A naive forecast predicts that the next period will equal the most recent actual. The seasonal naive version repeats the same period from last year. Neither is meant to be your forecast. They are the baseline every sales forecast has to beat: if your model or forecast call cannot beat a naive forecast on held-out quarters, it adds no value.
What Is a Naive Forecast?
A naive forecast uses the last known actual as the forecast for the next period, with no model, no adjustments and no judgment. If a SaaS company booked $1.6M of net new ARR last quarter, the naive forecast for this quarter is $1.6M. When this quarter closes, the forecast for the next one becomes whatever this quarter booked.The Institute of Business Forecasting defines a naive model as "a model in which minimum amounts of effort and manipulation of data are used to prepare a forecast." In the standard forecasting textbook, Forecasting: Principles and Practice by Rob Hyndman and George Athanasopoulos, the naive method sets all forecasts "to be the value of the last observation." The textbook also calls it the random walk forecast, because a naive forecast is optimal when the data follow a random walk, and it works well for many economic and financial series.
The word "naive" causes some confusion in search results. An older R News article ranks for the term but covers Holt-Winters exponential smoothing, which it calls naive only because it uses no outside variables. This page uses the narrow, standard meaning: copy the last actual, or copy the same period last cycle.
Naive Forecast Formulas: Naive, Seasonal Naive and Drift
Three simple methods cover nearly every benchmark use: naive, seasonal naive and drift. Hyndman's textbook uses these three plus the mean of all history as its four benchmark methods. The table shows each formula with an Excel or Google Sheets version, assuming quarters run down column A and actuals sit in column B starting in row 2.| Method | Formula | Excel or Sheets (forecast in row 10) | Best baseline for |
|---|---|---|---|
| Naive (random walk) | Forecast = last actual | =B9 | Series with no seasonality, such as a stable renewal base |
| Seasonal naive | Forecast = actual from the same season one cycle back (m periods ago) | =B6 for quarterly data (m = 4); for monthly data, the cell 12 rows up | Quarterly or monthly bookings, which almost always have seasonality |
| Drift | Forecast = last actual + h x (last actual - first actual) / (T - 1) | =B9+(B9-$B$2)/(ROW(B9)-ROW($B$2)) | Series with a steady trend and little seasonality |
| Mean | Forecast = average of all history | =AVERAGE($B$2:B9) | Flat, noisy series with no trend |
Why Is the Naive Forecast the Benchmark Every Forecast Must Beat?
Because it costs nothing, a forecast that cannot beat it is wasting the time spent building it. Hyndman's textbook is direct about this: new methods are compared to the simple ones, and "if not, the new method is not worth considering."Michael Gilliland at SAS turned the same idea into a management metric called forecast value added (FVA). He defines it as "the change in a forecasting performance metric ... that can be attributed to each particular step and participant in your forecasting process." The naive forecast plays the role of the placebo in a drug trial. Every step of the process, from the statistical model to the manager override, is scored by how much it improves on doing nothing. In the white paper's sample report, a statistical model cut MAPE from 25% to 20%, and an analyst override then pushed it back up to 30%, a negative FVA of 10 points against the model.
For a SaaS revenue team the steps are usually the rep submission, the manager roll-up, the forecast override in the weekly call and any model output. Each one costs hours every week. FVA asks whether each one makes the number better than the seasonal naive, and better than the step before it.
This matters because most sales forecasts are not very accurate to begin with. Gartner found that only 7% of sales teams reach forecast accuracy of 90% or more, and the median is 70% to 79%. Pete Furseth, ORM's COO, puts the usual accuracy on new and expansion business at around 90%, and notes that it "takes a lot of time and effort to produce and is not dynamic as conditions change." A naive benchmark tells you how much of that effort is buying accuracy.
Worked Example: Naive vs Seasonal Naive on SaaS Bookings
In this illustrative example, the plain naive forecast misses by a third, the seasonal naive misses by under 10%, and the sales team's forecast barely beats the seasonal naive. The numbers are made up to show the method. They are not ORM customer data.A company has net new ARR bookings (new plus expansion, excluding renewals) of:
| Quarter | Q1 | Q2 | Q3 | Q4 |
|---|---|---|---|---|
| FY2023 | $900K | $1,200K | $950K | $1,450K |
| FY2024 | $1,000K | $1,300K | $1,050K | $1,600K |
| FY2025 (held out) | $1,100K | $1,450K | $1,150K | $1,780K |
| Method | Q1 forecast | Q2 forecast | Q3 forecast | Q4 forecast | MAE | MAPE | MASE | FVA vs seasonal naive |
|---|---|---|---|---|---|---|---|---|
| Naive | $1,600K | $1,100K | $1,450K | $1,150K | $445.0K | 32.8% | 3.96 | -23.2 pts |
| Drift | $1,700K | $1,125K | $1,511K | $1,175K | $472.8K | 35.6% | 4.20 | -26.0 pts |
| Seasonal naive | $1,000K | $1,300K | $1,050K | $1,600K | $132.5K | 9.6% | 1.18 | baseline |
| Team forecast | $1,250K | $1,400K | $1,300K | $1,650K | $120.0K | 9.4% | 1.07 | +0.2 pts |
| Statistical model | $1,120K | $1,420K | $1,190K | $1,700K | $42.5K | 3.0% | 0.38 | +6.6 pts |
- The plain naive forecast is the wrong benchmark for seasonal bookings. It carries every strong Q4 into a weak Q1, so its errors are huge. Beating it proves nothing. The drift method fails the same way, because it extends a trend into a seasonal dip.
- The seasonal naive is the real bar. Copying last year's quarter missed by less than 10% on average in this example, with zero effort.
- The team forecast added 0.2 points. Weeks of calls, roll-ups and overrides produced almost the same accuracy as copying last year. That is the finding FVA is designed to surface, and the fix is to look at which step adds error, not to add more steps.
- The model clears the bar on held-out quarters. Its FVA of 6.6 points against the seasonal naive, and 6.4 points against the team call, is the evidence that it earns its cost. Pete's test for any vendor is the same: "during the implementation they should be able to hold out a quarter and measure the model accuracy on the hold out quarter."
The FVA column is the seasonal naive's MAPE minus each method's MAPE, so a positive number means the method beat the baseline. Check every figure in your own sheet: for the seasonal naive, the four absolute errors are $100K, $150K, $100K and $180K, so MAE is $530K / 4 = $132.5K, and MAPE is the average of 9.09%, 10.34%, 8.70% and 10.11%, or 9.56%.
How to Measure Against a Naive Forecast: MASE
MASE, the mean absolute scaled error, divides your forecast's average absolute error by the average error a naive forecast made on the training data, so a score below 1 means you beat that baseline. Rob Hyndman and Anne Koehler proposed it in 2006, and the accuracy chapter of Hyndman's textbook explains it in full. For seasonal data the scale uses the seasonal naive.In the example, the seasonal naive errors inside the training years were $100K, $100K, $100K and $150K (FY2024 against FY2023, quarter by quarter), an average of $112.5K. The team's held-out MAE of $120.0K divided by $112.5K gives a MASE of 1.07. The model's $42.5K gives 0.38.
In Excel or Google Sheets, with actuals in B2:B13 and a method's FY2025 forecasts in F10:F13:
=(SUMPRODUCT(ABS(B10:B13-F10:F13))/4)/(SUMPRODUCT(ABS(B6:B9-B2:B5))/4)
MASE solves two problems with MAPE. The textbook notes that percentage errors are "infinite or undefined" when an actual is zero, which happens in thin segments and new products, and that they put a heavier penalty on negative errors than on positive ones. A scaled error has neither problem, and it lets you compare accuracy across segments of very different size.
One caution when you read it: the team scored 1.07 even though it beat the seasonal naive on the held-out year. MASE compares against the naive's error in the training years, and FY2025 happened to be a harder year for every method. Compare MASE values to each other, with the seasonal naive's own held-out MASE (1.18 here) as the line to beat.
Naive Forecast vs Moving Average Forecast
Both are cheap baselines. The naive forecast copies one period, and a moving average averages several. A moving average forecast smooths out a single odd quarter, which the naive forecast copies forward in full. The SAS white paper counts a moving average as a valid naive model for FVA work, and notes that a 12-month window has "the advantage of smoothing out any seasonality."That smoothing is also the weakness for SaaS bookings. Averaging Q1 through Q4 flattens the very pattern a revenue team most needs to see, which is why the seasonal naive usually makes the tougher benchmark for quarterly data. Pete's point about the months inside a quarter applies too: "Within a quarter the third month is always the largest and the first month is always the smallest." A monthly forecast needs a seasonal baseline that repeats that shape, which a flat average cannot do. For how to model trend and seasonality properly, see time series forecasting.
Where a Naive Forecast Breaks
A naive forecast only knows the past, so it misses every turning point. It has no view of the pipeline already open, the deals that will be created and closed inside the quarter, a price change or a new competitor. When win rates fall or deal sizes shrink, the naive forecast keeps quoting last year until the actuals catch up.Use it three ways:
- As the floor in every accuracy review. Report each forecast's MAPE or MASE next to the seasonal naive's. The guide to measuring forecast accuracy covers the other metrics to track with it.
- As the test in backtesting. When you backtest a forecast model, score it on held-out quarters against the seasonal naive before anything else.
- As a short-history fallback. Microsoft's Dynamics 365 demand planning switches to naive rules when a series is too short to train a model, and calls it "only a fallback, not a replacement for regular forecasting models."
ORM builds a model on each company's own sales history, adjusts it for quarterly and weekly seasonality, and targets 95% forecast accuracy on new and expansion business, updated as the quarter moves. Whatever tool you use, the first question is the same one this page asks: on quarters it has never seen, does it beat copying last year?
Frequently Asked Questions
What is the formula for naive forecasting?
The naive formula is: forecast for the next period = the actual of the last period. Written as an equation, the forecast at T+h equals y at T for every future period h. The seasonal naive formula uses the actual from the same season one cycle back, so the Q3 forecast equals last year's Q3. In Excel, the naive forecast in row 10 is simply =B9.
What is the difference between a naive and a seasonal naive forecast?
A naive forecast repeats the most recent actual, so a strong Q4 becomes the forecast for Q1. A seasonal naive forecast repeats the same period from the previous cycle, so Q1 is forecast from last year's Q1. For SaaS bookings, which swing by quarter, the seasonal naive is almost always the better baseline and the harder one to beat.
Why is the naive forecast used as a benchmark?
It costs nothing, needs no judgment and is hard to argue with. Any forecasting method or process that takes time and money should beat it on the same held-out periods. Hyndman and Athanasopoulos use it this way in their forecasting textbook, and SAS's forecast value added method measures every step of a forecast process against it.
What does a MASE below 1 mean?
MASE, the mean absolute scaled error, divides your forecast's average absolute error by the average error of a one-step naive forecast on the training data. Below 1 means your forecast errs less than that naive baseline did. Above 1 means it errs more. Use the seasonal naive as the scale when your data is seasonal, which quarterly bookings usually are.
What is forecast bias vs. forecast accuracy?
Accuracy measures how big the misses are, whatever their direction. Bias measures whether the misses lean one way, such as a sales team that over-calls every quarter. A forecast can have a small average error and still be biased if every miss is high. Track both, and compare each against a naive forecast so you know whether the process is helping.
When should you actually use a naive forecast?
Use it as the floor in every accuracy review, and as a fallback when a series is too short to model, such as a new product or segment with a few quarters of history. Do not commit it to the board as the plan. It cannot see the pipeline, pricing changes or a shift in win rates, so it will miss turning points.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like naive forecast into prescriptive action for your team.
Schedule a Demo