The ROI question almost always arrives at the wrong moment. Not at approval, when it could still be answered, but twelve months later, when the budget is being renegotiated and nobody remembers what success was supposed to be measured against.
So this post walks through the year in the order it actually unfolds. What has to be fixed on day one, what is visible after four weeks, what after twelve, and why most attempts to prove value fail in month six rather than month twelve.
Day 1: The number you will need later does not exist yet
It starts with a sentence in the business case. Usually something like "We want to shorten onboarding" or "We want to reduce assembly errors". That is a goal. It is not yet a measure.
If you want to show an effect in month twelve, you need two things today that almost never appear in the business case: a metric collected outside L&D, and its current value. Not the number of people trained, but the number of complaints from plant B, the days until a new hire runs a shift alone, the questions to the specialist team per week. And you need it now, before anything has been rolled out.
That sounds trivial. It is the step most often missing, because it means work in another department and because nobody at project kick-off has time to collect a figure that will only matter in a year. Without that baseline there is no before-and-after later, only a story.
Day one is also when the cost side belongs on paper. Not just licences and production, but the hours of the specialists who supply and approve content. With classic authoring, one hour of e-learning takes 100 to 200 hours of development work (iSpring/Chapman), and a good share of those hours never show up as L&D cost in any budget. The post How much does a training video cost? adds it all up.
Week 4: The first numbers are activity, not impact
After four weeks the LMS produces its first reports. Assigned, started, completed, quiz score. These numbers are reliable, immediately available and almost always presented as success.
They are activity data, though. A high completion rate shows that people finished the course. It does not show that they work differently afterwards. A low rate, on the other hand, is a useful warning: on classic mandatory courses only a quarter of learners finish at all (Continu, 2025). If you are there in week four, look at the content before you talk about impact in month twelve.
What really matters in week four is something else: where people drop out and which quiz questions they get wrong. This is the only phase in which a course can still be fixed cheaply. Let it pass, and in month six you will be measuring the impact of a course you already knew was broken in two places.
Week 12: The metric moves, or it does not
Now the day-one number becomes interesting. After three months, operational topics show an effect if there is one: fewer errors on the line, shorter onboarding, fewer questions to the same few people.
This is the most uncomfortable point of the year, and no amount of writing makes it go away. If the metric moves, you still do not know whether training caused it. A new shift lead, a changed process or a quieter order book can produce the same effect. The clean answer would be a control group, a site or shift trained later. In practice you rarely get one, because nobody wants to explain to a department why it gets the improvement a quarter later.
If a control group is out of reach, an honest workaround helps: ask the managers of the affected areas what share of the change they attribute to the training, and apply that share conservatively. It is not science. It is defensible, as long as you say openly how the number was produced.
Month 6: Where most proofs fail
In month six the metric turns into a calculation. Benefit in euros, minus all costs, divided by costs. That is the familiar Phillips ROI formula, and the arithmetic is trivial. It fails in two other places.
The first is converting to money. Ten percent fewer complaints can be valued, because each complaint has a price. Two days less onboarding can too, through the staff cost of those days. "More confidence with customers" cannot, and you should not force it. Part of the benefit stays qualitative, and a case that admits this is more credible than one that converts everything into euros.
The second is the cost side, and this is the lever that rarely gets discussed. The ROI of a training depends not only on what it achieves but on what it cost to build and keep current. A course that needs four to six weeks of coordination and is re-produced with every process change has to achieve a great deal to pay off.
This is where Omnora's AI Learning Factory comes in. Content is built from an AI-guided interview that takes around 30 minutes of the experienced person's time, and production time per training hour typically drops by up to 80%. Updates run in the same process, quiz and SCORM export included, so the week-four data actually reaches the LMS cleanly. That does not make any training more effective. It lowers the denominator, which is the part L&D genuinely controls. How that adds up across several trainings a year is covered in the post on scaling training production without adding headcount.
One figure reliably turns up in this context: every dollar spent on e-learning returns thirty in productivity (IBM). It is well sourced and still useless for your budget meeting, because it comes from another company. What counts in month twelve is your own number, even if it is smaller.
Month 12: What the budget meeting actually asks
The meeting itself is rarely about the formula. The question is whether something changed that leadership is already watching, and whether L&D can show it plausibly. If you collected the right metric on day one, you can do that in three sentences. If you only have completion rates, you need a slide deck and still will not convince anyone.
A second point is almost always overlooked. Alongside the course, the training produced something that lasts: the specialists' knowledge captured to build the content. Onboarding shows this most clearly, as the post on onboarding with video describes. When that knowledge is verified and reusable, it also answers questions it was never built for. That belongs in the calculation, even if it only shows in year two.
If you have a training project on the table today, there is exactly one task that cannot wait. Write down the one number that should have moved in twelve months, and ask the department that owns it for today's value.
If nobody knows it, you have your first result.
