forecasting ROAS before launch - my method

harper-local

Pre-launch forecasting is mostly educated guessing dressed up in a spreadsheet. I've been doing local lead gen campaigns for years, and here's what actually works. Start with the revenue goal, then work backwards using realistic conversion rates by channel from historical data or industry benchmarks. If a client hit 3.4x last time, the most defensible forecast is 'somewhere around there again, but here's what we're changing and why it should move the needle.' Be transparent that these are assumptions, not guarantees. Clients who demand a guaranteed ROAS number before launch are setting you up for an awkward conversation later.

I also tap into Salesforce historical data from the last 3 years - segment by channel, product, region. For local, foot traffic attribution is key. I adjust for budget changes YoY, market conditions, seasonality. It's all educated guessing, but the more data you have, the better. And yeah, Claude just gave me a random number based on vibes - AI can't replace the experience of working with real campaign data.

bouncekiller

Pre-launch forecasting is basically educated guessing with a fancy spreadsheet costume. If they managed 3.4x last time, the safest forecast is "around there again, but here's what's different and why we think it shifts things." Start from their revenue target, plug in realistic conversion rates per channel from historical data or benchmarks, and be crystal clear these are assumptions, not promises. Any client demanding a guaranteed ROAS before launch is basically setting you up for a painful chat down the line.

logan-lifecycle

This can be complex and lot of it depends on what historical as well as forecasting data your organization has available. This is what I used to do: I targeted campaigns by target industry/markets, featured product and region. Then I tapped into Salesforce historical data for the last 3 years and created reports of campaigns converted to pipeline generation (or whatever conversion metric you track) and segmented by channel, product, and market. In my last organization targeted market was big driver for forecasting so I had a lot of data to work with here and used that to create a forecast. If my quarterly target is $5M and I see that the previous quarter, campaigns in that market were performing 30% of the target than I would use that % to forecast per campaign. Keeping in mind factors that may fluctuate the results like budget (has it increased, decreased YoY), market conditions (like tariffs or cost of goods), etc, and adjusting to realistically take into consideration these factors. I would say that forecasting is making educated guesses on how a campaign would perform and the more data you have, the better informed those predictions would be.

adcraft

This can be too complex to answer here, and a lot depends on the data you have (e.g., time series, data for all the proper variables in model specification).

For example, I may already know or have control over some predictors like the amount of money to be invested over time, the expected inertia (how much past revenue predicts future revenue), and the pricing and distribution strategies specifically for the new campaign.

I may have predictions from teams in finance and economics about GDP and environmental factors.

There are usually variables for seasonality, even for things like Holt-Winters.

We can even have something closer to a synthetic control, so we "synthesize" (create an artifical) version of the future campaign that is a combination of forecasts of other campaigns, even from competitors.

If I started with the goal in mind, I may have to work backwards with my regression results to make decisions to achieve that goal (e.g., how much I should invest in advertising, how frequently, and how often).

Working with different scenarios is also common. Can we still expect to achieve the goal under a pessimistic scenario?

And yeah, this is an example of how AI can't do what I do. AI helps with coding to run my analytics, not much more beyond that.

digitalnomad

Honestly, you don't forecast ROAS at the start. You test first to uncover winners - and those tests need to be research-backed, not just random stabs in the dark.

Once you've found what works, you can shift them into a dedicated winners campaign and start forecasting revenue from there with some actual confidence.

If there's existing data, you forecast based on those proven performers. But there are a few things I'd need to know:

🔹 Is it the same product or a new launch?
🔹 Any historical performance from similar campaigns?
🔹 What's the audience overlap?

Without those details, any forecast is just a wish dressed up in a spreadsheet.

brandvoice

Pre-launch forecasting is basically educated guesswork dressed up in a spreadsheet - and I think the most honest thing you can do is own that from the start.

start with what you already know: last campaign's channel-level ROAS, average CTR, and conversion rate by funnel stage. Then build a bottom-up model per tactic. Something like: "If we drop £X on paid search at a historical CPC of £Y and a CVR of Z%, that gives us N conversions at an average order value of £A." Add those up and see if you land anywhere near 4x. If you don't, you either shift spend toward better-performing channels, improve conversion somewhere in the funnel, or gently reset expectations.

The real value isn't the number you land on - it's showing the client the assumptions behind it. so when you end up at 3.6x instead of 4x, the conversation becomes "which assumption shifted?" instead of "you clearly don't know what you're doing." That honesty builds trust way more than a perfect forecast ever will.

zoecreative

Forecasting without launch data is basically informed guesswork, but the framework you've laid out is spot on. I break it down the same way: map spend per channel, plug in expected CPC/CPM and CTR based on historical benchmarks (or worst-case if it's a new client), then multiply by conversion rate and AOV. That gives you a raw revenue estimate per tactic.

The weighting part is where the art meets the numbers. Retargeting nearly always outperforms prospecting, so I'll weight that forecast higher, then roll everything up and stress-test against the 4x goal. If it doesn't land, reallocate budget until the blended ROAS looks plausible. No magic, just maths and a healthy dose of realism.

harper-local

Exactly this. If you've got the same product and clean data from last Q4, you're already ahead of the game. Pull the conversion rates, average order value, and cost per acquisition by week. factor in any market shifts-higher CPCs, new competitors-and adjust accordingly. that's your baseline. Build in a 10-15% buffer for surprises, then track daily against it. no guesswork needed