Base Rates and Reference-Class Forecasting
Ask any marketer to predict a launch and they will build a bottom-up story: features, positioning, target audience, expected conversion. Ask an unrelated third party the same question and they will ask a smarter one: "how often do launches like this hit their goal?" That question is the base rate, and it is usually a better predictor than the story.
Quick Summary
- The base rate is the historical frequency of an outcome across all cases like yours.
- Human intuition ignores it: we default to the "inside view" (this project's specific details) and skip the "outside view" (how often projects like this succeed).
- Reference-class forecasting: predict by starting from the base rate, then adjust for what makes your case different.
- Adopted correctly, it cuts overconfidence by 30-50% on average in academic studies of forecasting.
What It Actually Is
Bing Experimentation Platform and digital feature rollouts Calibrating team expectations on campaign and feature win rates against empirical historical baselines Analyzed outcomes across tens of thousands of controlled experiments to establish the objective outside-view baseline for growth hypotheses
Result: Established an empirical 10%–20% feature win rate base rate, preventing overconfident budget allocation on untested intuition (2015 – 2020).
SourceDaniel Kahneman and Amos Tversky named the two views in their Nobel-winning work on judgment. The inside view builds a prediction from the specifics of your project. The outside view places your project in a reference class of similar projects and starts from that class's actual track record.
Human brains almost always start with the inside view. It is more vivid, more familiar, and makes us feel like experts. Unfortunately, the outside view is more accurate almost every time.
A marketer predicting "this webinar will get 500 signups" from the inside view thinks about the topic, the speaker, the promotion plan. A marketer predicting from the outside view asks: what did our last 15 webinars average? That number is a better starting point than any story.
Why It Matters
The forecasting literature is unusually clean on this. Bent Flyvbjerg's studies of megaproject budgets found that reference-class forecasting reduced overrun rates significantly compared to bottom-up estimates. Kahneman describes the inside view as "the single most important source of error in most forecasts."
In marketing, the same failure mode shows up as overconfident goal-setting. Teams promise pipeline numbers based on the story of the campaign, not the historical hit rate of campaigns like it. When the number misses, the retro asks "what went wrong with execution?" when the honest answer is "the goal was set from the wrong reference point."
Base rates are also how the professionals bet. Every good investor, poker player, and epidemiologist starts from the base rate and adjusts. Marketers are the last team in most companies still forecasting inside-view first.
The Playbook: Four Steps
Strategic enterprise planning and new business launch forecasting Eliminating executive optimism bias in multi-million dollar product and campaign projections Implemented Kahneman & Lovallo's 4-step reference-class forecasting model by identifying comparable past initiative classes, computing base-rate distributions, and constraining upward adjustments
Result: 30% to 50% reduction in forecasting error and budget overruns compared to traditional bottom-up inside-view models (2003).
SourceStep 1: Pick a reference class. The class should be broad enough to have data (at least 5 to 10 cases) and narrow enough to be comparable (same channel, same customer type, same rough scale).
Step 2: Get the base rate. Compute the mean and range of the class. "Launches like this land between 300 and 900 signups; the average is 550."
Step 3: Adjust for real differences. Only for factors that actually differed across the reference class. "Our audience is 2x larger than the average" is a real adjustment. "This one feels more exciting" is not.
Step 4: Write it down. Include the prediction, the reference class you used, and your confidence range. Score it after the fact. Calibration only improves when it is measurable.
A Real Example
In 2023, an analytics-tools company was setting a Q1 goal for a new product launch. The team's inside-view forecast was 2,500 signups in 30 days, based on the launch plan, the audience, and the influencer list.
Their previous eight product-line launches had averaged 720 signups in the same window, with a range of 400 to 1,100. The team was predicting 3.5x the historical top of their own reference class, and no one had asked why.
They rebuilt the forecast from the base rate. Starting at 720, they adjusted upward for a larger audience list (+30%), better positioning research (+10%), and an influencer partnership (+15%), landing at ~1,180 signups. That became the plan-of-record.
Actual result at day 30: 1,240 signups. The base-rate forecast was accurate; the inside-view one would have been off by 100%.
When you have no internal data, look for public reference classes. Industry newsletters, State-of reports, and case study collections publish enough numbers that you can usually find a comparable class within a few searches. A messy public benchmark is a better anchor than a confident guess.
Common Mistakes
Mistake 1: Reference class of one. "Last year's launch did 400 signups, so we will do 400 signups." One case is not a class. Widen the sample even if it means slightly less comparable cases.
Mistake 2: Adjusting for hopes, not differences. Every launch feels special from the inside. The only legitimate adjustments are for factors that measurably differ from the class average.
Mistake 3: Skipping calibration. If you never score your predictions, you never learn whether your adjustments are helping or hurting. Predictions without scoring are entertainment.
Kahneman's warning applies here: "the inside view is the tempting one, and it is the wrong one." When the base rate says one thing and the story says another, default to the base rate and require evidence to override it, not the reverse.
Key Takeaways
- The base rate is how often outcomes like yours have actually happened; it beats bottom-up stories most of the time.
- Reference-class forecasting: pick the class, get the base rate, adjust only for real differences, write it down.
- Use the outside view first, the inside view second. Reverse the order most marketers use.
- One case is not a class; widen the sample even at the cost of tight comparability.
- Track prediction accuracy over time. Calibration compounds; guessing does not.