Sales Forecasting Methods That Work When You Deliver Your Own Orders

Most sales forecasting methods were designed for businesses that sell a small number of large deals over a long cycle. If you run delivery routes, you have the opposite problem: a large number of small orders on a short cycle, where the forecast has to survive being converted into driver hours by Thursday.

The method you pick should follow from your order book, not from which approach sounds most advanced. A wholesaler with 80 standing accounts and a caterer chasing twelve large events a year need different arithmetic, and the caterer who copies the wholesaler’s moving average will forecast a flat line through a business that does not have one. If you are still deciding whether the exercise is worth the time at all, the case for forecasting in a delivery operation covers what the number is actually for.

This post compares the five methods in general use, what each needs, where each fails, and which combination tends to work at which size.

The Bottom Line

• There are two families: quantitative methods that extrapolate from data you already have, and qualitative methods that structure judgement when you do not have the data.
• Repeat-order businesses should start with a historical baseline or moving average. Project and event businesses should start with pipeline forecasting. Nearly everyone ends up blending two.
• Method choice matters less than review frequency. Teams tracking weekly reached 87% forecast accuracy against 52% for those tracking irregularly (Gangly).
• No method handles seasonality for free. Every one listed here needs a manual adjustment for the holiday weeks that dominate food, floral and catering demand.

The two families of sales forecasting methods

Every method on this page belongs to one of two groups, and knowing which you are using tells you what its weakness will be.

Quantitative methods extrapolate from recorded data: past orders, conversion rates, deal values. They are reproducible, they improve as history accumulates, and they are blind to anything that has not happened before. A quantitative forecast will cheerfully project last February into this February and miss that you have since added three accounts and lost your second-largest.

Qualitative methods convert expert judgement into a number through a structured process. They handle new products, new markets and one-off events, and they carry whatever optimism the person doing the judging carries. They are the only option in a first year of trading.

The practical answer for most delivery businesses is a quantitative baseline with a qualitative overlay: extrapolate the standing book, then adjust by hand for what you know that the data does not.

Comparing the five main forecasting methods

Historical forecasting: projecting last year forward

Historical forecasting takes what you sold in a comparable past period and uses it as the estimate for the period ahead, adjusted for known growth. It is the simplest method that works, and for a wholesale or standing-order business it is often the most accurate thing available.

The key word is comparable. For a delivery business, the right comparison is the same week last year, not last month, because the shape of the year dominates. Week 7 is Valentine’s week for a florist whatever happened in January.

Where it fails is a changed customer base. If you added a major account in June, every pre-June comparison understates you by that account’s volume, and the fix is to forecast the base book from history and add new accounts separately by hand.

Best for: wholesalers, bakeries and distributors with a stable list of repeat customers and at least a year of history.

Moving average: smoothing out noisy weekly volume

A moving average takes the mean of the last several periods and rolls forward, dropping the oldest as each new one lands. A four-week moving average on weekly order counts is the usual implementation and takes about ten minutes to set up.

Its job is noise reduction. Individual weeks in a delivery business bounce around for reasons that carry no information: a bank holiday, a bad weather Tuesday, one customer ordering twice. An average absorbs that so you are not reacting to each wobble.

Its job is not peak prediction. An average built from ordinary weeks will always under-call a holiday week, sometimes by a lot. Use the average for the baseline and override it entirely for known peaks.

A weighted moving average, where the most recent weeks count for more, responds faster to a real trend change at the cost of a little more arithmetic. If your volume is growing month over month, weight it.

Best for: operations with high order counts and volatile week-to-week swings, as a baseline rather than a complete answer.

Pipeline forecasting: working from named opportunities

Pipeline forecasting lists every open opportunity, assigns each a probability based on how far along it is, and sums the weighted values. A quote sent is worth less than a quote verbally accepted, which is worth less than a signed contract with a start date.

This is the method the mainstream sales forecasting literature is mostly about, and it is the right one if your revenue arrives as a small number of named commitments: catering contracts, wholesale supply agreements, corporate accounts, event bookings.

Two cautions for delivery businesses. First, it only forecasts the pipeline, so a business with 90% repeat volume and 10% new business should pipeline-forecast the 10% and historically forecast the 90%. Second, stage probabilities need to come from your own closed-won history, not from a template. The classic failure is that every opportunity sits at 50% because nobody updated the stages, and forecast accuracy follows data quality closely.

Best for: caterers, event businesses and wholesalers whose growth comes from winning named accounts.

Bottom-up forecasting: building from individual customers or routes

Bottom-up forecasting estimates each customer, product or route separately and adds them together. It is the most accurate method available to a small operator and the most expensive in time.

For a delivery business the natural unit is the route or the account. You forecast each standing account from its own ordering pattern (account A takes 20 cases every Tuesday, account B takes 8 cases fortnightly and has been slowing), then sum to get the week. Because you built it from parts, you can also read the forecast by route directly, which is what scheduling needs anyway.

The method pays for itself when accounts behave very differently from each other, which is common once you pass about 20 accounts and before you pass about 200. Below that, you hold it in your head. Above that, the admin becomes untenable and you aggregate into groups or buy a tool.

Best for: mid-size books where a handful of accounts make up most of the volume and their patterns diverge.

Qualitative forecasting: structuring judgement when you have no data

Qualitative forecasting turns informed opinion into a number. In its simplest form you and whoever runs your routes sit down and estimate the period ahead from what you know about customers, the calendar and what is in the diary.

It is unavoidable in three situations: a new business with no history, a new product or route with no analogue, and a period with no precedent. It is also the only method that can incorporate things nobody has recorded: that a major account’s buyer has changed, that a competitor just closed, that a long-standing customer sounded unhappy on the phone.

Make it auditable by writing down the reasoning alongside the number, not just the number. When the forecast misses, the reasoning tells you which assumption was wrong, and that is the part you can fix.

Best for: year one, new product launches, and as the overlay on top of any quantitative baseline.

Which forecasting method should you use?

Pick by order book shape first and size second.

• Mostly repeat orders from a stable list: historical baseline by same-week-last-year, plus a qualitative overlay for events and new accounts.
• Mostly named contracts and events: pipeline forecasting for new business, historical for whatever recurs.
• High order count, volatile weeks: weighted moving average as a baseline, manual override for peaks.
• Twenty to two hundred accounts behaving differently: bottom-up by account, aggregated to routes.
• No history at all: qualitative, constrained by capacity, revised every single week until history exists.

Whichever you choose, the method is only half the job. The forecast still has to be built, reviewed and compared against what actually shipped, and the practical sequence for that is in the guide to creating a sales forecast you can plan routes against.

Why most delivery businesses blend two methods

Pure methods are rare in practice because most order books have two distinct parts: a predictable base and a variable top.

The base is standing orders from established customers. It is predictable, and historical or moving-average methods handle it well. The top is new business, events and one-offs, which history cannot see and which pipeline or qualitative methods handle instead.