Cost Avoidance Is Not ROI

Memo · 4 min read

To: the executive who signs the training budget Re: what the number means

Most training ROI figures are cost avoidance under a better name, and the conflation is why executives discount everything a learning function reports. The fix is to report two numbers under two names, with the method visible for each, and to accept that the honest return figure will be smaller than the one you could get away with.

The two numbers

Cost avoidance answers one question: what would it have cost to buy this capability elsewhere? If a vendor charges a market rate for a technical course and the internal team builds and delivers an equivalent one for less, the difference is real money and worth reporting. It is a benchmark. It says the function is cheaper than the alternative. It says nothing about whether the training changed anything.

Return on investment answers a different question: what did the business get back, net of everything it spent? The benefit side is a measured change in a business outcome (time to competency, rework, nonconformance, cycle time) converted to money, with the share attributable to training discounted for every other cause that was moving at the same time. The cost side is fully loaded: development, delivery, learner time off the floor, platforms, and the function's overhead. ROI is net benefit over cost.

Both numbers are legitimate. The problem is a report that adds them together, or that lets the first one wear the second one's name.

Why the conflation persists

The incentives explain it. A learning function funded as an expense wants the largest number it can produce, nobody downstream is equipped to audit a training claim, and the benchmark number is easy: a vendor quote and a multiplication. The outcome number is hard. It requires deciding, at design time, what business metric the program is supposed to move, instrumenting that metric before and after, and then doing the uncomfortable part, which is admitting that training was one of several things that changed.

So functions report the easy number and call it return. It works until the first finance review that asks what the figure would be with the vendor quote removed, and the credibility of every later claim goes with it.

How I structure measurement instead

Instrumentation is tiered, because every program does not earn the same instrument, and pretending otherwise is how numbers get inflated.

Every course carries reaction and learning measures as a standard. That is table stakes and proves nothing about the business.

Selected programs carry behavior and results measures: a defined operational metric, a baseline, and a post-training read. These are the programs where the business told us what it needed to change, which is also the only honest basis for choosing them.

Full ROI is calculated only where isolation of effects can actually be performed. The methods are standard: a comparison group where one exists, trend-line analysis where the metric has a stable history, and, as a last resort, expert estimation adjusted for confidence, with the attribution share discounted before the benefit is claimed rather than after someone challenges it.

The four benefit streams behind the $50M+ figure I report sit in different tiers, and the report says so. Vendor displacement is cost avoidance and is labelled that way. Time to competency and production gains are results-level measures that can carry a return calculation. Safety-critical failure prevention is probabilistic and is reported with its assumptions visible or not at all.

A worked example

Illustrative figures, chosen to show the mechanism. Not operational data.

A technical program shortens the interval from hire to autonomous work by two weeks for 300 technicians a year. At a loaded cost of $60 an hour and 40 productive hours a week, the gross benefit is 80 hours times $60 times 300, or $1.44M. A trend line shows time to competency was already falling by about a fifth of that amount for reasons unrelated to training, which leaves $1.15M. Expert estimates, adjusted for confidence, attribute half of the remainder to the program. Attributed benefit: $576,000. Fully loaded program cost, including learner time: $320,000. ROI: 80 percent.

The same program reported as cost avoidance would show the vendor rate for an equivalent course times the number of learners, perhaps $900,000, with no cost side and no attribution. Larger, simpler, and unfounded.

What the executive gets

A smaller number and a trustworthy one. Over time that trade pays: a function whose claims survive inspection moves from being budgeted as an expense to being funded as an investment, because its numbers can be put in front of a board. A conservative number that survives scrutiny is worth more to a learning function than a large one that does not.

Two lines on the report, two names, methods visible. Everything else follows from that.