What is quantitative forecasting? It uses past data and math to predict what's next. See five methods, worked examples, and when to trust the numbers.
Ask a sales manager how many calls they'll get next Tuesday, and you'll usually get a shrug, a guess, or a number pulled from thin air. The strange part is they already have the answer. It's sitting in last month's call logs.
So what is quantitative forecasting? It's the method that pulls that number out of the data. It uses past numbers and a bit of math to predict what's coming, whether that's next quarter's revenue, next week's call volume, or how many reps you need on the phones.
Alpharun works with Medicare brokerages like Chapter, covering hundreds of advisors across hundreds of thousands of calls, so the worked examples include real call-center math. By the end, you'll be able to run the math on your own numbers and, just as usefully, spot when the math is quietly lying to you.
What is quantitative forecasting?
Quantitative forecasting is a method for predicting future outcomes from historical numerical data and math or statistical models. Feed it enough clean history, and it projects the pattern forward.
It works best with two things in place. You need enough past data to see a real pattern, and that pattern has to be stable enough that the past is a fair guide to the future.
The opposite approach is qualitative forecasting, which leans on human judgment when the numbers run out (more on that split later). Most quantitative methods fall into one of two families: time series and causal.
How quantitative forecasting works: Time series vs causal methods
The two families answer different questions.
Time series methods look inward. They forecast a variable from its own history, so last week's call volume predicts next week's. You use them when a number moves in patterns you can see, like a steady climb, a weekly rhythm, or a seasonal spike.
Causal methods look outward. They predict one variable from other variables, like using ad spend and season to forecast inbound leads. You reach for them when you actually know the drivers and can measure them.
Neither reaches very far into the future. Both are strongest over short to intermediate horizons, and accuracy drops the further out you push. Forecasting next month is a reasonable bet; forecasting the same week two years out is closer to astrology. If you want to see where forecasts sit inside a broader sales analytics practice, that's the wider toolkit this fits into.
5 quantitative forecasting methods (with formulas and examples)
Here are the five methods worth knowing, from the plainest to the most involved. Each one gets a formula in words and a worked example on small numbers that you can follow.
Method | What it predicts from | Formula in words | Best when |
|---|---|---|---|
Naive forecast | The single last period | Next = last actual | You need a quick baseline |
Moving average | The last few periods | Average of the last N periods | Numbers wobble around a level |
Exponential smoothing | Recent history, weighted | Weighted mix of last actual and last forecast | Recent data should count more |
Seasonal time series | A baseline plus a seasonal index | Baseline × seasonal index | Demand swings by season |
Linear regression | A driver variable | Y = a + bX | You have a measurable driver |
1. Naive forecast
The naive forecast is the simplest one there is: whatever happened last period is your forecast for the next period.
Forecast = last period's actual.
If your team closed 118 deals last month, your naive forecast for this month is 118. That's it. It sounds too basic to bother with, but it's genuinely useful as a baseline. Any fancier method you try should beat the naive forecast, or it isn't earning its keep.
2. Moving average
A moving average smooths out the noise by averaging the last few periods, so no single spike throws it off.
Forecast = (sum of the last N periods) ÷ N.
Say the last four weeks of calls were 1,050, 1,180, 1,100, and 1,240. A four-week moving average gives you (1,050 + 1,180 + 1,100 + 1,240) ÷ 4, which lands at about 1,143. Pick a bigger N to smooth harder, a smaller N to react faster. A weighted moving average is the same idea, with more weight on recent weeks.
3. Exponential smoothing
Exponential smoothing also leans on recent data, but it does so with a dial. A smoothing factor called alpha (a number between 0 and 1) decides how much the newest reading counts.
Forecast = α × (last actual) + (1 − α) × (last forecast).
Say last month's revenue came in at $124,000 and your forecast for it was $115,000. With alpha set to 0.3, you get 0.3 × $124,000 + 0.7 × $115,000, or $117,700. Crank alpha up and the forecast chases recent swings. Turn it down, and it stays calm. It's a favorite for demand that shifts gradually.
4. Seasonal time series
Some numbers swing on a calendar you already know. Seasonal forecasting takes a baseline and multiplies it by a seasonal index that captures the swing.
Forecast = baseline × seasonal index.
If your normal month runs 1,000 orders and your December index is 1.35 (December usually runs 35% hot), your December forecast is 1,000 × 1.35, or 1,350. Build your indexes from a few years of history so one odd quarter can't skew them.
5. Linear regression
Regression is the causal method most teams meet first. It fits a straight line that predicts an outcome (Y) from a driver (X).
Y = a + bX, where a is the starting point and b is how much Y moves per unit of X.
Suppose the fitted line comes out to Sales = 20 + 0.05 × (website visits). Plug in 8,000 visits for next month, and you get 20 + 400, or 420 sales. Regression is only as good as the driver you pick; if visits and sales aren't actually linked, the line will mislead you with total confidence.
Quantitative vs qualitative forecasting
The difference between quantitative and qualitative forecasting comes down to what each one runs on. Quantitative forecasting runs on historical numbers and formulas, and qualitative forecasting runs on informed human judgment. Most teams need both, and knowing which to trust when is half the skill.
Factor | Quantitative | Qualitative |
|---|---|---|
Data used | Historical numbers, measured over time | Expert opinion, market read, deal context |
Best when | You have clean history and stable patterns | Data is thin, new, or the situation just changed |
Strengths | Objective, repeatable, scales across large volumes | Catches context and one-off events numbers miss |
Weak spots | Blind to anything not in the data | Subjective, hard to repeat, prone to bias |
Qualitative methods are worth naming, since you'll lean on them when the numbers dry up. The common ones are executive or expert opinion, the Delphi method (structured rounds of anonymous expert input), sales-force composite (aggregating your reps' own read on their pipeline), and market research or surveys.
Use both: the numbers tell you what usually happens, and judgment tells you when this time is different. A brand-new product launch is a judgment call. Next week's call volume on a mature line is a math problem.
When to use quantitative forecasting (and when it will mislead you)
Quantitative forecasting shines when three things are true: you have enough reliable history, the underlying pattern is reasonably stable, and you're forecasting over a short to medium horizon. Staffing a call center, projecting next quarter's enrollments, and planning inventory are all good fits.
It falls apart in a few predictable ways, and it's worth being honest about them:
No history to learn from: A new product, a new market, or a new script has no past for the model to read.
Sudden shocks: A regulation change, a viral moment, or a competitor's price cut can break the pattern overnight, and the model won't see it coming.
Dirty or thin data: If half your calls never get logged or scored, the model is confidently forecasting from a fraction of reality.
Bad data does the most damage because it's the hardest problem to spot. A precise-looking number built on a bad dataset is more dangerous than an honest guess, because people actually act on it. Garbage in, confident garbage out. Pair the math with human judgment, and fix your data before you fine-tune your method.
Quantitative forecasting example: Predicting call volume for staffing
Picture an inside-sales floor trying to figure out how many reps to schedule next week.
Start with the history. Say the last four weeks of inbound volume were 5,400, 5,600, 5,900, and 6,100 calls. A four-week moving average gives you 5,750 (23,000 ÷ 4), but that's below last week's actual because an average lags behind volume that keeps climbing.
Fit a trend line instead (the linear regression method above, with the week number as the driver), and volume is rising by about 240 calls a week, which puts next week near 6,350.
Now turn that forecast into a staffing plan. If one rep handles about 40 calls a day across a five-day week, that's roughly 200 calls per rep per week. Divide 6,350 by 200, and you land near 32 reps. Plan from the moving average alone, and you'd schedule about 29, three reps short. That single number is what turns "let's add some people, I guess" into a staffing decision you can defend in a budget meeting.
Layer in average handle time and target occupancy, and the estimate gets more realistic and usually higher, because no rep spends every minute of a shift on calls.
That more realistic number is the whole point of workforce optimization. Get the forecast wrong on the low side, and calls sit in a queue while sales leak out the bottom. Get it wrong on the high side, and you're paying idle reps to wait.
Seasonality makes the stakes bigger. Medicare brokerages see call volume climb every fall when the annual enrollment window opens on October 15, and a flat headcount plan gets buried. A seasonal index applied to the baseline is what keeps the floor ready for the spike before it lands.
How do you know if a forecast is accurate?
You judge a forecast by how far off it was, tracked consistently over time. Three plain measures do most of the work:
MAD (mean absolute deviation): The average size of your misses, in plain units. Off by 50 calls on average? That's your MAD.
MSE (mean squared error): Squares each miss before averaging, so it punishes big misses much harder than small ones. Use it when one giant miss hurts more than several small ones.
MAPE (mean absolute percentage error): The miss as a percentage, which makes it easy to compare forecasts across teams or products of different sizes.
The practical habit is simple: track your error every period, and hold every method to one bar: it has to beat the naive forecast. If your clever model can't out-predict "same as last week," it isn't worth running.
Forecast the headcount, then coach every call
Ask Alpharun preparing coaching notes for a rep from two weeks of calls, next to that rep's scores against the team.
Every quantitative forecasting method here has the same soft spot: a forecast is only as good as the history behind it. Clean, complete data produces a number you can staff and budget against. Thin data produces a confident-sounding guess.
A forecast tells you how many reps to put on the floor. How well those reps sell is a separate question, and manual QA only ever reviews a small fraction of their calls. When most conversations never get scored, your coaching and your call center business intelligence run on a sample too small to trust.
Alpharun scores every call against a playbook built from your own best conversations, so managers see how every rep actually sells. It also handles the repetitive scoring work, which gives managers time back for coaching.
With Alpharun, teams can:
Score 100% of calls automatically against a custom playbook, well beyond the small share a manual QA team can cover.
Spot rep and team performance trends as they form, while there's still time to act.
Find the behaviors behind your best calls and build them into a shared playbook every rep can learn from.
Deliver sentence-level coaching that pulls the middle of the team toward top-performer numbers.
Flag compliance and process misses on every call, in every period, across the whole floor.
Send managers a weekly digest report showing exactly what to coach next, plus short coaching notes for each rep right after calls.
Your forecast tells you how many reps to schedule, and Alpharun shows how well each one sells. Put the two together, and your contact center analytics cover both headcount and performance. Book a demo with Alpharun to see what your own calls have been trying to tell you.
Frequently asked questions
What is quantitative forecasting in simple terms?
Quantitative forecasting predicts future outcomes using past numbers and math models. For example, it uses the last four weeks of call volume to project next week's, giving you a data-based estimate you can plan around.
What are the main quantitative forecasting methods?
The main quantitative forecasting methods are the naive forecast, moving average, exponential smoothing, seasonal time series, and linear regression. The first four are time series methods that project a variable from its own history, while regression is a causal method that predicts an outcome from a driver variable.
What is the difference between quantitative and qualitative forecasting?
Quantitative forecasting uses historical data and formulas, while qualitative forecasting uses expert judgment and market context. Quantitative works best when you have a clean, stable history; qualitative takes over when data is thin, new, or the situation has just changed.
What are the four types of forecasting?
There's no single standard list of four. Most sources group forecasting methods into three families: time series methods (including the naive forecast), causal or regression methods, and qualitative methods. The first two are quantitative and run on data, while qualitative forecasting relies on informed human judgment.
When should you not use quantitative forecasting?
You should avoid relying on quantitative forecasting when you lack reliable history, the pattern is unstable, or your data is incomplete. New products, sudden market shifts, and poorly logged data all produce confident forecasts that turn out to be wrong.
