How to Forecast Trial and Repeat for a New Product (2026)

To forecast trial and repeat for a new product, estimate two separate numbers before launch: the trial rate (what share of your target buyers will buy it once) and the repeat rate (what share of those triers come back). Multiply both by a reachable market base, express the result as base, upside and downside scenarios, then re-forecast at weeks 1, 2, 4 and 8 as real purchase data arrives.

The reason for splitting it comes down to how launches actually die. A product can beat its trial target and still fail commercially, because the buyers were persuaded by sampling and launch discounting and never repurchase at full price. One blended demand number hides that.

This guide walks the full method, with the arithmetic shown at every step.

Short answer: build a baseline from the closest analog product, adjust it for price, position, pack size and distribution, convert pre-launch research signals into a trial assumption, apply a purchase-cycle model to get the second-purchase number, then present base, upside and downside scenarios rather than a single figure. Re-forecast once real repeat data starts landing, because financials lag behavior by months.

What You Need

Before touching a model, be clear about which decision the forecast is supposed to inform. A forecast built to justify an inventory buy is a different object from one built to set a marketing budget, and a forecast built to decide whether to kill a product is a third. Naming the decision keeps the forecast honest, because each one has a different tolerance for error and a different refresh schedule.

You also need a shared vocabulary. Most launch arguments that look like data disagreements are definition disagreements, where one person means a customer’s second purchase and another means annual repurchase. Writing the metric definitions down before the numbers go in kills that problem cheaply.

Metric	Formula	Typical window	Where it breaks
Trial rate	Target buyers who buy at least once / target buyers in the launch scope	6 to 12 weeks	Counts promotional and sampling buyers who never intended to repurchase
Repeat rate (repeat purchase rate)	Customers placing a second order / customers in the launch cohort	90 to 180 days	The window is shorter than the purchase cycle, so repeat looks artificially low
Customer retention rate	Active customers this period / active customers last period	Monthly for subscriptions, quarterly otherwise	Blends genuine churn with customers who simply paused
Time to repeat (purchase cycle length)	Median days between first and second purchase, by cohort	Measured per cohort, not averaged across the catalog	Irregular replenishment makes the median meaningless for a minority of buyers
Cumulative trial (penetration)	Buyers who have tried the item to date / category buyers to date	3 to 5 years	Understates penetration when distribution is uneven across stores or regions
Customer lifetime value	Repeat orders per customer multiplied by contribution per order, discounted	Multi-year	Inherits every error in the trial and repeat assumptions feeding it

Alongside the definitions, you need nine inputs: trial rate, repeat rate, purchase cycle length, retention curve shape, category market size, distribution reach, price, competitor benchmarks, and the evidence you actually have rather than the evidence you wish you had.

The last one matters most. Sort your inputs into three buckets: panel or retailer scan data (treated as ground truth), research you have run yourself (concept tests, taste tests, conjoint, and so on), and assumptions with no data behind them. Every forecast keeps all three visible, because the third bucket is where the arguing happens.

Step-by-Step

Here is the seven-step sequence in compressed form, then each step in full.

1. Define trial and repeat for the product.
2. Estimate the addressable trial opportunity.
3. Estimate first-to-second purchase conversion.
4. Build a repeat-purchase curve.
5. Calculate trial, repeat and revenue scenarios.
6. Stress-test the assumptions.
7. Validate and update the forecast after launch.

1. Define Trial and Repeat for the Product

Define the trial event and the repeat event in writing, and state whether each measure counts customers or transactions. The distinction changes every number downstream. A second transaction from one household is not a second customer, and in a product bought by couples or families that gap can be large.

Then state the purchase cycle, because it sets the length of your measurement window. A 12-week repeat window is meaningless for a product whose buyers replenish every six months, and generous for a consumable that repurchases in ten days.

• Consumables (food, drink, cleaning, personal care): trial is the first unit, repeat is the second purchase within one to two purchase cycles. Short windows, repeat arrives fast.
• Durable goods (appliances, furniture, tools): repeat is rare by design. Model attachment and accessories separately, because the second purchase is often a different product entirely.
• Subscriptions (SaaS, media, memberships): trial is the signup or free period, repeat is the first renewal or the first paid period after trial ends. Churn between trial end and month two is the number that matters.
• Delayed-repurchase categories (pet consumables with veterinary cycles, durable gear, seasonal purchases): the forecast horizon may need to extend past a year before the repeat signal is readable.

How to tell it worked: a colleague who was not in the room can read your definitions and calculate your metrics without asking a follow-up question.

2. Estimate the Addressable Trial Opportunity

Size the category market, then shrink it to the audience you can actually reach. Market potential and forecastable demand are different quantities, and using the first in place of the second is the single most common way new-product forecasts run 5 to 10 times too high.

The shrinkage usually comes from three filters: distribution reach (how many stores, locations or traffic sources you will have), mental availability (whether the audience knows the product exists and knows when to buy it), and category eligibility (buyers who already purchase in the category at all).

Apply your first trial-rate assumption to the reachable base, not the total market. Physical and mental availability are what drive trial in most categories, which is why two products with identical formulas and identical ratings can have wildly different first-year trial rates.

A practical shortcut for the trial assumption: start from the analog’s observed cumulative trial curve, note how far along it was at the same number of weeks post-launch, and treat that as your base case. Then adjust down for weaker distribution and up for stronger awareness.

Adjustment factor	Direction	What to check before moving the number
Price relative to the analog	Higher price cuts trial, lower price lifts it	Whether the price sits above or below the category’s reference price point
Position in the store or site	Eye-level and category adjacency lift trial	Where the item actually lands, not where the plan says it lands
Pack size or tier	Larger sizes lift trial and slow repeat; smaller sizes do the reverse	Whether the trial purchase can plausibly be the whole purchase
Distribution and numeric distribution	Fewer outlets, fewer facings, less cold space, lower trial	Numeric distribution as a share of the category’s total
Launch promotional depth	Deep discounting inflates trial and depresses measured repeat	Whether launch pricing ends before your repeat window closes
Awareness and mental availability	Distinctive assets and clear occasions lift trial	Whether buyers can name the brand when asked
Cannibalization	Skews trial toward existing buyers and away from new category buyers	Whether the item line extends or creates

How to tell it worked: you can name which stores, sites or channels are in scope and what share of the category they represent, and your trial number is defensible from that base alone.

3. Estimate First-to-Second Purchase Conversion

This is where most forecasts quietly go wrong, because teams apply a repeat rate borrowed from a different category, a different channel and a different time window. Benchmark selection matters more than the benchmark itself.

Work through six factors when setting the assumption:

• Benchmark fit. Match on category, channel, price tier and purchase cycle, not just on product type.
• Purchase cycle length. A rate measured over 90 days cannot be applied to a product whose buyers take 210 days to return.
• Satisfaction signals. Taste tests and review returns tell you whether the first purchase was good enough to repeat. Stated preference is a weak signal; observed repeat from the analog is a strong one.
• Habit formation. Products with a recurring occasion (breakfast, a daily commute, a weekly clean) repeat far better than products bought on a one-off occasion.
• Price sensitivity. If the trial purchase came at a heavy discount, the repeat decision is a different decision at a different price.
• Repeat versus retention. Repeat is a one-way trip from the first purchase. Retention describes an ongoing relationship. Do not borrow a subscription retention figure for a physical product.

For genuinely novel products with no analog, you are estimating rather than borrowing, and you should say so. Triangulate from the closest behavioral neighbor, the occasion it attaches to, and the trial price versus the repeat price. Then widen the confidence band deliberately, because novelty is exactly where analog methods fail.

How to tell it worked: you have written down where the number came from, and you have flagged it as either evidence-based or judgment-based.

4. Build a Repeat-Purchase Curve

Repeat is a curve, not a single number, and the shape tells you when to look. Start with first-to-second conversion, then extend to third and fourth purchases using declining rates, because the share of triers who make a third purchase is always smaller than the share who made a second.

Two structural features matter most. Repurchase intervals cluster around the purchase cycle, so repeat arrives in waves rather than as a smooth trickle. And depletion cuts in: once the most enthusiastic buyers have bought and come back, the remaining pool is progressively harder to convert, which is why penetration curves flatten rather than run forever.

Frequent buyers and occasional buyers are different populations. A small group buys weekly and dominates early unit sales; a much larger group buys quarterly and determines whether the item has a long tail. Segment the cohort if your data allows it, because a single average repeat rate hides the split that actually drives inventory planning.

One more mechanic: launch discounting shortens the first purchase cycle and lengthens the second. Where that is true, your measured repeat rate will look worse than the product’s steady-state rate, and the correction is to note the promotional calendar rather than to assume a product problem.

How to tell it worked: you can state the expected timing of the second purchase wave, which tells your team when to measure rather than when to worry.

5. Calculate Trial, Repeat, and Total Revenue Scenarios

Now do the arithmetic. Here is a worked example for a mid-priced consumable launching in 400 grocery stores. The numbers are illustrative planning figures for demonstration, not published category averages; your own panel and scan data replaces every one of them.

• Category buyers in the 400 stores: 2,000,000
• Numeric distribution against the category: 12%
• Reachability factor (awareness, shelf presence, consideration): 55%
• Addressable trial base: 2,000,000 multiplied by 0.12 multiplied by 0.55, or 132,000 buyers
• Base trial rate: 18%, giving 23,760 first-time buyers
• Purchase cycle: 45 days
• Base first-to-second conversion: 34%, giving 8,078 repeat buyers in the first two cycles
• Base repeat orders per converted customer over 12 months: 5.5

Base case volume: 23,760 first purchases plus 8,078 customers returning, each averaging 5.5 orders, or roughly 44,429 orders in the first 12 months.

Now vary the two assumptions that carry the most weight.

• Downside: distribution slips to 9% numeric, trial rate falls to 12%, and conversion holds at 26% because repeat price is higher relative to competitors. Addressable base falls to 99,000, first-time buyers fall to 11,880, and repeat buyers fall to 3,089.
• Upside: distribution holds at 12%, trial reaches 24% on stronger launch support, and conversion improves to 40%. First-time buyers rise to 31,680 and repeat buyers rise to 12,672.

The spread between downside and upside is roughly a 3x range on volume, and that spread is the honest output. A single number drawn from the middle of a 3x range is a decision to look confident, not a forecast.

Revenue follows from units multiplied by price, and revenue quality follows from the mix between discounted first purchases and full-price repeats. A launch that hits its unit target while half of year-one volume came at launch discount is a weaker commercial result than the same units look on paper.

Where your product has no price story, keep this section in units and let finance apply the margin assumptions. Planners argue more about revenue than volume, and volume is the number they can actually check against scans.

How to tell it worked: someone can trace any headline figure back to a named assumption, and no figure in the summary exists only in the summary.

6. Stress-Test the Assumptions

Identify which assumptions the result actually depends on. In most new-product models it comes down to six: trial rate, first-to-second conversion, purchase cycle length, price, distribution reach and retention decay.

Run sensitivity analysis by moving one assumption at a time and recording the effect on volume. If a 5-point change in conversion swings volume more than a 10-point change in trial rate, conversion is where your research budget belongs. That one table decides where to spend the next month.

Then ask what evidence could move each assumption, since an assumption nobody can test is just a preference in a spreadsheet. Trial rate responds to concept and taste tests, awareness studies and observed distribution data. Conversion responds to panel purchase histories and review and return data. Cycle length responds to replenishment-interval data from the analog or from a comparable subscription.

Search volume and social listening are directional signals at best. They tell you attention exists, not that a purchase will follow, and using them as a units input is how a forecast ends up anchored to a number no one can explain.

Finish by stating confidence out loud. A launch forecast built on one analog, in one market, at one price point, deserves a wider band than a forecast built on five analogs with matched distribution. Naming the band is what stops a cautious forecast from being overruled in a meeting by a confident one.

How to tell it worked: you have a ranked list of assumptions by leverage, and a written statement of how wide your band is and why.

7. Validate and Update the Forecast After Launch

Monitoring is not the same as watching revenue. Financials are lagging indicators, and by the time repeat weakness appears in a revenue report the cost of fixing it has usually already been spent.

Checkpoint	What to check	Decision it triggers
Week 1	Distribution actually delivered against plan, in-stock rate, facings, first-week trial rate	Fix the availability problem before judging demand; do not revise the trial assumption yet
Week 2	Repeat rate against the analog curve, promotional dependency, off-promo purchase share	Revise the conversion assumption; decide whether launch discount continues past the first cycle
Week 4	Time to repeat against the assumed cycle, review and return rates, search and awareness metrics	Reset purchase cycle length and the repeat curve shape
Week 8	Cumulative trial penetration versus the analog curve, distribution gaps, cannibalization reading	Re-forecast the full year, or scale back the rollout
Month 6 and 12	Depletion rate, repeat orders per customer, retention decay	Grade forecast accuracy and feed the result into the next launch model

Learn to separate learning lag from genuine failure. Early weeks understate repeat simply because the second purchase wave has not arrived for buyers whose cycle is long. If you measure a 45-day product at day 20, low repeat is arithmetic, not evidence.

Distribution gaps masquerade as demand gaps more often than teams admit. A product that is not on shelf or not visible cannot be trialed, and a low trial rate in week one usually means a delivery problem until proven otherwise.

Small and fast-moving brands get around the evidence problem by launching deliberately narrow, in a limited number of stores or one region, so trial and repeat can be measured on a controlled sample before the rollout scales. It costs reach and buys something rarer, which is a real read on repeat before the money is committed.

Finally, grade the forecast once the year closes, and keep a running log of which assumptions were wrong and by how much. A launch log is the only thing that compounds; without it, every forecast starts from scratch and inherits the same optimism that hurt the last one.

Watch for the pattern that shows up repeatedly in practice: financial performance is a lagging indicator of trial and repeat, and by the time it reports the problem, the corrective window has closed.

How to tell it worked: each checkpoint has a named owner and a decision attached to it, rather than a report that gets circulated.

Common Mistakes

These are the errors that most often turn into a bad launch decision, with the correction for each.

• Confusing repeat rate with retention. Repeat is a one-way conversion from the first purchase. Retention describes an ongoing relationship. Applying a subscription retention figure to a physical consumable inflates the forecast badly.
• Using total market size as accessible demand. Nobody reaches the whole category. Multiply by numeric distribution and reachability before applying any trial rate, or the number is fantasy.
• Ignoring the repurchase cycle. A repeat rate measured over 90 days tells you nothing about a product whose buyers return every six months. Match the window to the cycle.
• Using benchmarks without category context. A benchmark from another category, price tier or channel is decoration. Match on category, channel, price and cycle, or do not use it.
• Treating one forecast as a promise. A single central number invites anchoring and hides the range you actually believe. Present scenarios and state the band.
• Reading a promotional trial spike as product-market fit. Sampling and launch discounting pull forward buyers who never repurchase at full price. Judge repeat on off-promo volume.
• Letting stated preference stand in for purchase. Concept and taste tests measure what people say they would do. The canonical warning is a product that tested well in blind tasting and then failed once the original was withdrawn and buyers had to choose again at full price.
• Waiting for revenue to prove the model wrong. Revenue arrives months after the behavior that caused it. Watch first-to-second purchase instead.

Diagnostics split cleanly by stage, and knowing which stage failed tells you what to fix.

Symptom	Likely stage	Correction
Distribution below plan, trial low	Trial	Fix availability and facings before touching the demand assumption
Trial on plan, repeat below analog	Repeat	Check occasion fit, off-promo pricing and satisfaction signals
Trial spike, repeat flat	Repeat	Strip sampling and discounting to measure underlying demand
Buyers present, sellers thin	Trial	Fix numeric distribution and trade terms
Strong trial among existing customers only	Trial	Test the item against new-category buyers, not only your base
Second purchase arrives late	Repeat	Reset the cycle length rather than cutting the repeat rate
Revenue misses while behavior holds	Neither	Re-examine price and mix assumptions before the demand model

Frequently Asked Questions

Conclusion

Splitting the forecast in two is the whole idea. Trial and repeat fail for different reasons, on different timelines, and show up in different metrics, so a blended number cannot tell you which one broke.

Start with three things. Write down exactly what counts as a trial event and a repeat event, including the window and whether you count customers or transactions. Build a defensible base from the closest analog, adjusted for price, position, pack size and the distribution you have actually signed for. Then put your assumptions into low, base and high scenarios and state the band out loud.

From there, watch first-to-second purchase instead of revenue, because behavior moves first and the corrective window closes quickly. As of 2026, the habit of grading each launch’s forecast against what actually happened is still the exception rather than the rule, which means teams with a proper launch log are learning considerably faster than teams without one.