Leading and lagging L&D metrics: a six-line scorecard

The deck had four numbers on it. Learning hours delivered: 4,180. Employees trained: 612. Satisfaction: 4.4 out of 5. Spend: Rs 41 lakh.

The CFO let the slide sit for a few seconds and asked which of the four would be different if the programme had been cancelled in April. Nobody answered, because the honest answer was none of them.

I have watched that same silence in three budget reviews. It is not a reporting failure. Every one of those numbers is accurate, cheap to collect and easy to defend. They are simply the wrong instruments for the question being asked, and the person who gets blamed for the silence is usually the one who inherited the template.

Your most-used metric is a lagging measure that guides nothing

The Association for Talent Development's 2024 State of the Industry report, built on data from 498 organisations, lists what learning functions actually use to measure themselves. Employee satisfaction comes first at 76 percent. Employee retention is second at 62 percent, customer satisfaction third at 53 percent. For measuring the success of the function itself, the most common metrics were the number of employees trained and the number of learning hours delivered.

Now read what Kaplan and Norton wrote about exactly this habit in their 1996 California Management Review paper on linking the scorecard to strategy:

Many of the most popular non-financial measures, such as customer satisfaction and employee attitudes, have some of the same limitations as financial measures. First, they are lagging measures, reporting how well the organization's strategy worked in the past period but providing little guidance on how to navigate to the future.

That is the mechanism, and it is worth saying plainly. Satisfaction and retention tell you how the last cycle went. They arrive after the money is spent, they move for a dozen reasons that have nothing to do with your programme, and they cannot tell you what to change next month. Hours delivered and headcount trained are worse than lagging. They are activity counts that will rise reliably whenever you run anything.

A scorecard made only of lagging measures is a rear-view mirror with a receipt attached. A scorecard made only of activity counts is a receipt with no mirror.

So the fix is not to abandon lagging metrics. It is to stop reporting them alone. Every metric on your page should have a partner that moves earlier and points the same way, and you should be able to say which one you expect to move first.

Four tests a metric must pass to earn a line

Before adding anything, apply these. A metric that fails one is a report line, not a management instrument.

A denominator you control. "412 trained" is a count. "412 of the 460 people in the funded role families" is a metric, because the denominator tells you whether you reached the population or a convenient subset. Every number on the page needs its denominator printed next to it, in the same font size.

A decision it can change. Name the decision and the month. If the answer is "it informs the talent review," you have a slide, not a metric. The test is whether a bad reading would cause someone to do something differently within one quarter.

An owner outside L&D. If only the L&D team can move the number, the number measures L&D's effort. Time to independent sign-off, rework rates and vacancy days belong to the business. That ownership is what makes the metric survive a budget cut, because someone with revenue responsibility will defend it.

A cadence shorter than the decision cycle. A metric reviewed annually cannot steer an annual budget. If the decision happens in February, the reading has to exist in January.

The fourth test is where most teams discover their problem is not measurement at all. They have data, but it arrives after the decision it was meant to inform.

The six lines, in three pairs

Three leading metrics, three lagging, each pair pointing at the same outcome. The left column is cheap and fast. The right column is what decides funding.

Three rules keep the pairs honest.

Print both numbers or neither. A leading metric alone invites the team to optimise the cheap one. Coverage can be pushed to 100 percent by shortening the assessment. Practice completion can be pushed up by making the task trivial. The lagging partner is the brake.

Test the pair after two cycles. If the leading metric moves and the lagging one does not, the leading metric is measuring the wrong thing. That is a finding about your instrument, not about the programme, and it is the most useful review you will run all year.

One owner per row, named in the sheet. Not a department. A person, with a date next to the reading.

The observation and sign-off columns need a definition before they need a tool. If you have not already built the observation form, what a level 3 evaluation actually requires covers who observes and how often, and it is the part most programmes skip.

The order the work actually happens

Do not start with the dashboard. Start with the sentence that defines competence.

Step one: write the sign-off sentence. One line, in the business's language, agreed by the person who currently approves the work. For a credit role it might read: "Signs off a file up to Rs 25 lakh unaided, with rework under 10 percent." If you cannot get that sentence, you do not have a training outcome, you have a course. This step takes one meeting and it is the whole method.

Step two: name the observer. Someone who already reviews the work, not the trainer. A trainer marking their own cohort's behaviour change is not measurement.

Step three: fix the denominators. Funded headcount per role family, not headcount invited. Enrolled, not registered. Signed off, not attended.

Step four: set the cadence and the reader. Coverage monthly, sign-off quarterly, cost per competence per cohort. One page, one reader, one date.

Step five: pre-register the expectation. Before the cohort starts, write down what you expect each leading metric to do, and what lagging movement you would accept as consistent with it. This is the step that converts a dashboard into a test. It also protects you, because a written expectation from before the programme is the only defence against a result being reinterpreted afterwards.

If the baseline does not exist yet, build it first. A programme without a baseline cannot produce any of the six lines, and starting the cohort anyway just moves the problem one quarter later.

A worked example, with illustrative numbers

Twelve branch managers, credit appraisal, one cohort. All figures below are illustrative and rounded, but the structure is the one I use.

Costs: external fee Rs 4.20 lakh. Participant time, 12 managers at 16 hours each at a loaded Rs 900 an hour, Rs 1.73 lakh. Observer and manager time, 24 hours at Rs 1,200, Rs 0.29 lakh. Total Rs 6.22 lakh.

Results: 12 enrolled, 11 completed the assessed practice file, 9 reached independent sign-off, median 7 weeks.

Cost per enrolled learner: Rs 51,800. Cost per verified competence: Rs 69,100. Same programme, two numbers, and only the second one is defensible in a budget meeting, because the first one counts the three people who cannot yet do the job.

Coverage was 12 of 14 funded branch manager roles, so the reach line reads 86 percent, not 100. Time to sign-off is compared against the previous cohort's 9 weeks, not against zero. Behaviour at week six was observed for 9 of 9 by the regional credit head, using a four-item checklist built from the sign-off sentence.

Note what the table does not contain. No satisfaction score. No hours delivered. Not because they are wrong, but because they are already implied: if 9 of 12 are signing off unaided at week 7, the sessions were not hated. If they were, that would show up as attrition in the practice step, which is a number you can act on.

The one line a CFO will read

Not the cost per learner. This:

Rs 69,100 per manager who can now sign off a Rs 25 lakh file unaided, against a rework cost of Rs 1,400 per file and 40 files a month per branch.

Two things make that line work. The numerator is a verified competence, not a seat. The denominator is the cost of the thing the competence replaces, expressed in the business's own units. Every CFO already knows what rework, a vacancy day or a failed audit costs. Almost none of them know what a learning hour is worth, and they are right not to guess.

If you cannot fill the second half of that sentence, you are not ready to make the argument, and the fix is a conversation with the operations head, not more analysis.

Where this breaks

Three limits, stated plainly.

Nine people is not a sample. One manager dropping out moves cost per competence by 11 percent. These are operating numbers for steering a programme, not estimates of programme effect. Report them as counts and medians, never as percentages with two decimal places.

There is no comparison group. Time to sign-off also moves with file volume, manager availability and how strict the regional head feels in a given month. You are building a defensible range with a stated attribution caveat, not proving causation. Any vendor who offers you causation from a single cohort is selling you something.

Self-marked behaviour is worthless. If the person who ran the training also observes the behaviour, the observation line collapses into a satisfaction score with extra steps. This is the single most common way the method gets quietly ruined, and it happens because the trainer is the only person with time.

Monday's action

Pick one programme you have already funded and write its sign-off sentence. One line, in the business's language, with the name of the person who currently approves that work beside it.

If the sentence comes easily, you have your first lagging metric and the rest of the table follows in a fortnight. If it does not, you have found the actual problem, and it is not in your reporting. The programmes most likely to be cut at the next budget review are the ones where nobody can finish that sentence, which is why the platform question comes second and the sentence comes first.