Unit economics when data is still thin

When you have only a handful of customers, unit economics cannot yet be measured reliably, but they can be built. The honest approach is to write a per-customer model with explicit assumptions, show ranges instead of precise numbers, separate what you have observed from what you estimate, and say which data you will collect in the coming months to replace the estimates. An early-stage investor does not expect certainty; they expect you to know which levers matter and how far you are from proving them.

In short

• Build the model on a single typical customer, not on company totals.
• Always separate three categories: observed data, reasoned estimates, assumptions still to test.
• Use ranges, with a cautious, a central and an optimistic scenario.
• Focus on the two or three levers that move the result most.
• Write a validation plan: which data, by when, through which experiment.

What unit economics mean in practice

Unit economics answer a simple question: when you win a customer, what does it cost you and what does it leave you over time? The basic building blocks are few:

• revenue per customer over the period (monthly or annual);
• gross margin, meaning revenue minus the direct costs of serving that customer: infrastructure, third-party licences, support, payment fees, any set-up work;
• the expected length of the relationship, linked to how many customers stay over time;
• acquisition cost, meaning what you spend on sales and marketing to win one new customer;
• the payback time needed to recover the acquisition cost through margin.

With many customers, these values come from historical series. With few customers they have to be built one by one, and that is where method and transparency matter.

Separate observed, estimated and assumed

The most useful table you can prepare has one extra column: for each item, state the nature of the figure.

• Observed: you have seen it happen. The price signed in the first contracts, the real monthly infrastructure cost, the hours spent onboarding one customer.
• Estimated: derived from partial data with explicit reasoning. For example, support cost calculated from three months of tickets from two customers.
• Assumed: you have no data yet. For example, the renewal rate after year one, if no customer has reached the end of their contract.

This distinction changes the tone of an investor conversation. You do not have to defend numbers you do not have; you have to show that you know which ones are weak and how you plan to strengthen them.

Ranges and scenarios instead of single numbers

A single value for relationship length or acquisition cost suggests a precision that does not exist. Prepare three scenarios instead:

1. cautious: the least favourable assumptions you still consider plausible;
2. central: your best estimate today;
3. optimistic: what happens if the main levers go the way you hope.

For each scenario, write one line on why the assumptions are what they are. If the model only works in the optimistic scenario, that is important information for you before it is for investors: it tells you which lever you must prove first.

Find the levers that matter

Not all variables weigh the same. Do a simple exercise: change one variable at a time by a reasonable amount, for example price, gross margin, relationship length or acquisition cost, and see how much the final result moves. Usually two or three variables explain almost all of the variation.

Those are the levers to focus experiments and data collection on. In many B2B startups, for instance, the cost of onboarding each customer and the founder time spent selling are underestimated items: they are worth measuring carefully even when they look small.

Hidden costs not to forget

With thin data it is easy to leave out items that weigh more later:

• founder time spent on sales, onboarding and support, valued at a realistic cost even if you do not pay yourselves today;
• pilot discounts and terms that will not be repeated, or that will instead become the norm;
• customisations built for one customer and not reusable;
• fees charged by platforms, marketplaces or distribution partners;
• variable AI or cloud costs that grow with usage rather than with the number of customers.

If an item cannot be measured yet, enter it as an assumption and label it as such. A missing item is worse than a badly estimated one.

A plan to replace estimates with data

The model should come with a short plan saying which assumption you are testing and how. For each of the main levers, write:

• which figure you need;
• which experiment or event will produce it (a renewal, a campaign, a new price);
• by when you expect to have it;
• which decision you will take based on the result.

Update the model with every new data point and keep earlier versions. Showing how your estimates moved closer to reality, or further from it, is one of the strongest arguments a young startup can make.

What to do this week

1. Pick one typical customer and write their model: revenue, direct costs, length, acquisition cost.
2. Label every item as observed, estimated or assumed.
3. Prepare the three scenarios, with one line of reasoning for each.
4. Run the one-variable-at-a-time analysis and identify the two or three main levers.
5. Add founder time and the other hidden items.
6. Write the validation plan with dates and linked decisions.
7. If you want an outside review of your model, get in touch.

You will find a spreadsheet template for unit economics in Resources.

Subscribe to Dispatch, the PATAVIUM newsletter, for practical tools to build your startup every week.