Three numbers, none of them right
The processor dashboard knows about card customers. The bank knows about everyone who paid by transfer. The accountant knows about what was invoiced. All three are accurate about their own slice and none of them is your revenue.
Most founders resolve this by picking whichever is easiest to read — usually the dashboard — and mentally adding “plus the enterprise ones”. That works until someone asks for the actual figure, at which point it turns out the mental addition was three months out of date and one customer had quietly stopped.
The fix is not a system. It is an hour a month spent putting both sources into the same grid, with the definitions written down. It stays adequate for longer than most people expect.
The awkward stage
This piece is about a specific window: past the point where you can remember every customer, before the point where billing is properly wired up. It typically lasts a couple of years and it is when most fundraising conversations happen.
Three things are true in that window. Self-serve customers pay by card. Larger customers ask for an invoice and thirty-day terms, and you say yes because they are worth it. And nobody has time to implement a billing system that handles both, because the product needs building.
All three are correct decisions. The cost is that no system has the whole picture, and the gap grows precisely in the direction of your most valuable customers — which is the worst possible place for a blind spot.
Two payment routes, two data problems
| Card customers | Invoice customers | |
|---|---|---|
| Where the detail lives | Processor dashboard | Nowhere — only the bank |
| What the bank shows | A net payout covering many | One line per customer |
| Failure visibility | Reported and retried automatically | Silence |
| Timing | Charge date, payout a few days later | Whenever they get round to it |
| Typical size | Smaller, many | Larger, few |
| Main risk | Forgetting fees are netted off | A customer stopping unnoticed |
The third row is the asymmetry that matters most. A failed card is reported to you and usually retried automatically. An invoice customer who stops paying produces nothing at all — and they are your largest accounts, which is covered in full on failed payment and churn signals.
The second row is where double counting creeps in. A payout is not customer revenue; it is the delivery of revenue you already counted on the processor side. Counting both is the single most common way an early revenue figure ends up overstated.
Combining them properly
Four rules, and they resolve almost every question that comes up.
Card customers come from the processor, not the bank.The processor knows who paid what; the bank only knows a net total arrived. Use the processor’s own statement for that side — our payout statement converter turns it into rows.
Invoice customers come from the bank. One line per customer, grouped by payer, exactly as described on recurring revenue from bank statements.
Use the payout only to reconcile.It confirms that the processor’s figures actually arrived and reveals the fees. It never contributes revenue — that would count the same money twice.
Report gross of fees, and track fees separately. Processor fees are a cost of doing business, not a reduction in what the customer pays. Netting them makes your revenue look smaller and hides a cost line that grows with you.
One number that survives scrutiny
Normalised monthly recurring revenue, with the definition written at the top of the sheet. Three lines: gross or net of VAT, annual plans spread across twelve months, one-off fees excluded.
That last point is where early-stage numbers most often inflate. A setup fee, a paid pilot or a bespoke integration is real money and it is not recurring. Including it produces a figure that promises something next month that will not arrive, and it is the first thing an experienced reader tests.
Spreading annual plans is the second. Counting a year’s payment in the month it landed makes one month look extraordinary and eleven look like decline — the shape alone gives it away in a chart, without anyone needing to ask.
Write the definitions once and do not revise them to make a quarter look better. A number whose definition moved is worse than a lower number, because everything before it becomes unusable.
What investors actually check
Less than founders fear, and different from what they prepare for. Four things come up consistently.
Can you rebuild it? Diligence goes to bank statements and processor exports. A figure that cannot be traced back to those is a problem regardless of whether it happens to be accurate.
Has the definition been stable? A consistent, slightly conservative definition beats an optimistic one that changed in Q3. Changes get noticed and they colour everything else.
Is one-off revenue in there? The first thing anyone experienced tests, because it is the most common overstatement and the easiest to spot from the shape of the series.
What happened to the customers who left? Not the churn rate — the names, and what they said. Having recorded that at the time is the difference between a thoughtful answer and an improvised one.
None of these require a billing system. All of them require having been consistent for a year, which is the actual work and cannot be done retroactively the week before a raise.
Churn at low volume
With twenty customers, a churn percentage is close to meaningless. Losing one is five percent; losing two in a month is ten and looks catastrophic while being entirely normal variation.
Report the count and the names instead. “Two customers left this quarter, both small, both moved in-house” is more informative than any rate, and it is what a reader actually wants to know.
Report by revenue as well as by count, because at this size one departure can dominate. Losing your largest customer and losing your smallest are the same count and completely different quarters.
And watch contraction rather than cancellation. At low volume, a customer reducing their seats is a signal you can act on personally, which is one of the genuine advantages of being small — the full argument is on failed payment and churn signals.
Runway and burn from the same data
Revenue is the number founders present. Runway is the number they check on a Sunday evening, and it comes from the same converted statements with less work.
Net burn is money out minus money in, monthly. Runway is cash in the bank divided by that. Both are arithmetic on data you already have, and both are more urgent than MRR when the balance is finite.
Two cautions. Use a three-month average rather than a single month — one large annual payment either way distorts a single month badly. And separate one-off inflows such as a grant or a loan from operating income, or your runway will look longer than it is.
The forward-looking version is on cash flow from bank statements, and the founder-facing framing on the startup converter page.
A monthly hour
Once set up, the routine is about an hour, and it is worth protecting because everything above depends on it having been done consistently.
Convert
Bank statements and the processor export for the month. Minutes.
Assign new payers
Any names that appeared for the first time. Recorded once, so it holds.
Update the grid
Customers down, months across, normalised amounts.
Read it
New, expansion, contraction, churn — and the blank cells.
Do it on the same day each month. The consistency is what makes the series comparable, and a metric produced irregularly is one nobody trusts — including, eventually, you.
The first rebuild
Doing this for the first time is a different exercise from the monthly routine, and it is worth setting expectations: it takes half a day and it usually contradicts something you believed.
Start from the first paying customer, not from this year. Early months are tiny and quick to process, and a complete history is worth far more than it costs the moment anyone asks about growth rate.
Expect the total to be lower than you thought. Almost always, and for two reasons: one-off revenue was mentally included, and at least one customer stopped without anyone noticing. Neither is a crisis; both are better known.
Expect the shape to surprise you more than the level. Founders are usually roughly right about the total and wrong about the composition — an enterprise account quietly larger than the whole self-serve base is the classic finding.
Write the definitions before you compute anything. Deciding what counts after seeing the numbers is how a figure becomes something you have to defend rather than something you can simply state.
What goes in the board pack
One slide, six rows. More than that and the discussion becomes about the numbers rather than about the business, which is the opposite of what a board meeting is for.
| Line | Why it belongs |
|---|---|
| MRR, normalised | The headline, on a stated definition |
| Movements: new, expansion, contraction, churn | Explains the change without anyone asking |
| Paying customers | Guards against one large account flattering the total |
| Customers lost, by name | More useful than a rate at this size |
| Net burn, three-month average | The number that constrains everything else |
| Runway in months | The only one that ever gets acted on urgently |
The fourth row is the one founders leave out and boards ask for. Naming the customers who left, with a sentence about why, signals that you are tracking the thing that matters rather than the thing that looks best — and it prompts far more useful help than a churn percentage does.
Keep the same six lines every month. A pack whose contents change is one where the reader spends the first five minutes working out what changed, and that is five minutes taken from the conversation you actually wanted.
What it is worth
Three returns, and the largest is not the time saved.
Customers caught before they are gone. An invoice customer who stops paying is invisible without this, and they are your largest accounts. Catching one a year pays for the entire exercise many times over.
A fundraise that does not stall. Diligence questions answered in a day rather than a fortnight of reconstruction. This is a real difference in outcomes, and it is determined a year earlier.
Knowing your own business. Harder to quantify and probably the most valuable. Founders who build this grid usually discover something within the first hour — a customer already gone, revenue lower than believed, or an enterprise account quietly larger than the entire self-serve base.
The conversation this makes possible
The practical value of all this is not the slide. It is that a founder with a payer grid in front of them can hold a different kind of conversation, and the difference shows up in three places.
With a customer.Noticing a reduced payment in the month it happens allows a call that begins “I saw you dropped some seats — has something changed?” That call is welcome. The same call six months later, after a cancellation, is not.
With an investor. Being able to say which customers left and why, without preparation, changes how the rest of the meeting goes. It signals that the numbers are something you use rather than something you assemble for meetings.
With yourself. The least visible and probably the most useful. A grid makes it hard to keep believing a comfortable story about the business — that the self-serve base is growing, that the enterprise customers are incidental, that last quarter was an anomaly. Sometimes those beliefs turn out to be right; the point is that they get tested monthly rather than annually.
Six mistakes
Reporting the processor dashboard as revenue
It excludes every invoice customer — which is to say, your largest ones.
Counting payouts and charges both
The same money twice. The most common way an early figure ends up overstated.
Including one-off fees in MRR
The first thing an experienced reader tests, and the easiest to spot from the shape of the series.
Counting annual plans on the cash date
One extraordinary month and eleven that look like decline — visible in a chart without anyone asking.
Reporting churn as a percentage at twenty customers
Statistically meaningless. Report the count, the names and the revenue.
Revising the definition to improve a quarter
Makes every prior period unusable, and the change gets noticed.
It is worth naming the trap at the other end too. Some founders, having built the grid, keep extending it — cohort tables, lifetime value, payback periods, expansion ratios by segment. Every one of those is a real metric and almost none of them mean anything with thirty customers and eighteen months of history. They produce confident-looking outputs from inputs too thin to support them, which is worse than not measuring: a rough number invites judgement, a precise-looking one invites belief. Keep the six lines, keep them honest, and add sophistication when the data can carry it rather than when the spreadsheet can.
When to stop doing this by hand
Four signs, and none of them is customer count on its own.
Proration appears. Once customers upgrade mid-period, a payment total can no longer be normalised correctly, because the split is not in the data.
Usage-based components. A base fee plus variable usage cannot be separated from a single amount.
More than one currency. Exchange movement starts appearing as growth, and unpicking it by hand is genuinely difficult.
Someone else has to run it. The moment the monthly hour is delegated, the undocumented judgements in your head become a problem.
Until then, the spreadsheet is not a placeholder. It is a method whose assumptions you can see and change — which is more than most dashboards offer, and it is why founders who did it by hand tend to understand their own numbers better.
A last practical point that saves a lot of frustration: do this before you need it, not when someone asks. The founders who find this exercise painless are the ones who started while they had eleven customers and it took twenty minutes. The ones who find it brutal are the ones reconstructing two years of history the week a term sheet conversation started, from statements they have to download from three banks and a processor whose export format changed halfway through. The work is identical; only the deadline differs, and the deadline is what makes it expensive.
Where it stops
It is not a billing system: no invoicing, no payment collection, no retries, no plan management.
It is not statutory revenue. Your accountant will report a different figure under accounting rules, and both can be right — understand the bridge rather than forcing them to agree.
It does not replace talking to customers. The grid tells you someone stopped; it never tells you why, and the why is the part that changes what you build.
And it cannot invent history. If you have never tracked this, start now with what you have — the first consistent year is worth far more than a reconstructed three.