The honest answer is a count of resolved outcomes, not a count of years. A forecast model learns from deals that finished. If your pipeline is long and your volume is low, three years of CRM history might contain fewer usable examples than eighteen months at a high-velocity company.
Count Outcomes, Not Calendar Time
Every open opportunity is a question the model has not answered yet. Only closed records teach it anything, and it needs both sides. Closed-won records alone tell the model what buying looks like without telling it what stalling looks like.
The requirement compounds with segmentation. If you forecast enterprise and SMB separately, each needs its own body of resolved outcomes. Split further by region, product line, and motion, and the history you thought was abundant thins out fast. Segment only as far as the data supports, then stop.
Seasonality Needs Repetition
One year of history shows a shape. Two years show whether the shape is seasonal or was a one-time event. Without a second cycle, a model cannot tell a recurring Q4 surge from a single large deal that happened to land in December.
ORM's read on the pattern is that Q2 and Q4 usually run stronger than Q1 and Q3, and the third month of a quarter runs stronger than the first and second. Most teams underweight this. A model that has not seen the pattern twice will underweight it too.
Consistency Beats Volume
More history is worse than less history when the definitions moved. A stage that meant one thing in 2024 and something else in 2025 injects noise the model reads as signal.
ORM's position on this is direct. Everyone believes their data is uniquely bad and that it prevents accurate forecasting. That belief is wrong. Garbage in does not have to equal garbage out. As long as the data is consistent, accurate predictions follow.
Before you extend a training window backward, check whether the fields carried the same meaning that far back. Cutting history at the last definition change usually beats keeping records the model will misread.
Forecasting While the History Is Thin
Thin history does not block a forecast. It changes what the forecast can claim.
Forecast at the total level rather than the segment level. Use driver-based logic, where you model lead volume, conversion, and cycle time explicitly, instead of asking a model to learn weights it has too few examples to estimate. Report a range and say what it depends on. Then backtest as outcomes accumulate and tighten the grain when the data earns it.
Do not compensate for missing history by leaning harder on pipeline coverage. Coverage is an input, and treating it as the conclusion is a well-documented way to miss, as covered in the 3x pipeline coverage rule is wrong. Build the forecast from sources of revenue instead, following the structure in how to create a sales forecast.
Frequently Asked Questions
How much history does a revenue forecast model need?
Enough closed-won and closed-lost outcomes for the pattern to repeat within every segment you intend to forecast separately, plus two or more full annual cycles if your business has seasonality. A model that has seen a pattern once has seen an anecdote.
Can you forecast revenue without years of CRM history?
Yes, at a coarser grain. With thin history you forecast the whole business instead of each segment separately, leaning on driver-based logic rather than learned weights and reporting a range instead of a point. Precision comes back as outcomes accumulate.
Is dirty historical data disqualifying?
No. ORM's position is that every company thinks its data is uniquely bad and it rarely blocks accurate prediction. Consistent data supports good forecasts even when it looks messy. A field whose definition changed midway through the history is the real problem.
How much history do you need to capture seasonality?
At least two complete annual cycles, because a single cycle cannot separate a seasonal effect from a trend. ORM observes that Q2 and Q4 typically run stronger than Q1 and Q3, and that the third month of a quarter runs stronger than the first and second.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like how much historical data do you need to forecast revenue? into prescriptive action for your team.
Schedule a Demo