You've got eighteen months of cash in the bank. Then a key hire quits, a big client delays payment by sixty days, and your cloud bill jumps because someone left a staging environment running since spring. Your runway didn't shrink because your forecast was wrong on paper. It shrank because the forecast never accounted for the boring stuff.
That's the gap most budget forecast advice for tech startups leaves open. Generic "12 steps to a startup budget" guides will tell you to project revenue and expenses. Useful, sure. But a SaaS company doesn't budget like a bakery, and a marketplace doesn't budget like a SaaS company. The mechanics of recurring revenue, the shape of your burn, and the timing of when money actually lands in your account all change what the forecast has to do.
Key takeaways
- Build the forecast bottom-up from real drivers (headcount, contracts, usage), not from a top-down percentage you hope to hit.
- Cash flow matters more than the P&L for a startup. Revenue recognized is not revenue collected.
- Model three scenarios and know which one triggers which decision—hiring freeze, spend cut, fundraise.
- Your burn multiple and runway are the two numbers investors will actually interrogate.
- Update the model monthly. A forecast you build once and never touch is decoration.
Why tech startup budgets break where ordinary budgets hold
Traditional budgeting assumes revenue arrives roughly when you bill for it. A consulting firm sends an invoice, gets paid in thirty days, and the forecast is basically the invoice schedule. Fine.
Software doesn't work that way. You sign a customer to an annual contract, recognize the revenue across twelve months, but the cash might come upfront or quarterly depending on how you negotiated. Meanwhile you're paying engineers every two weeks regardless of whether anyone paid you. That mismatch is where startups get surprised.
Cash is not revenue, and the forecast must show both
I learned this the hard way on a project where the P&L looked healthy for two quarters while the bank account kept dropping. The revenue was real. It was just sitting in accounts receivable because our largest customer had negotiated net-60 terms and treated them as optional. Nothing in the budget model flagged it, because the model tracked revenue, not collections.
So build two views side by side: an accrual view (when you earn it) and a cash view (when it hits the bank). They will diverge, and the divergence is the whole point. If you only build one, a delayed payment can wipe out a month of runway you thought you had.
Burn rate and runway should anchor everything
Net burn is what you spend minus what you bring in each month. Runway is cash divided by net burn. Simple math, and yet people fudge the inputs constantly—counting restricted cash as available, ignoring a tax payment due next quarter, forgetting the annual software renewal that hits in one lump.
Here's the thing: your runway is the number that decides your next move. If you have nine months, you start raising. If you have four, you're cutting. Build the forecast so that runway is visible at a glance, not buried in a spreadsheet tab nobody opens.
How do I forecast a budget? Start from drivers, not percentages
People often ask me some version of this, and my answer is always the same: build it bottom-up. Top-down ("we'll grow revenue 40% and keep costs flat") feels fast and is almost always fiction.
Step 1: list your actual cost drivers
For most tech startups those come down to a handful:
- Headcount—salaries, employer taxes, benefits, the recruiter fee when you hire
- Infrastructure—cloud, hosting, third-party APIs, per-seat tools billed monthly
- Sales and marketing—ad spend, events, the commission structure you promised
- G&A—legal, accounting, insurance, the office lease with its annual escalator
Build each line from a unit. Not "infrastructure: €4,000/month" but "€X per active customer per month times projected customers." When the customer count changes, the cost moves with it—which is what actually happens when you grow.
Step 2: forecast revenue from the model, not the hope
Recurring revenue needs to be forecast from cohort behavior, not a single growth rate. Take your existing customers, apply an assumed churn rate, then add new customers at a realistic monthly acquisition pace, then apply expansion. That gives you MRR that reflects how the machine actually turns.
One detail people miss: new customers don't arrive evenly. If your sales cycle is ninety days, a hire made in January doesn't produce revenue in February. Put the lag in the model.
How to forecast sales for a startup?
Forecast sales from your pipeline and conversion rates, not from a target you'd like to hit. Pull the number of qualified opportunities, the historical close rate at each stage, and the average contract value—then multiply. If you don't have enough history yet, use your best estimate of the sales cycle length and be explicit that it's a guess. A labeled assumption is honest; a silent one is a trap.
The lag nobody models
Sales forecasts usually fail not because the conversion rate is wrong but because the timing is. A deal you close in the last week of the quarter may not pay until well into the next one. Model the collection date, not the close date, and your cash forecast stops surprising you.
Stress-test with scenarios
Run three versions: base, pessimistic, and optimistic. The pessimistic case is not "half revenue"—it's "our biggest customer churns and one enterprise deal slips a quarter." Then write down what you'd do if that hit. Which expenses are cuttable on thirty days' notice? Which hires get delayed? Having the answer written down before the crisis is worth more than any forecast accuracy.
Can ChatGPT make me a budget?
It can help you build the skeleton—a spreadsheet structure, a list of line items, a draft scenario framework. What it cannot do is know your business. It doesn't know your churn rate, your average contract value, the contract terms your biggest customer negotiated, or the timing of your next funding round.
So use it as a drafting tool, not a source of truth. Feed it your real numbers if you're comfortable with that, treat its output as a starting template, and rebuild the logic yourself. The value of building the model is that you understand every line. If you outsource that understanding, you can't defend it in a board meeting—or adjust it when conditions change.
What is the typical budget for starting a startup business?
There's no honest universal number—it depends entirely on your model and how much of the build you can do yourself. What I can tell you is the shape of it. In the earliest phase, most of the spend goes to people (founders often take minimal salary) and tools. If you're building software, infrastructure costs are usually modest at low usage. If you're building hardware, your costs are heavily front-loaded into tooling, prototypes, and compliance testing before you've sold anything.
What varies most is the runway you need to reach a defensible milestone—a working product with real users, or the first paying customers. Aim to raise or set aside enough for that milestone plus a buffer, because the milestone will take longer than the plan says. It always does.
| Model | Where the money goes early | Forecast priority |
|---|---|---|
| SaaS | Engineering, cloud, sales hires | MRR, churn, CAC, burn multiple |
| Marketplace | Two-sided acquisition, take-rate economics | Liquidity, repeat rate, subsidy per transaction |
| Hardware | Tooling, prototypes, certification, inventory | Bill of materials, lead times, working capital |
Unit economics belong in the forecast, not in a separate deck
If you track CAC and LTV only in investor materials and not in the operating model, you're flying without instruments. Put them in. Customer acquisition cost divided against lifetime value tells you whether growth actually creates value or just burns cash—and it changes how you budget sales and marketing.
Same with the burn multiple: net burn divided by net new revenue. A high number means you're spending heavily to grow slowly. Investors look at this. So should you, monthly.
Review it monthly, or it's decoration
Build the forecast, then actually update it. Reconcile actuals against the plan every month. Where did you beat? Where did you blow through? What assumption was wrong? The reconciliation is how the model gets better, and how you catch the small deviations before they become a crisis.
And keep it simple enough that you'll actually do it. A model with forty tabs is a model you'll abandon by the third month.
The forecast that saves a startup isn't the one that predicted everything correctly. It's the one that told you, in plain terms, how much runway you had and what you'd do when it got short.