The most-cited sales training statistic — that 87% of what reps learn is forgotten within 30 days — comes from a 2016 vendor blog post, not a study. That post said 12 weeks, not 30 days, and credited unnamed Xerox research. Here are the statistics in this category that do hold up, each with its source.

The oldest version of the 87% claim we could locate is a post published 15 September 2016 by Ralph Grimse. It says “87% of the new knowledge learned in sales training is lost within 12 weeks” and credits “Xerox research.” No publication, author, year, sample size or method is named.

• Twelve weeks became thirty days somewhere in the retelling. Pages repeating “87% within 30 days” today cite either nothing at all or “the forgetting curve” in general.

• The forgetting curve everyone cites rests on two experiments. Each used exactly one subject memorizing nonsense syllables: Hermann Ebbinghaus in 1885, and a 2015 replication by Jaap Murre and Joeri Dros.

• A review of 14 research articles covering 69 experimental conditions and more than 1,000 learners found forgetting ranging from 0% to 94%. There is no single rate.

• A meta-analysis of 117 studies found that factual knowledge decayed over time while skills and on-the-job behavior held steady or improved.

Where the 87% statistic comes from

1. Why the number appears as 30 days

2. What the forgetting curve research measured

3. How much people actually forget

4. Whether skill decays like knowledge

5. What the research says makes training stick

6. What we found on our own site

7. What to ask a vendor

8. Frequently asked questions

It comes from a vendor blog post, not a study. The earliest instance we could find is “Information Overload: Closing the Last Mile in Sales Training” by Ralph Grimse, published 15 September 2016 and now hosted at sbigrowth.com. It states that “87% of the new knowledge learned in sales training is lost within 12 weeks,” attributed to “Xerox research.” No paper, author, year, sample size or method is given.

We searched for a published Xerox study matching that claim and could not find one. The same post carries a second figure — “84% is lost after only 90 days,” credited to Sales Performance International — which is equally unsourced. That is why the two numbers so often travel together.

Because the timeframe shrank in transit. The original claim was 12 weeks. The version circulating now is 30 days. That is a materially different assertion about the same figure, and nobody appears to have noticed the change.

One current page, titled “Sales Training Programs: Why 87% of Training Is Forgotten Within 30 Days”, offers this as its support: “This isn’t just a theory; it’s based on the well documented forgetting curve.” That is a gesture at a body of work, not a citation.

The same thing has happened to the 70% claim. One page published in December 2025 says employees “forget up to 70% of new information within 24 hours,” and credits it to Ebbinghaus, “validated by Murre & Dros (2015).” The same number circulates elsewhere as 70% within a week. A figure that changes its timeframe by a factor of three or seven while keeping its precision is folklore. We are not the first to say so — Clario published a similar correction in July 2026.

One person, memorizing nonsense. Hermann Ebbinghaus ran his 1885 experiments on himself — he was the only subject in all of them — using consonant-vowel-consonant syllables like WID and ZOF. He measured not recall but “savings,” the time saved relearning a list the second time.

Murre and Dros replicated the work in PLOS ONE in 2015, also with a single subject: Joeri Dros, the paper’s second author, who learned 70 lists of 104 nonsense syllables each. His savings score at 31 days was 0.041. Because savings measures retention, that reads as roughly 96% forgotten — not 4.1% forgotten, which is how the number is often repeated.

The authors flag the result themselves. Correcting for a steady increase in learning time over the study, they write, “the corrected savings measure would be 0.137 for the 31 day interval instead of 0.0410. This, however, is still well below the values for the three others, which are in the 0.20 range.” Their own paper treats the figure as an outlier against the comparison data.

None of that transfers cleanly to a rep learning an objection-handling sequence. Nonsense syllables are designed to carry no meaning, no structure and no prior association — the three things that make real material memorable. A percentage drawn from that setting cannot be applied to sales content without saying so.

Anywhere between none of it and nearly all of it, depending on what was learned and how. Will Thalheimer’s research report How Much Do People Forget? (Work-Learning Research, 14 December 2010) reviewed 14 research articles covering 69 experimental conditions and more than 1,000 learners.

Forgetting across those conditions ranged from 0% to 94%. In the one-to-two-day range alone it ran from 0% to 73%. Thalheimer’s conclusion is blunt: “Rules-of-thumb that show people forgetting at some pre-defined rate are just plain false.” Any vendor quoting a single retention percentage as a general fact is quoting something the evidence does not support.

No, and this is the part the folklore gets backwards. Taylor, Russ-Eft and Chan’s meta-analysis of behavior modeling training (Journal of Applied Psychology, 90(4), 692–709, 2005) pooled 117 studies across six training outcomes.

Their finding on decay is stated plainly: “Although BMT effects on declarative knowledge decayed over time, training effects on skills and job behavior remained stable or even increased.” So the thing that fades is the product briefing. The thing that holds is the behavior someone practiced. Vendors quoting an invented knowledge-loss number to sell behavior change are arguing against their own product.

Practicing retrieval, and spacing it out.

In Roediger and Karpicke’s test-enhanced learning experiments (Psychological Science, 2006), Experiment 1 gave 120 undergraduates a prose passage to either restudy or be tested on. Five minutes later the restudy group was ahead, 81% to 75%. A week later that reversed: the tested group recalled 56% against the restudy group’s 42%. Their second experiment used 180 undergraduates. The group that studied once and then took three tests read the passage an average of 3.4 times. The group that only restudied read it 14.2 times, and still recalled less a week later: 40% against 61%.

On spacing, Cepeda and colleagues (Psychological Bulletin, 132(3), 354–380, 2006) synthesized 839 assessments of distributed practice across 317 experiments in 184 articles, and found spaced study reliably beat massed study.

Both bodies of work were run on students learning verbal material, not reps learning discovery calls, and we would rather say that than overstate the fit. What carries across is the mechanism: producing an answer from memory beats reviewing it, and doing that repeatedly over weeks beats doing it once.

It did not pass everywhere. A call-to-action banner on our own sales training ROI page claims “Reduce Ramp-Up time by 47%” with no source attached. Our published case study on the same subject reports something different. Ramp fell from 90 days to 15 at India’s largest online insurance aggregator, a 10,000-person deployment. That is an 83% reduction, not 47%. We can source the second number. We cannot source the first, and it is being removed.

Elsewhere on our blog, figures are credited to Gartner, Forrester, CSO Insights and McKinsey by firm name alone, with no publication, year or sample size. That is the same failure this article is about, at a smaller scale, on our domain.

We are naming it because a page arguing for better sourcing has no standing otherwise. We also want to be clear about what we have not published. We score real customer calls, so we could measure retention on a real sales team and report it. We have not done that yet. Until we do, we have no retention percentage of our own to offer, and we are not going to borrow one.

Ask for the study. Specifically: who published it, in what year, with how many participants, and what was measured — recall, savings, skill performance or business results.

Those four questions eliminate most of the numbers in this category in under a minute. If the answer is a link to another vendor’s blog, you have found the end of the trail rather than the source. The same test applies to outcome claims, including ours: a number without a sample size, a date and a method is not evidence, whoever is presenting it.

Is the Ebbinghaus forgetting curve wrong?

Not wrong, but much narrower than its popular use suggests. Ebbinghaus established that memory for newly learned material declines steeply at first and then flattens, and that relearning is faster than first learning. Those findings have held up. What has not held up is the practice of lifting a specific percentage from his curve and applying it to workplace training. His 1885 experiments used one subject — himself — memorizing syllables chosen specifically to be meaningless, and the 2015 replication also used one subject. A curve produced under those conditions describes the shape of forgetting, not its rate for a rep learning a product objection.

How long do sales reps actually remember what they learn in training?

There is no published figure for this, and anyone offering one is estimating. The closest honest answer comes from Thalheimer’s review, which found forgetting ranging from 0% to 94% across 69 experimental conditions, depending on the material, the learner and the delay. What the evidence does support is a distinction: factual recall — product specs, pricing tiers, competitor positioning — decays fastest, while practiced skills hold up considerably better, per the Taylor meta-analysis of 117 studies. If you want a number for your own team, the only sound way to get one is to measure it on your own team.

Does AI roleplay solve the retention problem?

It addresses the two mechanisms the research supports, which is not the same as solving a problem the research cannot even size. Retrieval practice and spacing both require reps to produce answers from memory, repeatedly, over weeks — and the constraint on doing that has always been manager time, not willingness. AI roleplay removes that constraint by making practice available without a second person present. What it does not do is make a forgetting percentage disappear, because no credible forgetting percentage for sales training exists to begin with. Judge it on whether reps practice more often and perform better, not on a statistic.

Why do so many sales training companies repeat these numbers?

Because the numbers are useful and nobody checks. “87% forgotten in 30 days” makes a compelling case for buying reinforcement, which is what most of the sites repeating it sell. The citation chain runs vendor blog to vendor blog, and each repetition makes the next one look better sourced. We have published category statistics ourselves without tracing them, which is part of why we went looking. The practical fix is the four-question test above, applied evenly — to competitors, to analysts, and to us.

The honest summary of the sales training statistics in circulation is short. The most-cited one has no source. The research behind the forgetting curve rests on two single-subject experiments with nonsense syllables. The best available review found forgetting rates anywhere from 0% to 94%.

What the evidence does support is narrower and more useful: knowledge decays, practiced skill largely does not, and retrieval spread over time is what moves material from one category to the other. If you are choosing how to build sales capability, that distinction is worth more than any percentage on this page.

Most teams ration practice by manager availability. If you want to see what it looks like when every rep has it instead, see what AI sales roleplay involves or ask for a walkthrough with your own scenarios.