Cash-flow forecasting when revenue is lumpy

Why standard forecasting fails for lumpy businesses

Monthly recurring revenue is a fintech concept. It describes subscription businesses where the same money arrives on the same date every month. For a construction firm, a specialist consultancy, a healthcare practice, or an events company in Australia, revenue does not work that way. Projects land irregularly. Retainers make up some of the base, but most of the income arrives in chunks: a deposit at signing, a progress payment mid-project, a final payment on completion. When those chunks fall in the same quarter, the business looks healthy. When they fall in different quarters, the same business looks like it is struggling.

A 12-month P&L projection handles this by averaging. If the business earns $480,000 per year, the model shows $40,000 per month. That number is useful for a lender assessing annual capacity. It is useless for a business owner managing the next 90 days.

Consider a professional services firm that earns $400,000 between July and September and $80,000 between January and March. The annual P&L looks fine. The owner feels comfortable through September, takes on an extra staff member in October, upgrades the office fitout in November, and then arrives at February with $80,000 in revenue, $120,000 in quarterly fixed costs, a BAS due, and super contributions overdue. The crises that hits in February was visible in September, if the right tool was in use. It was not visible in the annual P&L at all.

This is not a rare edge case. It is the normal operating rhythm for a large share of Australian SMBs. Construction and trades are seasonal. Professional services businesses cluster their project work around client budget cycles, which often align with financial year ends. Healthcare practices are quieter in December and January. Events businesses have dead quarters by definition. If your revenue is not flat across 12 months, a flat-month model is not your friend.

What a 13-week model does differently

A 13-week model works at weekly granularity across a rolling 91-day window. Every week you drop the oldest week off the front and add a new week at the back, so you always have three months of visibility. The rolling structure keeps the forecast grounded: the first four weeks are mostly confirmed, the middle four are probable, and the final five are directional. That layering is useful information. It tells you where your confidence ends.

The model uses three revenue buckets rather than a single income line.

The first bucket is committed revenue: money that is contractually due. Signed project contracts, active retainers, confirmed milestone payments. You know the week the payment is expected based on the invoice terms in the contract. This is the only revenue you count at 100 per cent.

The second bucket is likely revenue: projects in advanced negotiation, repeat clients you have a strong relationship with, proposals submitted and verbally accepted but not yet signed. Apply a 70 per cent probability weighting to everything in this bucket. If a prospect has told you the job is theirs to lose but has not sent back the signed agreement, that invoice does not go into committed. It goes into likely at 70 cents on the dollar. This discipline sounds pessimistic. It reflects reality. Deals that seem certain fall through more often than founders want to admit.

The third bucket is discretionary spend: expenses you can switch on or off within a few weeks. Contractor hours, software subscriptions that are nice to have, conference attendance, upgraded marketing spend. These are tagged in the model with a toggle. If the balance in week nine looks uncomfortable, you turn the discretionary items off for that period. This is not cutting corners. It is the operational lever that makes the model actionable rather than just informational.

Building the model in five steps

The five-step build process below takes two to three hours the first time. Once the structure is in place, the weekly update takes fifteen to twenty minutes.

Step 1: List every committed payment by week of receipt

Open your signed contracts and active retainer agreements. For each one, identify the payment date or the invoice trigger event. Log the expected receipt date, not the invoice date. If your standard terms are 14 days from invoice and you plan to invoice on the first of the month, the receipt lands around the fifteenth. Use that date. Do not use the invoice date. Cash flow is about cash in the bank, not invoices in the system.

Step 2: List every fixed outgoing by week

Payroll, rent, software subscriptions, loan repayments, insurance premiums, and any other committed expense. Log the week money actually leaves your account. Many businesses pay suppliers on 30-day terms, so the cash outflow is a month after the expense is incurred. Use the actual payment week.

Step 3: Add likely revenue at 70 per cent

Review your pipeline. For every opportunity in advanced negotiation or with a verbal agreement, estimate the payment date if it closes and add 70 per cent of the invoice value to the relevant week. Do not include early-stage prospects. The 70 per cent haircut is not negotiable. It exists precisely for the deals that almost closed.

Step 4: Add discretionary spend with a toggle

List all non-essential spending planned for the next 13 weeks. Mark each item clearly as discretionary in the model. When you calculate your running balance in step five, you will run the model twice: once with discretionary spend included and once without. The gap tells you your flexibility range.

Step 5: Calculate the running balance and flag danger weeks

Starting with your current bank balance, add each week's inflows and subtract each week's outflows. The result is your projected end-of-week balance. Flag any week where the projected balance falls below four times your average weekly operating cost. That is your danger threshold. Four weeks of operating cost is the minimum buffer for a business with lumpy revenue because a single delayed payment from a major client can absorb two to three weeks of runway instantly.

Below is a worked example using four weeks of data for a nine-person professional services firm in Brisbane with $1.1M annual revenue, including a BAS payment due in week three.

Week three shows a closing balance of $24,500 after the BAS payment. With weekly operating costs of approximately $9,200, the four-week buffer threshold is $36,800. The model flags week three as a watch period twelve days before it arrives. That is enough time to chase an outstanding invoice, draw down a line of credit temporarily, or defer the discretionary spend from week four. Without the model, you arrive at week three and react. With the model, you prepare.

Australian-specific inputs that most templates miss

Generic 13-week templates built for US or UK markets miss three inputs that are structural features of the Australian business environment. Each one is a predictable, large-value outgoing that concentrates in a specific week. If it is not in your model, your model is wrong by design.

BAS lodgment timing

Most Australian SMBs lodge quarterly. The standard lodgment deadlines are 28 February, 28 April, 28 July, and 28 October, with payment due on the same date unless you have an arrangement with the ATO. The GST component typically represents 10 per cent of your GST-inclusive revenue for the quarter, minus input tax credits. For a business with $275,000 in quarterly revenue and $55,000 in creditable acquisitions, the net GST liability is around $20,000. That leaves your account on the lodgment date. Map every quarter-end BAS payment into your model as a fixed outgoing in the relevant week. It is the single largest predictable drain that businesses consistently underestimate because it does not appear as a regular monthly line.

Superannuation contributions

The Superannuation Guarantee rate for 2024-25 is 11.5 per cent of ordinary time earnings per the ATO. Contributions are due quarterly: 28 October, 28 January, 28 April, and 28 July. For a business with five staff earning an average of $85,000 per year, the quarterly super obligation is approximately $12,244. That amount hits the bank account in the same week as payroll, with the super payment going out separately to the clearing house. Many businesses manage payroll without tracking super timing separately. Put it in the model as a distinct line item in the relevant week.

Annual leave loading

Under most modern awards and enterprise agreements, employees who take annual leave are entitled to a 17.5 per cent leave loading on top of their base rate. For a business with five staff each taking two weeks of leave across the December-January period, the leave loading component adds approximately $3,000 to $5,000 in payroll cost concentrated in the weeks when leave is taken. The December quarter is already lower-revenue for many businesses. Leave loading in the same period compounds the cash pressure. Log it by week when you know which staff are taking leave.

How AI changes the upkeep

The most common failure mode for a 13-week model is not building it wrong. It is building it correctly and then not updating it. A cash flow forecast that was accurate six weeks ago and has not been touched since is worse than no forecast, because it creates false confidence. The model needs to be refreshed every week with actual receipts, any new signed contracts, updated pipeline probabilities, and confirmed outgoing payment dates.

Done manually, that weekly update takes 20 to 30 minutes if your accounting system is well-maintained and your pipeline data is current. In practice, it takes longer because both of those conditions are rarely met simultaneously. The update gets skipped. The model drifts. Within four to six weeks it no longer reflects reality, and the owner stops trusting it. It sits in a folder and collects digital dust.

An AI layer connected to your accounting software changes this. Tools built on top of Xero or MYOB can pull confirmed invoice receipts automatically, update the committed revenue bucket in real time, flag invoices that are overdue and should have moved from committed to uncertain, and calculate the updated running balance without manual input. The weekly update that takes 25 minutes manually can be reduced to a five-minute review of what the system has already updated.

AI can also identify patterns in your receivables data that are not visible in a single forecast cycle. If a particular client consistently pays 12 days late, the model should reflect that. If your construction projects routinely have final invoices paid 8 weeks after practical completion rather than the contracted 30 days, the likely revenue bucket should apply a timing adjustment, not just a probability adjustment. These patterns exist in your historical data. AI surfaces them. A human updating a spreadsheet weekly does not have the bandwidth to identify them.

The result is a model that is always current, always reflects real payment behaviour, and requires less discipline to maintain. The strategic discipline is still yours: you decide which opportunities go into the likely bucket, you decide what counts as discretionary, you decide when to draw down a credit facility. The operational upkeep is automated.

The template and your next step

A 13-week model built from scratch in a blank spreadsheet takes a few hours and requires some care to structure correctly. The running balance formula, the discretionary toggle logic, and the danger-week flagging all need to be set up before the data goes in. Getting the structure wrong means the model produces numbers that look right but do not account for timing correctly.

The Bizkook 13-Week Cash Flow Template is a pre-built Google Sheet with the structure already in place: committed and likely revenue buckets with automatic probability weighting, fixed and discretionary outgoings separated, a running balance that recalculates automatically, and conditional formatting that flags any week where the projected balance falls below the four-week operating cost threshold. It also includes a tab for Australian-specific inputs with quarterly BAS and super payment date prompts built in.

You provide the numbers. The template provides the structure. The first version of your model should take under two hours to complete.

For businesses already using Xero or MYOB and interested in automating the weekly update, our AI consulting team can configure a connection that keeps the committed revenue bucket current without manual input. That typically reduces ongoing model maintenance from 20 minutes per week to a five-minute review.