and the floor is 11,000. A final price of 20,000 leaves 9,000, which is not profit: it has to cover the weeks spent finding the client, overhead and tax, none of which appear in the ten days.

Two things follow. The quote lists what is included — the fixed-format report, a stated number of revisions, pilot support — and what is not: post-launch model and server costs, adaptation to a changed spreadsheet format, rework from a changed requirement. And payment is staged against visible artifacts. On a 20,000 job: 8,000 on a confirmed scope and schedule, 8,000 when the agreed samples run in the agreed environment and reconcile with a manual count, 4,000 after a trial week with scripts and handover. Scoping and pricing is the longer version of the same argument.

The test

Before the next free demo, write three things down: the name of the person who will check the output, the name of the person who can say yes to money, and the date the sample arrives. If any of the three is blank, the demo is entertainment. The correct next message is a request for the sample, not a build.