How do we forecast and steer a €1B business?

This year, Alan will become a €1B+ annual revenue company. I've spent the last two years as the person responsible for forecasting, at almost any given moment, where we'll end up by December 31st.

What I've learned is this: forecasting a complex business isn't a science. It isn't really an art either. It's more like navigation. You're moving fast, visibility is limited, and being wrong by a little is fine, but being wrong by a lot has consequences.

The "Landing" is what we call our main forecasting process. The name gives it away: the goal is to estimate, months in advance, where we'll land at year-end (in terms of members, revenue, and gross profit). Like any landing, smooth is better than rough. In this context, "smooth" means maintaining a stable, accurate estimate over time, updating it regularly without wild swings.

Before explaining how we approach it, it's worth being honest about why forecasting is so difficult: it lives in the intersection several uncomfortable realities:

And then there is the compounding complexity problem. The more a business grows, the more moving parts it accumulates.

At Alan, we have a particularly rich version of this challenge. Our forecast needs to cover:

Four countries (for now!): France, Belgium, Spain, and Canada, each with different market dynamics, regulatory environments, and maturity stages.

Multiple ARR dynamics, each with its own logic and behavior:

Multiple product lines: health insurance, prévoyance (disability & life insurance), occupational health, each with different unit economics and different forecast mechanics.

Multiply the dimensions and you get a lot of cells in a lot of spreadsheets. The temptation is to let the model grow with the business. We've resisted it, deliberately.

There is a coordinator for the Landing. That's me. But I would describe my role less as a pilot and more as an air traffic controller. I only “fly” a few of the streams, but I have visibility on all of them.

Each dynamic has a dedicated owner, someone deeply embedded in that area of the business. The churn number isn't something estimated from the outside. It comes from the person closest to the data, accountable for it.

This matters for two reasons.

This works because of our radical transparency: everyone sees everyone else's numbers, can understand how they are estimated and where they comes from.

My role concretely as coordinator is to orchestrate the “art” behind this balance of numbers. I work hand-in-hand with the owner of each stream to identify the pockets of growth that are conservative and the ones that are ambitious. As we gather further signal, I manage the trade-offs across the whole model—for example, increasing new business expectations if we gain strong signals from a specific segment, while adjusting repricing assumptions based on inflation trends.

I am able to do this because I have built context over time in two ways:

The boring truth is that forecasting quality is mostly determined before the process starts: a forecast is only as good as the data it's built on.

This matters because it means we spend our time interpreting numbers, not debating whether the numbers are right. The conversation becomes "why are we beating targets on repricing?” rather than "wait, what's included in this repricing figure?"

This is perhaps the most underrated part of our approach.

When it comes to sales-driven new business, which is one of the buckets that can create the most volatility in the Landing, our sales teams use a detailed forecast methodology year-round to steer their activity. Concretely, this means:

A good forecast isn't just an accurate number at the end of the year. It's an early warning system.

If a dynamic starts moving in the wrong direction (for example new business behind pace in one channel), a well-maintained forecast surfaces that signal early, while there's still time to act. That's the real value: not the number itself, but the reaction time it creates.

It also feeds strategy: our multi-year plan, our investor materials, our headcount model. And perhaps most importantly, it becomes a shared language. When people know the numbers are well-owned and trustworthy, they make decisions faster, with less internal negotiation.

This is not a solved problem.

Forecasting a business like Alan still involves real uncertainty, and we've had our share of surprises. What we've found is that uncertainty is manageable, as long as the process is rigorous. You can't eliminate forecasting error. You can ensure you detect it early, u

nderstand where it's coming from, and update accordingly.

There's also an ongoing tension between model simplicity and reality's complexity. The right answer isn't to model everything. It's to model the things that matter most, and resist the temptation to add detail that adds noise rather than signal.

The Landing will keep getting more complex as Alan grows. Our job is to make sure the process doesn't grow with it.