Why do so many seed-stage pitch decks fall apart on slide 11? Because that's where the sales forecast lives, and the numbers are usually pulled from thin air. Anyone can type "£40k MRR by month 12" into a spreadsheet. The problem is defending it when someone asks which assumptions got you there.
Building a sales forecast for a new startup with no revenue history is genuinely hard. You cannot extrapolate from last year, because there is no last year. So you build from the ground up, from first principles, and you show your working. This guide walks through how to create a sales forecast you can actually defend—with the formula, the data sources, the revision cadence, and the honest limits of what a model can predict.
Key takeaways
- A startup forecast is bottom-up: leads × conversion rate × average deal size, not a top-down market-share guess.
- Before you have revenue, generate proxy data fast: pre-sales, a waitlist, paid pilots, a founder-run sales sprint.
- Model three scenarios (worst, most likely, best) and update the forecast on a fixed cadence, not when you remember.
- ChatGPT and similar tools can help structure assumptions and stress-test logic—they cannot generate the conversion rates for you.
- Investors and lenders expect roughly three years of forecast in a business plan, with month-by-month detail for the first twelve.
How to create a sales forecast when you have zero sales history
The standard formula is unforgiving in its simplicity:
Sales forecast = number of opportunities × conversion rate × average deal value
Every honest forecast reduces to those three inputs. Everything else—the fancy dashboard, the cohort charts—is presentation. Get the three inputs wrong and the output is fiction with a decimal point.
The bottom-up method, step by step
Start with the smallest unit of demand you can observe. Not the total addressable market (that number is for the investor narrative, not the forecast). Take one channel—say, outbound email to a specific segment—and estimate how many qualified leads you can generate per week.
Here's a real trap I walked into on a B2B tool I helped launch: I estimated 40 qualified leads per month from cold outreach because that's what a blog post told me was "typical." After four weeks of founder-led sending, the actual number was 11. We had booked a forecast on a number we had never tested. The lesson stuck: never forecast a channel you have not personally worked for at least three weeks.
Where the conversion rate actually comes from
For a pre-revenue startup, conversion rates come from three legitimate places:
- Your own early tests—even 20 outbound emails and 2 replies gives you a starting rate.
- Direct competitor conversations—ask potential customers what they currently pay for and what triggers a purchase.
- Adjacent benchmarks you can defend—if you sell a B2B SaaS product at a similar price point to a segment you've worked in before, your prior experience is a legitimate anchor.
What you must not do is invent a number and call it a benchmark. That's the fastest way to get your forecast dismissed.
How to forecast sales for a new product with no history
You need data. If the market hasn't given it to you yet, go create it. This is the single biggest gap in most startup forecasting advice: everyone tells you to "use market benchmarks," nobody tells you how to generate your own signals in six weeks.
Four ways to generate real demand data fast
- Pre-sales or pre-orders. If the product allows it, take money before you build. A deposit converts intent into evidence.
- Founder-run sales sprint. Block two weeks, personally pitch 30 prospects, log every objection. Thirty calls give you a defensible conversion rate to within a rough margin.
- Paid pilot cohort. Charge a small fee for early access. Free users lie; paying users don't.
- Waitlist with a payment gate. A waitlist of 500 emails is worth almost nothing. A waitlist where 40 people put down £20 is worth a forecast.
How much data is enough? A rough rule I use: 20–30 real conversations with your target buyer before you trust a conversion rate. Fewer than that and you're guessing with extra steps.
Triangulate, then discount
Take your three input estimates and deliberately shade them pessimistic. If your sprint produced a 12% reply-to-meeting rate, forecast on 8%. If your average deal value looked like £4,000, model £3,200. The forecast you present should be the one you'd still hit if things go moderately wrong. Optimism belongs in the "best case" column, not the base.
Can ChatGPT do forecasting?
Partially. It can structure your assumptions into a clean model, pressure-test the logic, write the narrative around the numbers, and generate a formula skeleton you paste into a spreadsheet. What it cannot do is hand you a credible conversion rate for your specific market at your specific price point—and any rate it invents is exactly the kind of fabricated figure that gets a forecast torn apart in a diligence call.
Where it genuinely helps: asking "what are the three weakest assumptions in this model?" or "what would make this forecast 50% too optimistic?" runs that a tired founder on a Friday evening often skips. I use it as a red-team tool, not an oracle.
What to ask it instead
Feed it your actual inputs—your lead volume from the sprint, your observed reply rate, your price—and ask it to walk through the sensitivities. Which input, if it moves by 20%, breaks the whole plan? That answer is worth more than any generated number.
How do I create a financial forecast for my start-up?
The sales forecast is one slice of the financial forecast. The financial forecast adds costs, cash flow, and runway. The order matters: build sales first, because headcount and burn depend on it, not the reverse.
Lay out a table like this—one row per month for the first twelve, then quarterly for years two and three:
| Line item | Month 1 | Month 3 | Month 6 | Month 12 |
|---|---|---|---|---|
| Qualified leads | 11 | 18 | 30 | 45 |
| Conversion rate | 9% | 11% | 12% | 13% |
| New customers | 1 | 2 | 4 | 6 |
| Avg. deal value | £3,200 | £3,300 | £3,400 | £3,500 |
| New revenue | £3,200 | £6,600 | £13,600 | £21,000 |
Keep every assumption in its own row. When reality diverges—and it will—you can see exactly which row moved. A lumped-together revenue figure tells you nothing about why it was wrong.
What are the 7 steps to forecasting?
Distilled version, in order:
- Define the period and granularity (monthly for year one).
- List your revenue streams—one line each.
- Choose the forecasting method: bottom-up for a startup, historical-based once you have 12+ months of data.
- Establish each input from real observation, not memory.
- Multiply through to a base case.
- Build worst and best scenarios around it.
- Set a review trigger and stick to it.
How often should you update the forecast?
"Regularly" is useless advice. Set a cadence: monthly comparison against actuals, with a full model rebuild every quarter. Add one trigger—if any single input is off by more than 20% versus actuals, rebuild immediately rather than waiting for the month-end.
This matters more than getting the initial numbers right. A forecast is a living assumption stack, not a document. The startups that survive are the ones that noticed in week three that their conversion rate had halved, not the ones that held a beautiful spreadsheet until the runway ran out.
Should I trust my first forecast?
No. Treat version one as a hypothesis. I have never seen a first-pass startup forecast land within 30% of reality—mine included, more than once. What makes it valuable is the discipline of writing down your assumptions so you can be proven wrong quickly and cheaply.
The founders who model carefully don't predict better than everyone else. They just find out sooner.
