Your LMS already holds better data than any indust
A colleague asked me last spring what a good course completion rate looks like. I gave her a number I had read somewhere, she put it in a deck, and it went to her board. About a week later I tried to find the source again and could not. The number traced back to a vendor blog citing another vendor blog citing a report from 2019 that no longer resolved. I have no idea whether it was ever true. It is probably still in that deck.
That is the normal state of benchmarking in Learning and Development. We quote engagement averages and completion rates with a confidence nobody has earned, mostly because the alternative feels like a research project we do not have time for. It is not. Most L&D teams are already sitting on data that beats anything they could borrow, and the work of turning it into something publishable is smaller than it looks.
Borrowed Benchmarks Describe Someone Else's Company
The problem with an industry average is not that it is wrong. It is that you cannot tell what it is measuring. A completion rate pulled from a vendor report might come from compliance modules at a bank, where completion is mandatory and the number means nothing, or from optional skills content at a startup, where the same number means quite a lot. Both get averaged into one figure that describes neither.
Your own numbers do not have that problem. You know what your modules are for, who was required to take them, and what was happening in the business that quarter. A completion rate you calculated yourself comes with all the context that makes it interpretable, which is exactly what gets stripped out when a statistic starts traveling.
What I Learned By Running One
I do not work in L&D. I run a digital PR and link building agency, and earlier this year we published a study of our publisher network, State of Link Building and Brand Mentions 2026, covering 16,625 sites in 53 languages. It is a different subject entirely, but the process of building it taught me things that apply to any team publishing internal data for the first time.
The first was that our data had been sitting there for nine years. We built the database at ESBO Ltd to do the job, not to study it, and it took an embarrassingly long time to notice that a decade of pricing records was itself a finding. L&D teams have the same asset in their LMS. Nobody set it up to answer research questions, which does not stop it from answering them.
The second was that the study corrected me in public. I had written a few months earlier that publishers charge two to five times more to carry restricted content. The actual median surcharge came out at 24%. My guess had been the top of the range, not the norm, and the correction is in the report because leaving it out would have made everything else less trustworthy. Expect your own data to do this to you. If it confirms everything you already believed, check your method rather than celebrating.
The third was that the discipline mattered more than the conclusions. Putting a sample size next to every single number sounds fussy until you try it, at which point you discover which of your claims rest on 4,000 records and which rest on eleven. That exercise alone will improve how your team talks about learning data, whether or not anyone outside the company ever reads the result.
Four Rules That Keep A Small Study Honest
State your sample and your filters before you state your finding. "Completion for our 14 optional skills modules, 2,180 enrollments, employees with more than six months tenure" is a sentence that lets a reader judge you. "Our completion rate is 71%" is not.
Use medians instead of averages for anything money or time related. One executive program at forty times the cost of everything else will pull an average somewhere useless, and the median will not move.
Say what your data cannot see. Ours over-represents publishers that sell placements, and we wrote that in the report. Yours probably over-represents whoever remembers to mark a module complete. Naming the limit costs you nothing and protects you from the one reader who spots it.
Write down the method precisely enough to repeat next year. A single-year figure is trivia. The same figure measured the same way three years running is the thing people actually want.
What This Buys You
Internally, it changes the conversation with finance. A CFO who ignores an industry engagement statistic will engage with a number drawn from your own people, because it describes a population they recognize and can argue with. Arguing is fine. It means they are treating the number as real.
Externally, it does something L&D teams tend to undervalue. Original data is the most quotable thing a team can produce, and it travels further than opinion pieces do. That has become more true as AI assistants have started mediating what people find. Ahrefs studied AI Overview visibility across 75,000 brands and found branded mentions correlated with visibility at 0.664 against 0.218 for backlinks, and Muck Rack's analysis of over 25 million AI-cited links found earned media accounted for 84% of citations while paid content accounted for 0.3%. The practical version for an L&D team: a study with your organization's name on it gets cited and repeated in a way that a sponsored post never will. Brand mentions built on real evidence outlast the campaign that produced them, which is most of what our clients at ESBO Ltd are buying when they buy digital PR.
Start With One Question
Do not design a research program. Pick one question you already argue about internally and answer it properly. Whether managers who complete a program have different retention on their teams a year later. Whether the modules people finish are the short ones or the relevant ones. Also, whether your instructor-led sessions actually beat the recordings once you control for who chooses which.
Pull the data yourself, write down what you filtered out, and put the sample size next to the answer. If the finding is boring, you have learned something anyway and it took you a week. If it is not boring, you have something nobody else in your industry can quote from anyone but you.
