The point estimate and the false-certainty trap
Single numbers, predictable miss
The single-number forecast — “next month will be 9,400 contacts” — is the most common form of planning communication and one of the most credibility-corrosive. The actual almost never lands exactly on the point. Every miss is read as a forecast failure. Over a year the planning function accumulates a reputation for being “always off” even when the model was reasonable. The maths is brutal: a forecast that’s within 4% MAPE will, in any given month, be visibly “wrong” by some amount, and the audience that demanded the single number is the same audience that will judge it against the actual.
Why ranges build credibility that points corrode
A range — “9,100 to 9,700, most likely 9,400” — is more honest, more often hit, and quietly more persuasive over a year of conversations. Three things happen.
The actual lands inside the range most months. The planning team is right by construction. The audience stops reading every variance as a failure and starts reading it as the noise it actually is.
The big misses get visible upfront. A range that’s wider than the audience expected signals that the planner is genuinely uncertain about something, and the audience can plan around the uncertainty rather than being surprised by it.
The conversation with finance changes. “The forecast is 9,400” invites the question “what if it’s wrong?” A range answers the question without it being asked, and finance reads the planner as commercially literate rather than over-confident.
How to construct a useful prediction interval
Three pragmatic approaches.
Empirical bands from history. Look at the forecast errors of the last 12 months. The 5th to 95th percentile of those errors gives you a 90% empirical band. Apply it around the point forecast as a range. This requires no statistical machinery and works surprisingly well for stable operations.
Model-based intervals. Most statistical forecasting methods (Holt-Winters, ARIMA, Prophet) produce prediction intervals as part of their output. Use them. The interval is usually wider than planners expect — that’s the honest answer, not an artefact to be tuned out.
Scenario bracketing. Forecast a central case, an upside, and a downside, each with named assumptions. “Central case: campaign as planned. Upside: campaign launches a week early. Downside: regulatory event in week 4.” Each is a defensible forecast in its own right. The range emerges from the spread.
Handling the “finance only wants one number” objection
The most common pushback is that finance, or the leadership team, doesn’t want a range — they want a number to plan against. This objection is usually less true than it sounds.
Finance does need a number for the budget cell. They don’t need that number to pretend to be more certain than it is. The pragmatic answer: present the range, name the central case as the planning point, and explicitly state the conditions under which the upside or downside become more likely. Finance gets the budget number; the conversation around uncertainty stays honest; and when the actual lands somewhere other than the point, the planner has already framed the variance as expected rather than a failure.
The operations that adopt this discipline find finance starts citing the range itself in their own conversations. The audience learns to think in ranges when the planner consistently presents in ranges. Over a year, the credibility shift is substantial.
The presentation that lands
A range presentation has a structure that works.
State the central case as a single number. “Our central forecast for Q4 is 28,400 contacts a month.” Finance hears the planning number.
State the range. “The realistic range is 26,500 to 30,200.” Specific, not vague.
Name the drivers that move you within the range. “The main upside risk is the marketing push starting two weeks earlier than planned; the main downside is if regulatory event X lands in October.” The audience now knows what to watch for.
End with the planning assumption. “We’re building the schedule against 28,400 with the flex layer sized to absorb up to 30,200 without overtime.” The decision is concrete; the contingency is named.
Conclusion
The point estimate is the most common forecasting presentation and one of the most credibility-corrosive. The range forecast is more honest, more often right, and over a year quietly more trusted by finance. The discipline to adopt isn’t technically difficult — empirical bands from history work for most operations — but it requires the planner to give up the false confidence of a single number. Planners who make that trade find finance reads them differently within a quarter and treats their numbers differently within a year. Planners who don’t find every miss is replayed as a failure and credibility quietly drains.
Next in the series: Treating noise as signal (and signal as noise).
Pair this with forecasting with ranges, Poisson and natural noise, and your forecast is probably more accurate than you think.