New

Peaky raised €1.1M, led by JK Invest

→

All articles

Company

/

Written by Peaky

How to build a driver-based forecast in under a day.

1 day
From Excel upload to live model
1 day
From Excel upload to live model

Most financial forecasts at scaling companies are not driver-based. They are spreadsheet-based.

The distinction matters. A spreadsheet-based forecast is a collection of hardcoded numbers connected by formulas. When something changes, you have to find the cells, update the values, and validate that the downstream math still works. A driver-based forecast is a model where the numbers are derived from a small set of underlying drivers. When something changes, you update the driver, and the model rebuilds itself.

If you have ever spent four days rebuilding a model after the close, you have lived the difference. This post is about building the kind of forecast that does not require those four days.

Why most teams don't have one

The reason most teams do not have a true driver-based forecast is not lack of skill. It is that the work of building one historically took weeks, and the work of maintaining one took more discipline than most finance functions had the bandwidth for.

In the era of spreadsheets, driver-based modelling was an aspirational exercise. People talked about it. Few teams actually committed to it because the cost was too high.

What has changed is the tooling. AI agents can do the mechanical work of building and maintaining a driver-based forecast in hours instead of weeks. The practical barrier is gone. What remains is the discipline question: do you actually want one.

What driver-based means

A driver-based forecast has three layers.

The driver layer. The small set of inputs that determine everything else. For a SaaS company, this is typically things like number of AEs, pipeline conversion rate, average deal size, churn rate, hiring plan, key vendor costs. The drivers are the assumptions you actually decide on.

The formula layer. The math that turns drivers into outputs. Revenue equals AEs times productivity times pricing. Burn equals headcount times average compensation plus non-headcount costs. The formulas are constants. Once they are set, they do not change unless the business model changes.

The output layer. The numbers everyone else cares about. Revenue, burn, runway, cash position, EBITDA, margin. These are not assumptions. They are consequences of the drivers and the formulas.

When the forecast is driver-based, you only ever change the drivers. The formulas hold. The outputs recompute. The model stays internally consistent.

When the forecast is spreadsheet-based, you change whatever cell is in front of you and hope the downstream math still works.

The six-step build

Here is the actual build sequence. It can be done in under a day with the right tools.

Step 1: Name the drivers (1 hour)

Sit down with a blank page. Write down the inputs that determine your business outcomes. For most SaaS companies, the list is between 15 and 25 drivers. Not 200. Not 5. The number that fits on a single page.

Revenue drivers: number of sellers, productivity per seller (in net new ARR), churn rate, expansion rate, pricing realisation.

Cost drivers: headcount by function, average compensation by function, planned hires by quarter, non-headcount opex by category, key vendor contracts, marketing spend by channel.

If you cannot fit your driver list on one page, you have too many. Reduce.

Step 2: Name the formulas (2 hours)

For each output you care about, define the formula in plain English. Revenue equals sellers times productivity minus churn plus expansion. Burn equals headcount times compensation plus opex. Cash position equals starting cash plus revenue collected minus burn paid.

Write these out as actual sentences before you put them into a model. The exercise will surface inconsistencies in your mental model.

Step 3: Ground the drivers in reality (2 hours)

For each driver, pull the latest actual data. What is your current churn rate. What is your actual productivity per seller. What is your hiring pace. What is your real CAC.

The temptation is to skip this and use targets instead of actuals. Resist. A driver-based forecast that is grounded in target numbers is a wish list, not a plan. Ground it in what is actually true today.

Step 4: Build the model (2 hours, with AI agents)

This is where modern tooling changes the math. In Excel, this step is two weeks. In Peaky, with an agent doing the building, this step is two hours.

The agent reads the drivers and formulas you defined. It builds the model. It maps each driver to its source data in your operational systems. It generates the outputs.

You review. You correct. You adjust. The model is functional by the end of step 4.

Step 5: Validate against the last closed period (1 hour)

Run the model against the last closed month. Compare the model's output to what actually happened. The reconciliation should be close. If it is not, the drivers are wrong, the formulas are wrong, or both.

This step is the difference between a model you can trust and a model that produces plausible-looking numbers that nobody can defend.

Step 6: Connect the drivers to live data (1 hour)

Each driver should be linked to its source system. CAC reads from your ad platform. Headcount reads from your HR system. Churn reads from your billing platform. Pipeline reads from your CRM.

Once the connection is in place, the drivers update automatically as the underlying data refreshes. You do not have to rebuild the model again.

Total elapsed time: under a day with the right tooling. Two weeks with Excel. Two months at a company that has never done it before.

The way the model stays alive

A driver-based forecast that is connected to live data does not need to be rebuilt monthly. It updates itself.

What you do instead is review signals. When a driver drifts outside tolerance, you see it. You decide whether the right move is to update the assumption, change the underlying reality, or wait for more data.

The work of finance moves from rebuilding to deciding. This is the whole point of driver-based forecasting at scale.

Common mistakes to avoid

Three mistakes are worth flagging.

Too many drivers. If your driver list does not fit on a page, you have too many. The point of a driver-based model is that it is comprehensible. A 200-driver model is just a spreadsheet in disguise.

Drivers that are actually outputs. Revenue is not a driver. Revenue is an output of (sellers x productivity) minus churn. If you treat revenue as a driver, you have not built a model. You have built a wish list.

Targets disguised as drivers. The plan should not start with "we will hit 30% growth". It should start with "given our productivity, hiring plan, and churn, this is what growth would look like". If 30% is not what falls out, the plan needs to change, not the model.

What to do next

If you have not yet built a true driver-based forecast, the question is not whether to do it. It is when. The cost of waiting compounds the same way drift does.

Block a day. Use the six steps above. Use AI tooling to do the mechanical work so you can focus on the decisions about drivers and formulas.

If you want to do the build in Peaky, you can start a trial at peaky.ai. You will have the model running by the end of the day.

The financial infrastructure for scaling without surprises.

Previous

Turning your ledger into a live financial model.

Next

Why we founded Peaky.

Keep exploring

Getting started

From your Excel plan to a live plan in a day.

Integrations

The systems Peaky connects to, and what each one brings in.

The Forecast Audit

Nine questions to find out whether your forecast is a report or a tool.

Peaky app icon

Sync or sink.

Ask AI for a summary of this website