A 13-week cash flow forecast is most accurate in its first four weeks, where it is built largely from committed data, and least accurate in weeks nine to thirteen, where assumptions take over. That decay is not a flaw in the format. It is the reason the format exists: thirteen weeks is roughly the longest horizon over which a weekly, cash-level forecast stays reliable enough to act on.

This page covers what the commonly quoted accuracy percentages are worth, why accuracy falls away across the window, what a real accuracy standard looks like where one exists, and how to measure your own. If you are setting one up from scratch, our guide to implementing a 13-week cash flow forecast covers the build itself.

What the commonly quoted percentages actually are

Ask an AI assistant, or read almost any article on the subject, and you will meet the same gradient: roughly 90 to 95 per cent accuracy in weeks one to four, 85 to 90 per cent in weeks five to eight, and 70 to 85 per cent in weeks nine to thirteen.

Those figures are targets, not measurements. They circulate as advice on what a well-run forecasting process should aim at, and no published study measures 13-week forecast accuracy against realised cash for businesses of this size. The versions in circulation do not even agree with each other: the first four weeks appear variously as 90 to 95 per cent, as 95 per cent or more, and as 85 to 95 per cent, and the same numbers are presented as observed performance on some pages and as something to aim for on others.

So treat them as a rule of thumb about shape rather than a benchmark to hit. The shape is right, for structural reasons worth understanding. The percentages will tell you far less about your forecast than a month of your own variance tracking.

Why accuracy decays across the thirteen weeks

The decay is structural. Each part of the window is built from a different grade of information.

Weeks one to four are mostly committed data. Invoices you have already raised, bills already approved, payroll, rent, loan repayments and tax amounts already known. The cash events exist; the only real uncertainty is timing, and even that is bounded by due dates and standing orders.

Weeks five to eight are a mixture. Some invoices are raised, some are not yet. Recurring costs are predictable; variable ones are estimates. Timing uncertainty widens, because a customer who pays fifteen days late moves cash across week boundaries.

Weeks nine to thirteen are mostly assumption. Revenue in this part of the window often depends on invoices you have not raised, work you have not won, or renewals that have not been confirmed. Two variables dominate the error here: customer payment timing, because the gap between an invoice's due date and the date it is actually paid compounds across the window, and pipeline assumptions, because forecast sales convert later, smaller, or not at all.

Payment timing is measurable at a national level, and the UK numbers show why it dominates. Under the reporting regulations that require large UK businesses to publish their payment performance, large businesses took an average of 32 days to pay suppliers in 2025 and paid 15 per cent of invoices after the agreed terms, according to Department for Business and Trade figures. Both have improved since reporting began in 2018, when the average was 35 days and a quarter of invoices were late. The sector spread is wide: manufacturing averaged 45 days, financial and insurance activities 21. If your customers are large businesses, that is the distribution your receipts are drawn from, and it is also a moving target, which is why last year's collection pattern is an imperfect guide to this quarter's.

There is a fourth source of error that has nothing to do with prediction: known cash events that simply never make it into the forecast. Quarterly VAT, annual insurance renewals, corporation tax instalments. These are data-capture failures, and they are common enough that they distort more forecasts than genuine surprises do.

What actually drives forecast accuracy

Since no published percentage will tell you where you stand, the useful question is not "what number should I hit" but "what determines whether my forecast is dependable". Four things, in practice.

The freshness of the ledger data at build time. A forecast assembled from a bookkeeping export that is a week old inherits a week of drift before any judgement is applied. Rekeying figures by hand adds transcription error on top. Nobody has put a reliable number on how much staleness costs you, and the claims that do circulate come from software companies describing their own products. The direction is not in doubt though: a forecast built on last week's ledger is describing last week. We cover what current data does and does not mean in what real-time cash flow visibility actually means.

Payment-timing assumptions. A forecast that books every invoice on its due date will be reliably wrong, because customers do not pay on due dates. Forecasting against expected payment dates, informed by how each customer has actually behaved, is the single largest accuracy improvement most finance teams can make.

Completeness of irregular outflows. The VAT quarter and the insurance renewal are perfectly predictable; they just have to be in the model.

Update cadence. A 13-week forecast updated weekly corrects itself: last week's misses inform this week's assumptions, and week fourteen enters the window as week thirteen leaves it. A 13-week forecast updated monthly is, for most of the month, a stale document with a reassuring name.

One thing does not appear on that list: any tool's ability to remove uncertainty itself. Software removes input error. It does not make an unpredictable customer predictable, and it cannot firm up a pipeline that has not closed. If your receipts are genuinely volatile, weeks nine to thirteen will stay uncertain in any product, including ours. What good tooling changes is that the uncertainty you are looking at is real uncertainty, not stale data or missed entries dressed up as uncertainty.

Where a real accuracy standard does exist: lender covenants

There is one place where somebody has written down how accurate a 13-week forecast has to be and attached consequences to it. When a lender funds a business through a restructuring, the facility agreement typically requires a rolling 13-week forecast and sets a permitted variance against it. Miss the tolerance and it is a covenant breach.

Executed agreements are more forgiving than the accuracy gradient implies. Permitted variance commonly sits in the region of 10 to 20 per cent, with 15 per cent recurring most often. Three features of how it is tested matter more than the percentage:

• Variance is tested cumulatively, over a rolling multi-week period. Four weeks is the common window. A single bad week inside a compliant four-week period is not a breach.
• Only adverse variance counts. Coming in better than forecast is not a failure. The test is one-directional.
• Receipts and disbursements are tested separately, rather than the closing cash balance, so that an under-forecast on payments cannot be hidden by an under-forecast on receipts.

Two caveats. These clauses appear in distressed and lender-monitored situations, mostly in larger companies and mostly under US practice, so the specific percentages are not a standard for a healthy 30-person business. And a covenant is a tolerance a lender will accept, not a measurement of what forecasts achieve.

What does transfer is the design. The people with the most money at stake, who have every reason to be strict, do not test a single week against a single number. They test accumulated error over a month, they ignore pleasant surprises, and they look at inflows and outflows separately. That is a considerably better model for judging your own forecast than a percentage per week band.

How accurate does a 13-week forecast need to be for planning purposes?

Accurate enough to make the decision in front of you, which differs by week.

The near weeks carry operational decisions: making payroll, covering the VAT bill, the date your balance crosses a minimum threshold. These decisions need cash-level confidence, and the committed data in weeks one to four supports it.

The later weeks carry directional decisions: whether you can afford a hire, when to time a capital purchase, whether a funding drawdown is needed this quarter or next. These decisions do not need penny accuracy. They need the trend and the range to be trustworthy.

For planning purposes, consistency matters more than precision. A forecast that runs five per cent optimistic every week is dependable, because you can plan around a known bias. A forecast that swings between optimistic and pessimistic is not, even if it averages out closer. And the diagnosis runs in one direction: if your first four weeks are missing badly, the problem is almost never forecasting skill. It is inputs, which is fixable.

How to measure your 13-week forecast accuracy

Published benchmarks cannot tell you whether your forecast is good. A weekly variance habit can, within about a month.

Step 1: Snapshot the forecast before you roll it. Each week, save the closing version of the forecast before updating it. Without the snapshot there is nothing objective to compare against.

Step 2: Compare actual cash movement to the forecast for the week just ended. Take actual receipts and actual payments from your accounting platform and set them against what the forecast said, keeping inflows and outflows separate so they cannot net each other off.

Step 3: Judge the result over a rolling four weeks, not a single week. Accumulate the variance across the last four weeks before deciding whether the forecast is performing, which is how lenders test the same document. One bad week is usually one late payment, not a broken forecast.

Step 4: Track variance by week number, not just in total. A miss in what was forecast as week one tells a different story from a miss in what was forecast as week ten. Recording variance against the original week position shows you where in the window your assumptions break down.

Step 5: Separate the repeating misses from the one-offs, and fix the assumption underneath. One late payment is noise. The same customer paying three weeks late every month is a pattern. Move expected payment dates to match observed behaviour, add the irregular outflows that caught you out, and revisit pipeline conversion assumptions the misses have tested.

One practical warning about percentages. If your inflows and outflows are close to matching in a given week, net cash movement can land near zero, and a percentage variance calculated against a near-zero figure produces a meaningless number that swings wildly on rounding. Where that happens, measure the variance in pounds or dollars rather than as a percentage, and measure receipts and payments separately rather than netting them first.

After four or five cycles you will know your real accuracy by week position, which is more useful than any published figure, because it is yours.

How Float fits

Float is built around the two accuracy drivers a tool can actually influence: input freshness and payment timing.

Float connects to Xero or QuickBooks Online through a one-way, read-only connection and imports your bank transactions, invoices and bills automatically every 24 hours, with an on-demand sync when you want the forecast current to the minute. Each sync pulls reconciled balances and cleared data from your accounting platform, so the forecast is built from your ledger rather than from a copy of it, and the rekeying errors and stale snapshots that quietly degrade spreadsheet forecasts do not arise. On the timing side, every unpaid invoice and bill is mapped to the date you actually expect it to be paid, not just its due date, and you can update expected dates or split an invoice into part payments as reality unfolds.

For the assumption-heavy end of the window, Float's scenario planning lets you build unlimited what-if versions alongside your base forecast, so the uncertainty in weeks nine to thirteen becomes a set of concrete cases you can compare: the plan if the big invoice lands on time, and the plan if it slips a month. A cash threshold limit tracks the date you would cross your minimum balance under each version. The 13-week rolling view sits alongside a monthly forecast extending to three years, so the operational window and the planning horizon read from the same live data.

What Float does not do is manufacture certainty. It makes sure the uncertainty you are planning around is genuine, and gives you the cases to plan around it. If you want to see the build process end to end, our 13-week implementation guide walks through it, and the product overview shows the forecast itself.

Frequently asked questions

How accurate are 13-week cash flow forecasts for planning purposes?

Accurate enough for the decisions each part of the window carries: the first four weeks are built largely from committed invoices, bills and payroll, so they support cash-level operational decisions, while weeks nine to thirteen rest on assumptions and support directional planning rather than precise figures. Commonly quoted accuracy percentages are targets rather than measured benchmarks, so the dependable measure is your own weekly variance between forecast and actuals.

Is there a reliable industry benchmark for cash flow forecast accuracy?

No. There is no published study measuring 13-week forecast accuracy against realised cash for businesses of this size, in the UK or elsewhere. The commonly cited gradient of roughly 90 to 95 per cent in the first four weeks declining to 70 to 85 per cent by weeks nine to thirteen is a set of targets rather than measured performance, and the versions in circulation disagree with each other. Your own tracked variance is a sounder benchmark than any published number.

What variance do lenders allow on a 13-week cash flow forecast?

In restructuring situations where a lender requires a rolling 13-week forecast, executed facility agreements commonly permit variance in the region of 10 to 20 per cent, with 15 per cent the most frequently recurring figure. The testing method matters as much as the number: variance is generally accumulated over a rolling four-week period rather than judged week by week, only adverse variance is counted, and receipts and disbursements are tested separately rather than as a net closing balance. These tolerances come from distressed, mostly US, mostly larger-company situations, so treat the percentages as context rather than as a standard for a healthy business of 11 to 50 people.

Why are the first four weeks of a 13-week forecast the most accurate?

Because they are built from cash events that already exist: invoices already raised, bills already approved, payroll, rent and tax amounts already known. The main remaining uncertainty is payment timing, which is bounded by due dates and observed customer behaviour. From week five onwards, a growing share of the forecast depends on invoices not yet raised and sales not yet won.

What causes a 13-week cash flow forecast to lose accuracy?

The two dominant causes are customer payment timing, because the gap between due date and actual payment date compounds across the window, and pipeline assumptions, because forecast revenue converts later or smaller than planned. Stale or rekeyed input data and missing irregular outflows, such as VAT quarters and annual renewals, add error that has nothing to do with prediction at all.

Should I measure forecast accuracy as a percentage?

Not always. Percentage variance works when the figure you are dividing by is substantial, but in a week where receipts and payments nearly cancel each other out, net cash movement can be close to zero and the percentage becomes unstable and unhelpful. In those weeks, measure the variance as an absolute amount, and compare receipts and payments separately rather than netting them together first.

How do I measure the accuracy of my own forecast?

Save a snapshot of the forecast each week before rolling it, then compare actual receipts and payments from your accounting platform against what that snapshot predicted, keeping inflows and outflows separate. Accumulate the result over a rolling four weeks rather than reacting to a single week, and track the variance by week position across a month or so of cycles. That tells you your real accuracy at each point in the window.

How often should a 13-week cash flow forecast be updated?

Weekly. A weekly cadence rolls a new week thirteen into the window as the completed week drops out, and it feeds each week's actual results back into the assumptions, so the forecast corrects itself continuously. A 13-week forecast updated monthly spends most of its life out of date.

Does syncing accounting data automatically improve forecast accuracy?

It removes one whole category of error: the drift and transcription mistakes that come from building a forecast on stale exports and manual rekeying. Float, for example, imports invoices, bills and reconciled bank data from Xero or QuickBooks Online every 24 hours over a read-only connection. What automation cannot do is improve the quality of your assumptions about future sales or payment behaviour, which is where later-week accuracy is decided.

Who can see and edit our forecast in Float?

Float uses role-based permissions with three roles: Admin, Editor and Viewer, so the finance lead controls who can change the forecast and who can only read it. The connection to your accounting platform is one-way and read-only, meaning Float can never alter anything in Xero or QuickBooks Online, and two-factor authentication is mandatory for Xero users.

How much does a tool like Float cost?

Pricing depends on the size of your business and the number of entities you consolidate, and current plans are listed on our pricing page. Every plan starts with a 14-day free trial, and connecting your accounting platform takes about three minutes.

Your forecast's accuracy is knowable within a month of tracking it, and improvable from the first week. Start a 14-day free trial of Float and see your next thirteen weeks built from your own Xero or QuickBooks Online data, with the first sync running in about three minutes.