- A useful eCommerce forecast splits revenue into returning customers (predictable, from your order history) and new customers (driven by ad spend and acquisition cost). Forecasting one total revenue line hides both.
- Build it from your own data: an order export with customer IDs, dates and values, plus your ad spend. A spreadsheet is enough for most stores doing $1M to $50M a year.
- Forecast in ranges (low, base, high), compare against actuals every month, and use the gaps to fix your assumptions. A forecast you never check is just a guess with formatting.
Why do eCommerce brands need a sales forecast?
Inventory buys, ad budgets, hiring and cash planning all assume a level of sales. When that assumption is “last year plus 20%,” the business either runs out of stock in its best month or ties up cash in slow inventory.
Sales forecasting does not need to be precise to be useful. It needs to be explicit. When the forecast is built from named drivers (customers, order rates, order values, ad spend), you can see which assumption was wrong when reality differs, and you can fix it. A single trend line cannot tell you why it missed.
The profit angle matters too. Revenue forecasts drive spending, and spending drives margin. A forecast that assumes new customers will arrive at last year’s acquisition cost, when costs have risen, will push you to overspend.
What data do you need?
You can build the model below from two sources most stores already have.
| Data | Where it comes from | What you use it for |
|---|---|---|
| Order export: order date, customer ID or email, order value, discounts, refunds | Shopify, WooCommerce or Magento order export | Returning customer rates, order values, seasonality |
| Ad spend by month and channel | Ad platform exports or your reporting dashboard | New customer volume and acquisition cost |
| Planned promotions and launches | Your marketing calendar | Adjustments for known events |
| Inventory constraints | Your inventory or ERP system | Caps on what you can actually sell |
Use at least 24 months of order history if you have it, so you can see each month of the year at least twice. Use net revenue (after discounts and refunds), not gross sales, so the forecast matches the money you keep.
If your data lives in several systems, a data warehouse makes this easier, but it is not required to start. A clean order export in a spreadsheet is enough for a first version.
How does the simple model work?
The model splits monthly revenue into two parts, because they behave differently.
Forecast revenue = returning customer revenue + new customer revenue
Part 1: returning customer revenue
Returning revenue comes from people who have bought before. It is the most predictable part of your sales, because you already know who these customers are and how often they come back.
Returning revenue = active customers × monthly repeat rate × returning order value
- Active customers: customers who bought in the last 12 months (or a window that fits your purchase frequency).
- Monthly repeat rate: the share of active customers who ordered in that same month last year. Calculate it per month, because it changes with the season.
- Returning order value: the average net order value from returning customers in that month last year.
Cohort analysis improves this part. Customers acquired in a heavy discount month often repeat less than customers acquired at full price. If your recent cohorts repeat at a lower rate than older ones, use the recent rate. Our guide to cohort analysis in Shopify shows how to build the table.
Part 2: new customer revenue
New customer revenue has two sources: customers you pay to acquire and customers who arrive without paid media (organic search, word of mouth, direct).
New customer revenue = (planned ad spend ÷ expected new customer CAC + organic new customers) × first order value
- Expected new customer CAC: what you expect to pay per new customer at the planned spend level. Use recent months, not last year, because costs change. Our guide to customer acquisition cost covers how to calculate it.
- Organic new customers: new customers in months with little paid activity, or a baseline you estimate from last year adjusted for trend.
- First order value: first orders often differ from repeat orders, so measure them separately.
One warning on CAC. Acquisition cost usually rises as spend rises, because each extra dollar reaches a less responsive audience. If you plan to double spend, do not assume CAC stays flat. Use a higher CAC for the extra spend. Our article on ad budget allocation explains how to think about marginal returns.
Part 3: adjustments for known events
Add or subtract for things you know about that last year’s numbers do not reflect:
- A promotion that ran last year but not this year, or the reverse
- A product launch or a discontinued best-seller
- Calendar shifts, such as Black Friday falling on November 27 in 2026
- Stock limits: you cannot sell what you will not have
Write each adjustment down with its reason, so you can check it later.
Worked example: forecasting one month
A store (illustrative numbers) is forecasting March. It has 20,000 active customers. Last March, 4.5% of active customers ordered, but recent cohorts repeat a little less, so the team uses 4.0%.
| Driver | Value | Source |
|---|---|---|
| Active customers | 20,000 | Order export, last 12 months |
| Monthly repeat rate (March) | 4.0% | Last March, adjusted for weaker recent cohorts |
| Returning order value | $70 | Last March, returning customers |
| Returning revenue | $56,000 | 20,000 × 4.0% × $70 |
| Planned ad spend | $60,000 | Media plan |
| Expected new customer CAC | $50 | Last three months, at similar spend |
| Paid new customers | 1,200 | $60,000 ÷ $50 |
| Organic new customers | 600 | Last March, adjusted for trend |
| First order value | $60 | Last three months |
| New customer revenue | $108,000 | 1,800 × $60 |
| Known adjustment: spring launch | +$10,000 | Pre-orders and launch email plan |
| Base forecast | $174,000 | Sum |
Notice what the split shows. Returning customers bring about a third of the revenue at no acquisition cost. New customers bring more revenue, but they cost $60,000 in ads to get. The profit picture depends heavily on contribution margin per first order and on how many of those new customers come back.
How do you turn one forecast into a range?
A single number invites false confidence. Build three versions by changing the drivers you are least sure of.
| Scenario | Repeat rate | New customer CAC | Launch adjustment | Forecast |
|---|---|---|---|---|
| Low | 3.5% | $60 | +$5,000 | $49,000 + $96,000 + $5,000 = $150,000 |
| Base | 4.0% | $50 | +$10,000 | $174,000 |
| High | 4.5% | $45 | +$15,000 | $63,000 + $116,000 + $15,000 = $194,000 |
Plan inventory for somewhere between base and high on your best sellers, and plan cash and fixed costs for the low case. Plan ad spend so the business still makes money in the low case.
How do you check and improve the forecast?
The forecast is only as good as the habit of checking it. At the end of each month, compare forecast and actual for each driver, not just the total.
Forecast error % = (actual − forecast) ÷ forecast
| Driver | Forecast | Actual | Error | Lesson |
|---|---|---|---|---|
| Returning revenue | $56,000 | $52,000 | −7% | Repeat rate was 3.7%; recent cohorts weaker than assumed |
| Paid new customers | 1,200 | 1,050 | −13% | CAC was $57, not $50 |
| Organic new customers | 600 | 640 | +7% | Close |
| Total revenue (launch delivered as planned) | $174,000 | $163,400 | −6% | Both misses came from cost and retention, not demand |
The total error is 6%, but the driver view shows where it came from. Acquisition cost was higher than planned and retention was weaker. Those are two different problems with two different fixes: rethink spend levels or channels, and look at the repeat purchase rate of recent cohorts. A total-only forecast would have only told you “sales were low.”
Keep a log of forecast errors by month. After six months, you will see whether your model runs high or low and on which driver, and you can correct it.
How does the forecast connect to the rest of the business?
The forecast becomes useful when other plans read from it:
- Inventory: convert revenue by product line into units, then into purchase orders with lead times.
- Ad budgets: check the planned spend against a target marketing efficiency ratio or a maximum acceptable new customer CAC.
- Retention: if returning revenue is the weakest driver, invest in lifecycle email and SMS before adding more ad spend.
- Reporting: track forecast versus actual in the same dashboard as your other core metrics. Our guide to eCommerce KPIs lists the ones that belong next to it.
Checklist: build your first forecast
- [ ] Export 24 months of orders with customer ID, date, net value, discounts and refunds
- [ ] Mark each order as first or returning
- [ ] Count active customers (bought in the last 12 months) at the start of each month
- [ ] Calculate monthly repeat rate and returning order value for each calendar month
- [ ] Calculate recent new customer CAC at your current spend level
- [ ] Estimate organic new customers per month from low-spend periods or last year
- [ ] List known events, calendar shifts and stock limits, each with a reason
- [ ] Build low, base and high scenarios
- [ ] Compare forecast to actual by driver at each month end
- [ ] Log errors and adjust assumptions every quarter
FAQ
Do we need forecasting software or machine learning?
Not to start. Most stores get more value from a driver-based spreadsheet they understand than from a model they cannot explain. Statistical tools become useful when you forecast many products or regions at once, and they work best when fed the same clean data this model needs.
How far ahead should an eCommerce forecast go?
Forecast monthly for the next 12 months, and update it every month. Near months can be detailed by product line, while months further out can stay at the level of returning and new customer revenue. Inventory lead times usually decide how far ahead you need detail.
What if we do not have two years of data?
Use what you have, and lean more on the drivers you can measure now: current active customers, recent repeat rates and recent CAC. Without a full year of history, seasonality is the weakest part, so keep wider ranges between your low and high scenarios until you have more data.