Skip to content
EncryptTechBD
Home Industries Our Work Blog Pricing About Contact Discuss Your Project
বাংলা (BN)
Data & Analytics 5 min read

Sales forecasting with business data explained

How sales forecasting works with the data a Bangladesh business already has, from clean history and seasonality to baselines, accuracy and human overrides.

E
EncryptTechBD · 0 comments
Charts and graphs on a laptop
Photo: Raymond Sam / StockSnap

Sales forecasting fails in two opposite ways. A business forecasts from a single number in a spreadsheet because it is fast, then misses a stock or cash decision. Or it buys a model, feeds it dirty data and trusts the output because it looks scientific. The useful middle is a transparent forecast that shows its assumptions and gets better every month.

This explainer is for a Bangladesh business that sells through retail, distribution, ecommerce or field sales and wants to forecast demand with the data it already has.

Forecast, target or pipeline?

A target is what you want. A forecast is what you expect. A pipeline is what might happen. Mixing them corrupts all three. A sales target should not be used as the demand forecast, and a pipeline should not be treated as committed revenue without evidence of conversion.

The forecast also needs a decision attached. A number that does not change a purchase order, a staffing plan or a cash plan is a report, not a forecast.

How a forecast is built

Forecasting is a sequence: define the unit, assemble history, separate pattern from noise, model, test, review and publish. The unit matters. A forecast for total revenue is not useful for stock; a forecast for every SKU may be too noisy. Choose a level that matches the decision: category, branch, channel, key account or SKU.

The horizon should match the lead time. A bakery may forecast daily; an importer with a long lead time needs a longer horizon and a wider range. Short horizons are usually more accurate, so do not promise the same precision for a six-month plan as for next week's production.

The data that matters

  • Sales history by date, location, channel and product, net of returns and cancellations.
  • Price, discount and promotion history.
  • Stock-outs and lost sales, so the history is not mistaken for demand.
  • Calendar and events: Eid, Pohela Boishakh, monsoon, winter, exams and paydays.
  • New product launches, discontinuations and market changes.
  • Marketing spend, field visits and competitor activity where available.
  • Lead times and supplier constraints.

Most forecasting projects spend more time fixing this list than tuning a model. That is normal. A clean three-year history with promotion flags is worth more than a sophisticated model on messy data.

Methods, from simple to advanced

  1. Naive: same period last year, or last month.
  2. Moving average: smooth the noise.
  3. Exponential smoothing: weight recent periods more heavily.
  4. Regression: relate sales to price, promotions, weather or footfall.
  5. Machine learning: use many drivers and hierarchical data when there is enough history.

Always compare against the naive baseline. A model that cannot beat "same month last year" is not ready for a stock decision. The statsmodels documentation covers time-series and regression models, while scikit-learn documents the error metrics used to compare them.

A worked example

A distributor forecasts monthly demand for three categories across two warehouses. Last year's same month was 1,200 units. The category has been growing, so the baseline is adjusted upward by a trend factor. A known promotion adds an expected lift, and a stock-out in the same month last year means the history understates demand, so the team adjusts again.

The result is not one number. It is a low, expected and high case, with the assumptions written next to each. The warehouse planner uses the high case for safety stock and the expected case for purchasing, then reviews the gap after the month closes.

Illustrative arithmetic: 1,200 units multiplied by a 1.05 trend factor gives 1,260. A promotion estimate of 100 units gives an expected case of 1,360. If the promotion is uncertain, the low case stays at 1,260 and the high case adds the full lift. These are planning assumptions, not a market forecast.

Measure accuracy in a way that changes decisions

  • Error by category, branch and horizon, not just one overall number.
  • Bias: are you consistently over or under?
  • Forecast value added versus the naive baseline.
  • Service level, stock cover and write-offs.
  • Forecast stability: how much the number moves before the month starts.

Use percentage error measures with care when volumes are low or zero; a large percentage on a tiny base can look dramatic. Absolute error and weighted error often tell a more useful story across a product range.

Limits

Forecasts are conditional statements, not promises. They break when a new competitor enters, a policy changes, a key supplier fails or a promotion behaves differently. New products have no history. Long lead times mean the forecast must be made earlier and with wider intervals. Human judgment is still needed for information the model cannot see, but every override should be logged and reviewed.

It also helps to separate independent demand from demand created by supply. If the sales team pushes stock into distributors, shipments are not the same as end-customer demand, and a forecast built on shipments will overstate what the market is buying.

The monthly operating rhythm

  1. Refresh the data and check for missing or duplicated periods.
  2. Generate the baseline and the model forecast.
  3. Review exceptions with sales, operations and finance.
  4. Record overrides with a reason and an owner.
  5. Publish one number set for purchasing, staffing and cash.
  6. Compare actuals, update the model and repeat.

A forecast that is reviewed once a year is a budget. A forecast that is reviewed monthly becomes an operating capability.

Where this fits in our work

If you need help building a forecasting routine on your sales and operations data, start with Operations. The pricing page explains scope. For the reporting layer, read the business KPI dashboard explainer.

Sources

Services that deliver this

Want this implemented properly?

Pricing →

Every engagement starts with a written scope. If a requirement is outside it, we quote it separately before starting.

Responses (0)

No comments yet. Add the first one below.

Leave a response

Your email address will not be published. Required fields are marked *.

Keep reading

More field notes