Every forecast rests on a set of assumptions. Revenue growth depends on win rates, pricing, and market volume. Headcount cost depends on hire timing, salary levels, and attrition. Gross margin depends on input costs, supplier terms, and product mix.
Most finance teams know their key assumptions. Few of them have a systematic framework for documenting those assumptions, tracking when they change, understanding the sensitivity of the forecast to each one, and communicating the assumption set clearly to board and management.
The result: forecasts that are technically correct but fragile under challenge. When a board member asks what happens to EBITDA if win rates drop 5 percentage points, the answer requires rebuilding the model manually rather than reading the answer from a pre-built sensitivity. When an assumption changes, it is updated in one tab but not in the supporting narrative. The forecast model and the story about the forecast drift apart.
What a Driver Tree Actually Is
A driver tree is a hierarchical decomposition of a financial output into its underlying drivers. Starting from a summary metric which includes revenue, EBITDA, operating cash flow, the tree breaks the number down into its component parts, and each component part into the operational or commercial inputs that determine it.
A simple revenue driver tree
- Revenue
- New business revenue: pipeline volume x win rate x average contract value
- Renewal revenue: installed base x renewal rate x average renewal value
- Expansion revenue: installed base x expansion rate x average expansion value
Each leaf node in the tree is a named assumption: a specific number, a range, or a formula linked to an external data source. The owner of that assumption is responsible for reviewing and updating it each forecast cycle.
Why Most Driver Trees Break Down in Practice
- Assumptions are implicit in the model rather than documented as named parameters
- Multiple versions of the same assumption exist in different tabs with no clear hierarchy
- No one has ownership of specific assumptions, so they do not get reviewed systematically
- Sensitivities are calculated occasionally rather than maintained as a live feature of the model
- When business conditions change, assumptions are updated informally without updating the documentation or the sensitivity analysis
Where AI Strengthens Assumption Management
Assumption Registry
AI maintains a named assumption registry: every driver that feeds the forecast model, its current value, its owner, the last date it was reviewed, and the source data or rationale behind it. The registry is a live document, not a static tab that gets forgotten.
When a forecast is presented to the board or CFO, the assumption registry is the supporting document that answers the question: where did these numbers come from?
Automated Sensitivity Analysis
AI runs sensitivity tables against the driver tree automatically. For each named assumption, AI calculates the impact on the key financial outputs like revenue, EBITDA, cash flow if the assumption moves by a defined range. These sensitivity tables update whenever the base model updates.
When a board member asks what happens to EBITDA if renewal rates drop 3 percentage points, the answer is already in the model. The CFO reads it from the sensitivity table rather than commissioning a model rebuild.
Assumption vs Actual Tracking
AI compares forecast assumptions against emerging actuals each period. If the win rate assumption was 28% and the current quarter is tracking at 22%, AI flags the gap. The FP&A team decides whether to update the assumption, investigate the driver of underperformance, or hold the assumption and explain the variance.
This is the feedback loop that keeps the forecast model honest. Assumptions that are persistently wrong get challenged systematically rather than quietly maintained because no one looked at them.
Assumption Change Log
Every time an assumption is changed, AI records the prior value, the new value, the date, and the reason. The change log is available for management review, board questions, and audit purposes. When a CFO asks why the revenue forecast changed between the Q2 board pack and the Q3 board pack, the assumption change log provides the answer.
Building a Driver Tree That AI Can Maintain
Step 1: Identify the five to ten most sensitive assumptions
Not every driver in the model matters equally. Run an initial sensitivity analysis to identify which assumptions, if wrong by a plausible amount, would materially change the financial outcome. Focus documentation and review effort on those.
Step 2: Name every critical assumption explicitly
Every assumption should be a named cell or parameter in the model, not a hardcoded number embedded in a formula. A formula that reads "=revenue_growth_rate * prior_year_revenue" is maintainable. A formula that reads "=1.12 * C24" is not.
Step 3: Assign ownership
Each assumption should have an owner who is responsible for reviewing it each forecast cycle. Revenue assumptions owned by FP&A in partnership with sales. Headcount assumptions owned by FP&A in partnership with HR. Cost assumptions owned by FP&A in partnership with the relevant department heads.
Step 4: Connect to source data where possible
Assumptions that can be derived from source data like win rate from CRM, renewal rate from the customer success platform, hire timing from the HRIS should be linked to that data rather than manually entered. AI updates these automatically as source data changes.
What AI Cannot Do in Assumption Management
- Determine whether an assumption is right. AI can tell you that the win rate assumption is 28% and that it has been wrong for three consecutive quarters. It cannot tell you whether the right assumption is 22%, 25%, or whether the business needs to change its go-to-market approach to hit 28%. That is a commercial judgment.
- Resolve assumption conflicts between functions. When finance, sales, and the CEO disagree about the right revenue growth assumption, AI provides the data. The resolution is a conversation.
- Communicate uncertainty. The art of presenting forecast assumptions to a board — how much confidence to express, which sensitivities to highlight, how to frame range versus point estimate requires experienced judgment about what the audience needs to know.
Start Here
Take your current forecast model and identify every hardcoded number that represents an assumption about how the business will perform. List them. For each one, name it, identify who owns it, and calculate what happens to EBITDA if it moves 10% in either direction.
That exercise turning implicit assumptions into a named, ranked list is the foundation of a driver tree. It also reveals which assumptions are driving most of the forecast risk, which is where both AI monitoring and management attention should focus.
