Part 5 of the six-part series From Course Factory to Capability Function.
Previously in this series: Who Should Do the Work: Humans, AI, Agents, or Machines
Installment 2 introduced the economic discipline: production effort is not consumption time, cash spent is not value realized, and most redesign claims blur them together. This installment is about what you measure instead, because a redesign that keeps the old metrics has changed its tools but not its accountability.
Start with an honest look at your current dashboard. Completion rates. Courses launched. Seats filled. Satisfaction scores. These numbers describe how busy L&D was. They say nothing about whether anything changed in the business. A team can hit every one of these targets while the performance problems that prompted the work remain exactly where they were.
Here is why this persists: activity metrics are easy to collect and safe to report. No one gets fired for 94% completion. Performance evidence is harder. It requires agreeing with the business on what would count as a result, then living with the answer. Most L&D teams were never asked to do this, so they never built the muscle. The redesign demands it.
Before any significant piece of work begins, write a one-page measurement plan. Four lines, no more.
Line one: the decision. What business decision does this investment serve? “Reduce onboarding time for field technicians from 90 days to 60” is a decision. “Deliver onboarding training” is an activity. If you cannot name the decision, you are not ready to build. Go back to the diagnosis from installment 3 and get clear first.
Line two: the evidence. What observable change would show the decision worked? Time to first independent job completion. Error rates in the first 90 days. Customer complaints tied to new-hire work. Pick the measure the business already trusts, not a learning metric you invent for the occasion. If operations tracks it, use their number. Borrowing credibility beats manufacturing it.
Line three: the source. Where does the evidence come from, and who owns it? If the data lives with operations or finance, agree upfront how you will get it and how often. Put a name to it. A measure you cannot access on a schedule is a wish, not a plan. This conversation happens before the work starts, not after, because after is too late to negotiate data access.
Line four: the review date. When will you look at the evidence and decide whether to continue, change, or stop? Pick a real date and put it on someone’s calendar. Without a date, measurement becomes a report nobody reads and a conversation nobody has.
Then report with the four-way separation intact. Installment 2 drew the economic lines; here is how they show up in practice. When you claim a result, say which of the four moved.
“We released 200 designer hours by automating first drafts.” That is capacity release. It is real, but it is not savings until you say what those hours moved to. If they moved to diagnostic work with business units, say that. If they evaporated into slack, say that too.
“We gave 10,000 employees an hour back by cutting a redundant module.” That is consumption time returned. It matters, but an hour returned to a calendar is not an hour of value created. Do not multiply it by an average salary and call it savings. That math insults the finance people you need on your side.
“We cancelled two vendor contracts worth $180,000.” That is cash. It is the cleanest of the four, and the rarest. Claim it plainly when you have it.
“Field technicians reach independent work 11 days sooner, and first-90-day error rates dropped 18%.” That is realized value. It is the only one of the four that describes a change in the business. Everything else is a means to this.
Each is honest on its own. Blending them into “we saved $2M” is not. The moment you collapse the four into a single dollar figure, you lose the trust of everyone who knows the difference. Keep them separate and your credibility compounds.
One more discipline: measure the counterfactual where you can. Compare the trained group with a group that was not trained. Compare performance before and after, holding other factors as steady as you can. You will not always get a clean comparison. Business conditions shift, rollouts overlap, sample sizes disappoint. Say what you did, say what you could not rule out, and report the range instead of a point estimate. An honest estimate with stated limits beats a precise fiction every time. The leaders you serve can tell the difference, and they remember who respected their intelligence.
The teams that do this stop defending L&D’s budget and start informing the business’s decisions. That is the point of the redesign: not a cheaper factory, but a function whose value can be seen and tested.
