Forecasting Sales for a New Product With No History
By Newtown Analytics · · 7 min read
For ecommerce, marketing & growth leaders
A new product sales forecast has no sales history to extrapolate, so it has to be built from other evidence: how similar products performed, how many people you can reach, how many will try the product and how many will buy again. The goal is not a precise number but a defensible range and a plan to replace guesses with actuals quickly. This guide covers four approaches (analogs, funnel, market sizing and test-market scaling), works through numerical examples of two of them, shows how to set low, base and high cases, and explains how to link the forecast to the first inventory order and to cash. It is part of budgeting and forecasting and applies the methods in how to choose a forecasting method.
Why new products are different
| Challenge | Consequence |
|---|---|
| No history | Time-series methods cannot be used |
| High uncertainty | Ranges are wide, so a single number is misleading |
| Optimism | Product teams tend to overestimate. See forecast bias in financial planning |
| Cannibalization | Part of the sales may come from your existing products |
| Supply decisions are made before demand is known | Over-ordering ties up cash; under-ordering loses sales |
| Early data are noisy | Launch-week spikes mislead |
Approach 1: Analogs
Find past launches that resemble the new product in category, price, channel and target customer, and use their sales curve as the model.
Worked example. A similar product sold 2,000, 3,500, 4,200 and 3,800 units in its first four months (13,500 in total). The new product will launch with fewer stores and a smaller marketing budget, and the team judges its reach at about 70 percent of the analog's.
| Month | Analog units | Scaled by 0.70 |
|---|---|---|
| 1 | 2,000 | 1,400 |
| 2 | 3,500 | 2,450 |
| 3 | 4,200 | 2,940 |
| 4 | 3,800 | 2,660 |
| Total | 13,500 | 9,450 |
Check: 13,500 × 0.70 = 9,450. Use several analogs when possible, and look at the spread between them; if the three best analogs sold 8,000, 13,500 and 21,000 units in four months, a forecast of 9,450 sits near the lower end, and the range is the real finding. Document why each analog was chosen and what differs.
The weakness of analogs is subjectivity in the choice of analog and the scaling factor. Ask people who did not work on the product to challenge both.
Approach 2: The funnel model
Build the forecast from the steps a customer takes: see the product, visit, buy, and buy again. This works well for direct and online sales, where each step can be measured.
Worked example. A hypothetical product launches online with planned advertising reach of 2,000,000 impressions in the first month.
| Step | Assumption | Result |
|---|---|---|
| Impressions | 2,000,000 | 2,000,000 |
| Click-through rate | 1.2% | 24,000 visits |
| Conversion rate | 2.5% | 600 orders |
| Average order value | $55 | $33,000 |
| Repeat orders within 90 days | 20% of first orders | 120 orders |
| Repeat revenue | 120 × $55 | $6,600 |
Check: 2,000,000 × 0.012 = 24,000; 24,000 × 0.025 = 600; 600 × $55 = $33,000; 600 × 0.20 = 120; 120 × $55 = $6,600. First-quarter revenue from this launch cohort is about $39,600. These rates are placeholders. Use your own data from similar campaigns, or from small tests, to set them, and show a range.
The funnel makes the drivers explicit: if conversion is 1.5 percent instead of 2.5 percent, orders fall to 360, a 40 percent shortfall. A small paid test before the main launch can measure click-through and conversion for a few hundred dollars and replace the largest guesses.
Approach 3: Market sizing
Estimate the number of potential buyers, the share you can reach and the share who will buy.
Potential buyers × Reach × Purchase rate × Units per buyer per year = Annual unitsIllustration: 400,000 potential buyers × 15% reached × 8% buying × 1.5 units per year = 7,200 units. (400,000 × 0.15 = 60,000; × 0.08 = 4,800; × 1.5 = 7,200.) Market sizing is useful as a sanity check on the other approaches. If a funnel model implies 50,000 units from a market that sizes at 7,200, one of them is wrong. See top-down vs. bottom-up revenue forecasting.
Approach 4: Test market and scaling
Launch in a limited region, channel or customer group, measure sales per store, per visit or per customer, and scale. Control for differences between the test and the full market (store mix, income, awareness), and watch the repeat rate, which takes time to emerge. Pre-orders, waitlists and pilot customers provide similar early evidence for products where a test is practical.
Approach 5: Trial and repeat for consumable products
For products bought repeatedly, forecast first purchases (trial) and repeat purchases separately, because they follow different logic.
Worked example. A hypothetical household product targets 500,000 households.
| Step | Assumption | Result |
|---|---|---|
| Target households | 500,000 | 500,000 |
| Awareness after launch campaign | 30% | 150,000 |
| Trial among aware households | 10% | 15,000 |
| Repeat rate (share of triers who buy again) | 35% | 5,250 repeaters |
| Repeat purchases per repeater in year 1 | 4 | 21,000 |
| Year 1 units | 15,000 + 21,000 | 36,000 |
If the repeat rate is 25 percent instead of 35, repeaters fall to 3,750, repeat units to 15,000 and total units to 30,000, which is 16.7 percent lower. The repeat rate is hard to know before launch, so it deserves a pilot or a conservative assumption. It also decides whether the product has a future: trial can be bought with advertising, but repeat has to be earned by the product. If trial is strong and repeat is weak, the forecast will show a spike and a decline, and inventory should not be built for the spike.
Scale the process as the product matures
A launch forecast is a bridge to a normal one. Once you have a few months of sales, move to the driver approach in how to forecast ecommerce sales for online channels, and to a structured assumptions model such as a bottom-up revenue model when the product becomes a line in the annual plan.
Combine approaches into a range
Use at least two methods and look at where they agree.
| Case | Four-month units | How it is set |
|---|---|---|
| Low | 5,500 | Weakest analog scaled by 0.70, funnel with conversion at 1.5% |
| Base | 9,450 | Analog scaled by 0.70, consistent with the funnel |
| High | 14,000 | Analog at full reach, strong repeat |
(Illustrative.) Plan inventory, staffing and cash around the low and base cases, and define in advance what you will do if sales track the high case: expedite production, release more advertising budget. A forecast with an agreed response plan is more valuable than a precise number.
From units to the first purchase order
New products have a particular supply problem: the lead time to produce is longer than the time to learn demand. A framework:
- Order enough for the low case plus a buffer for the lead time of a re-order.
- Negotiate a smaller first order with a quick-turn re-order option if suppliers allow.
- Set re-order triggers based on weekly sales against the forecast range.
- Plan the markdown or exit if sales track below the low case.
See demand planning and inventory optimization for safety stock and lead time logic.
Replace guesses with actuals quickly
| When | Action |
|---|---|
| Week 1 to 2 | Do not extrapolate launch spikes; check conversion and traffic against assumptions |
| Week 4 | Re-forecast using actual funnel rates and early repeat data |
| Month 3 | Compare with the analog curve; adjust scaling |
| Month 6 | Move to standard methods once there is enough history |
Set triggers: for example, re-forecast if cumulative sales are outside the low-high range for two consecutive weeks.
Cannibalization and incremental margin
If part of the new product's sales replaces existing sales, the incremental revenue is smaller than the forecast. Suppose the new product sells 9,450 units at $55, but customers would otherwise have bought an existing product at $50 for 30 percent of those units. Gross new revenue is 9,450 × $55 = $519,750. The displaced existing sales are 9,450 × 0.30 × $50 = $141,750. Incremental revenue is $519,750 - $141,750 = $378,000. Evaluate the launch on incremental margin, not gross sales. See unit economics.
Cash
A launch consumes cash before it produces it: development, tooling, first inventory, marketing, then sales, then collections. Build a cash forecast for the launch period. See cash flow management.
Common mistakes
- Using the best analog only.
- Forecasting a single number.
- Extrapolating launch-week spikes.
- Ignoring cannibalization.
- Ordering for the high case with no way to re-order.
- Not tracking actuals against the forecast from day one.
- Not documenting assumptions.
FAQs
How do you forecast sales for a new product?
Use analogs from similar launches, a funnel model, market sizing and test-market evidence, combine them into a low, base and high range, and update quickly with actuals.
What is an analog forecast?
A forecast that uses the sales pattern of a similar past product, scaled for differences.
How do I decide the first order quantity?
Cover the low case plus a buffer, negotiate quick re-orders, and set triggers based on early sales.
How accurate are new product forecasts?
Less accurate than forecasts for established products. Use ranges and update frequently.
Should I include cannibalization?
Yes. Incremental sales and margin matter more than gross sales.
Keep reading
This article is for general information. It is not accounting, tax, legal or investment advice, and examples use illustrative numbers rather than client data. Confirm decisions that carry financial or legal consequences with a qualified professional.