How many assessment tools an L&D team actually needs

Six instruments. One renewal spreadsheet. And a question nobody in the room could answer.

I have sat in three reviews like this in the last two years, most recently with a 240-person manufacturing company where the assessment line item had crossed into eight figures of rupees a year. The HR head opened the sheet and asked the only question that matters: which decision would have gone differently last year without this tool? The room looked at the sheet. The vendor names were familiar. The reports were familiar. Nobody could name a single hire, promotion or training batch that had changed because of any one of the six.

The renewals were signed anyway, because dropping a tool means owning the risk of the gap it might have filled. Keeping it means owning nothing.

That silence is the real finding, and it is not a procurement failure. Every instrument on that sheet is defensible on its own terms. Together they are a cost with no owner, and the only person who can find the overlap is the one who can see all six rows at once. That is you.

Every tool was bought against a label, not a decision

The stack grows because labels multiply faster than decisions do.

Count the labels on offer to an Indian HR head this year: engagement, culture, psychological safety, resilience, learning agility, potential, derailers, emotional intelligence, personality, team effectiveness, cognitive ability, values fit. Count the decisions those labels are supposed to serve: who to hire, who to promote, who to develop and in what, where to deploy, who to exit or manage out. Roughly a dozen labels against five decisions. Any label can be attached to any of the five, which is why each purchase feels justified and the total keeps climbing.

Three things keep it climbing.

Each instrument has a sponsor. The engagement survey belongs to the CHRO, the technical test to the plant head, the 360 to L&D. Nobody's job description says "notice that two of these measure the same thing." Overlap is not a blind spot. It is an unassigned task.

Procurement asks the wrong question. The evaluation form asks whether the instrument is valid, reliable, benchmarked. Every serious vendor answers yes with a technical manual. Nobody asks whether the construct is already covered by something the company pays for, because that question belongs to a person who is not in the room.

Renewal is the default and cancellation is the exception. Renewing needs a signature. Cancelling needs a case, an owner, and an answer to "what if we lose the trend line." The asymmetry does the rest.

Start with a decision ledger, not a tool inventory

Do not start by listing instruments. Start by listing decisions, and force every instrument to attach to one.

Build one row per instrument with five columns. Keep it ugly and keep it in a spreadsheet, not in a deck.

Two rules turn this into an audit.

A row with no decision in column two should not exist. If the honest answer is "it feeds the talent review deck," that is not a decision, that is a slide.

A row whose column four is empty or older than 18 months gets challenged at renewal, not defended. The tool may still be worth keeping, but now someone has to say why out loud.

You will usually find that two rows have the same entry in column two. That is the first real signal, and it costs nothing to find.

Group instruments by construct family, not by label

Labels are marketing. Families are measurement. Sort every instrument on the sheet into one of six families.

The rule that falls out of this table is blunt: two instruments from different families are almost never redundant, and two from the same family usually are. The savings live inside a family. So does the discomfort, because the two tools inside a family have different brand names, different reports and different sponsors, and the person who bought the second one is sitting in the room.

This is also where a competency framework earns or wastes its money. If your framework defines twelve competencies and each one has its own instrument, the framework is generating the sprawl rather than organising it. The test is whether the same evidence can serve two competencies without double counting, which is the argument in how to build a competency framework managers actually use.

Test the overlap on your own people, not on the brochure

Two instruments in the same family might still measure different things. You do not settle this by reading two technical manuals. You settle it by putting both scores on the same people and correlating them.

Take 30 to 40 employees who already have scores from both instruments, pull the two total scores, and compute the correlation. Any spreadsheet does this in one function. Then read it with this table.

At 0.75 the two instruments share 56% of their variance. That is not a subtle overlap. It means most of what the second tool tells you, the first tool already said, and you are paying twice for the same signal plus a small remainder.

Two cautions, because this test gets misused. Thirty people gives you a correlation with a wide confidence interval, so use it to sort instruments into "clearly overlapping" and "clearly not," not to defend a third decimal place. And correlate total scores, not subscales. Subscale correlations on small samples are noise with a decimal point.

Then test whether the second score changes anyone's decision

Overlap is only half the question. The other half is whether the second instrument moves anyone across a line that matters.

Take the last 60 to 100 people both instruments scored. Rank them on instrument A alone and put them into the bands you actually use, such as ready now, ready in a year, not ready. Then rank them on A and B combined, with whatever weights you would honestly use, and count how many people changed band.

If fewer than one in ten changes band, the second instrument is not paying for itself. It is adding precision nobody acts on, at the price of a licence, an administration window and a report that a manager has to read.

This is what the validity research predicts. Gains from a second predictor are real but sub-additive, and they shrink with overlap. In Schmidt and Hunter's meta-analysis of 85 years of selection research, general mental ability predicts job performance at .51. An integrity test predicts .41 on its own. Put the two together and the pair reaches .65. The gain is .14, not .41. Conscientiousness, at .31 alone, lifts the combination to .60, a gain of .09. A work sample, the single most valid method in the table at .54, adds .12 to ability. Every one of those second instruments is worth buying. None of them is worth what its standalone validity suggests.

That arithmetic is the strongest argument you will ever have against a seventh tool. It says the second instrument's value is not its brochure number, it is its number minus what the first instrument already captured.

A worked example, with illustrative numbers

A 240-person company, six instruments, Rs 14.6 lakh a year. The numbers below are illustrative, built to show the method rather than to describe a client.

The two ability tests were bought four years apart by two different site heads. Correlation on 46 supervisors: 0.81. One of them was producing a second copy of the same signal, and the marginal test changed two band placements out of 46, both of them people who were already borderline. Dropping it saved Rs 1.9 lakh and, more usefully, removed a meeting where two reports disagreed about the same candidate.

The two climate surveys were merged into one instrument with a fixed core and rotating modules, which kept the board trend line and cut one administration. The personality questionnaire and the 360 stayed, because at 0.44 they were answering different questions: one about how a person describes themselves, one about how their team experiences them. That is a real pair, and merging it would have been a mistake dressed up as efficiency.

Net: six instruments to four, Rs 3.1 lakh a year saved, and, for the first time, a sheet where every row names a decision.

Where consolidation is wrong

Four situations where the arithmetic says merge and the right answer is to stop.

The labels are similar and the constructs are not. Self-reported emotional intelligence and ability-based emotional intelligence sound like the same product. They are not. The authors of the current revision of the MSCEIT, the main ability-based measure, preregistered an expectation that self-estimated emotional intelligence would correlate at or below .25 with their ability test. A correlation that low means the two instruments are close to independent. If you cancel the ability test because you already run an EQ self-report, you have not consolidated two measures of one thing. You have deleted a measure of something else, and you will not find out until someone asks why the development plan for a low-empathy manager recommends more of the same training.

The same instrument is used at two different stakes. A personality score collected for development, where the person knows it will not be used against them, is not the same data as a personality score collected for promotion. The stakes change the answers. Reusing a development score in a selection decision is a consolidation that costs you the trust of everyone who answered honestly.

The evidence does not travel for free. The SIOP Principles treat transportability as a claim to be demonstrated: validity evidence gathered in one setting generalises to another when important similarities between the settings are shown. A consolidated platform's norms were built on somebody else's population. If switching instruments to save a login breaks your ability to compare this year's cohort with last year's, you have traded a defensible number for a cheaper one.

Your sample is too small to conclude anything. If 12 people have both scores, the correlation could be anything. Run the marginal test instead, or wait until the next cycle. A confident decision built on 12 data points is how stacks get rebuilt two years later.

What to do this quarter

Do not audit the whole stack. Audit what is about to cost you money.

1. Pull the renewal calendar and mark every instrument renewing in the next 90 days. Usually that is two or three.
2. Build the five-column ledger for those only. If a row has no decision, the renewal conversation is over before it starts.
3. Sort those instruments into families. Any two in the same family go to the overlap test, using 30 to 40 people who already have both scores.
4. Where the correlation is above 0.75, run the marginal test and count band changes. Bring that number to the renewal meeting instead of an opinion.

The next renewal that lands on your desk is the place to start. Do not sign it until every row has a decision and a date. If you want the whole stack in one place instead of four logins, that is a different problem, and the comparison page is where we argue it. For the step that comes after this one, turning whatever survives the audit into a plan someone can act on, the gap analysis that ends in a one-page plan covers it.