Every training ROI calculation you've ever seen in a board deck was built backwards. Someone decided the number needed to look good before anyone ran the analysis, and then the methodology was reverse-engineered to hit that number. This isn't cynicism — it's how the entire discipline of training ROI was designed to function from the start. The Phillips ROI Methodology, the Kirkpatrick extensions, the "isolate the effects of training" frameworks that every L&D certification teaches: they all share a structural flaw that nobody in the room ever wants to name, which is that the moment you attempt to isolate training's contribution to a business outcome in a system with dozens of confounding variables, you have stopped doing measurement and started doing narrative construction.
Senior L&D leaders know this. Most have known it for years. And yet the ROI slide persists, because the alternative — admitting that a clean causal number is unattainable — feels like professional surrender in a room full of finance people who think in percentages. So the number gets produced anyway, dressed in the language of rigor, and everyone in the room nods because nobody wants to be the person who says the emperor's spreadsheet has no clothes.
The entire premise of training ROI rests on isolating the effect of a single intervention from everything else happening in an employee's working life during the same period. A sales team completes a negotiation skills programme in March, and by Q3 their close rates are up 12%. Was that the training? Or was it the new incentive structure that rolled out in April, the departure of two underperforming reps in May, the competitor who raised prices in June, or the fact that Q3 is always stronger for that product line? Nobody actually knows, and the honest answer is that no isolation technique — not control groups, not trend-line extrapolation, not participant self-estimation — can cleanly separate these threads in a live enterprise environment where a hundred variables move simultaneously.
What happens instead is that L&D teams adopt a technique called "estimate and adjust," where participants or managers are asked what percentage of a performance improvement they'd attribute to the training, and then that self-reported percentage gets treated as data. It is not data. It is a guess, dressed in a number, laundered through a formula, and presented as evidence. The formula doesn't validate the guess; it just makes the guess look like it survived a rigorous process.
Consider who is calculating the ROI and why. It is almost always the L&D function itself, reporting on the value of its own programme, to a budget owner who approved that programme and would prefer not to have approved a mistake. Every party in that chain has a reason to want the number to land somewhere respectable. The training vendor wants renewal. The L&D leader wants budget for next year's programme. The executive sponsor wants validation for a decision they already made. There is no adversarial party in the room whose job is to argue the number down, the way an auditor argues down a revenue recognition claim. Without that adversarial pressure, ROI figures drift toward whatever number makes everyone's next conversation easier.
The uncomfortable truth is that a training programme can be genuinely excellent and still produce an ROI calculation that means nothing, because the calculation was never actually measuring the programme — it was measuring the story everyone needed to tell about the programme.
This is the part capability directors resist hardest, because it sounds like an argument for abandoning measurement altogether, and it isn't. It's an argument that the wrong kind of measurement is worse than no measurement, because it creates false confidence. A programme that quietly underperforms but produces a 340% ROI slide will be renewed, expanded, and copied into other business units, while a programme that's actually working but resists tidy quantification gets quietly cut because nobody built it a story finance could love.
Walk into most enterprise procurement conversations for training and you'll find the ROI number functions less as evidence and more as a ticket to the next stage of conversation — a credential that says "this vendor understands the game," rather than a genuine input to the decision. Buyers who've been burned before have started asking different questions instead: what does competence actually look like six months after the programme ends, what specific behaviours changed, what did managers observe directly rather than infer from a dashboard. This is a harder conversation to have, and it produces less impressive slides, but it produces conversations that are actually true.
This shift matters most in fast-moving technical domains, where the gap between "completed the course" and "can apply the skill under real conditions" is enormous and where a bad ROI story can either kill a genuinely necessary investment or, worse, keep funding a bad one. Take a Generative AI training programme as an example: the temptation to produce a headline ROI figure within a quarter is intense, because everyone wants proof the investment was justified. But meaningful change in how teams actually use these tools — how they prompt, how they verify output, how they redesign workflows — takes longer to show up than any quarterly ROI slide allows, and forcing the number early just guarantees a fabricated one.
The alternative isn't giving up on evaluation; it's giving up on the pretence that a single causal percentage can honestly represent something as complex as human capability change inside a business. What replaces it is less elegant and more useful: direct observation of behaviour change by people who actually manage the work, tracked over a longer horizon than any single fiscal quarter, triangulated against multiple imperfect signals rather than compressed into one falsely precise number. It's messier to present. It doesn't fit neatly into a single slide. But it has the singular advantage of being true, which the traditional ROI figure, however polished, generally is not.
None of this means training investment decisions should be made on faith. It means the rigor needs to move upstream, into programme design and delivery quality, rather than being manufactured retroactively in a spreadsheet built to justify a decision that's already been made. Organisations serious about this shift are already rethinking how they select and sequence programmes rather than how they report on them afterward, and that recalibration is visible in how thoughtfully some teams are now approaching their upcoming training batches — treating the planning stage, not the ROI slide, as the place where real accountability lives.
The number on the slide was never the point. The behaviour in the building was always the point, and it's time L&D stopped pretending otherwise.
We run live, instructor-led programmes on Orbit. Browse our upcoming batches.