Training needs analysis vs identification: where the budget leaks

In February a plant head sends L&D a list. Fourteen courses, ranked by enthusiasm, with a note that says "please schedule before June". By April it is a calendar. By November nobody can say what the fourteen courses changed, and the same fourteen arrive again next February, slightly reordered.

Everyone involved believes they did training needs analysis. What they did was training needs identification, and then they spent the analysis budget anyway.

Two processes hide behind one acronym

The terms are used interchangeably in most job descriptions and most vendor decks. They answer different questions and produce different decisions.

Training needs identification is intake. It establishes that a gap exists and who has it. The JICA training needs assessment manual, used for civil service capability work, puts this as step one of five: identify the problem and needs, then design the assessment, collect data, analyse, and feed back. Identification is the door. It is not the room.

Training needs analysis is the examination of that gap: whether training can close it at all, what kind, for whom, in what order, and whether it is worth the money. CIPD draws the line sharply in its learning needs factsheet, describing a training needs analysis as a one-off event aimed at a specific training activity, distinct from an ongoing capability review, and warning that whatever gap is agreed, the organisation should consider all appropriate ways to meet it "rather than defaulting to a course".

Most Indian L&D teams are strong at the left column and absent from the right one. The only decision actually taken is scheduling.

Why the list survives contact with a budget

A request for "advanced negotiation skills" is safe. It is a noun, it exists in a catalogue, three vendors will quote for it, and if it goes wrong nobody can say the manager asked for the wrong thing. A statement of need is riskier, because it requires naming a behaviour, a standard, and a number. So the organisation defaults to the safe artefact.

Three mechanics keep it alive:

1. The request transfers risk upward. A manager who asks for a course has done their duty. A manager who says "my team loses renewals because pricing approval takes nine days" has exposed a process they may own. The filter has to make the second statement cheaper to say than the first.
2. Budgets are annual and categories are pre-approved. Finance has a training line and a process line. A gap that turns out to be a process problem cannot be funded from the training line, so it gets described as training to fit the envelope. This is not dishonesty. It is accounting.
3. Nobody is accountable for the gap after the course. The vendor is accountable for delivery, L&D for logistics, the manager for attendance. The behaviour on the shop floor has no owner in the paperwork.

The four-question filter below exists to make the right column cheap enough to actually run. It takes two to three hours per request, not a consulting engagement.

The four-question filter

Run every request through these in order. Stop at the first failure and send the request to a disposition, not to a bin.

Question 1: What did you see, and where? Not a category. Two named instances with a date and a place. "On the 14 March renewal call with the Nagpur account, the manager conceded a 12 percent discount without asking for anything in return." If the answer is "they generally lack ownership", you are being given a feeling. Ask for the last time it happened.

Question 2: What should have happened instead, and where is it written? There must be a standard, a checklist, a script, a policy, or a named person who does it well. If no standard exists, the gap is a clarity problem, and training will teach people to be confident about an undefined behaviour. Write the standard first; you will often find half the gap closes by itself.

Question 3: Can they do it when asked, with the file in front of them? This is the non-training filter, and it is the one that saves the most money. Mager and Pipe's performance analysis, summarised in the ERIC record of their 1970 workbook, separates a discrepancy caused by a skill deficiency from one caused by absent consequences or poor feedback, and prescribes a different remedy for each: formal training, practice, reinforcement, or simply adequate feedback. In practice you test it with one work sample. Sit two or three people from the group down with a real case and ask them to do the task. If they can, the gap is environmental: no time, no authority, no data, no consequence. Run the sample as a short assessment rather than a conversation, because people describe themselves generously.

Question 4: Which number moves, by how much, by when? Name a metric that already exists in a system of record, its current value, and a target date. If the request cannot name one, it is a want. If it names one that nobody can pull, the measurement is the project, not the course, and you should read the order in which impact becomes provable before spending on delivery.

What the answers produce

The want ledger matters. It is a single sheet with four columns: request, requester, reason deferred, review date. Wants are not rejected, they are deferred with a name attached. A request that survives two reviews and still has a manager willing to defend it usually turns out to be a real need, and now has a sponsor.

One request, end to end

Numbers here are illustrative, the shape is not.

A retail bank's L&D manager receives a request: advanced negotiation skills for 40 relationship managers. A vendor quotes 4.2 lakh for four days of classroom delivery, plus 40 participant days at an internal loaded cost of about 4,500 per day, which is another 1.8 lakh. Total commitment, 6 lakh, before anyone has asked a question.

• Q1. The regional head names three accounts lost on pricing in the last quarter and one relationship manager who consistently holds margin. Concrete enough to continue.
• Q2. There is a credit policy and a discount matrix. There is no standard for how a renewal conversation should run. So half the gap is undefined behaviour.
• Q3. L&D takes the top eight relationship managers and two who are struggling, and gives each the same live account file with 90 minutes to prepare a renewal plan. Six of the eight strong performers produce the plan the regional head wanted. The two strugglers do not, and neither do four of the mid-group who were never tested.
• Q4. The metric exists: average discount on renewals, pulled monthly from the CRM. Current 8.4 percent, target 6.9 percent by the next quarter close.

The disposition is not the four-day course. It is a written renewal standard (two pages), a one-hour briefing for all 40, a targeted two-day workshop for the eight who failed the work sample, and a change to the approval workflow so nobody can discount more than 5 percent without a documented trade. Cost: roughly 1.1 lakh. The 4.9 lakh that stays in the budget is not a saving to celebrate. It is the amount that was about to be spent on a problem nobody had described.

What this does to the plan

When the filter runs first, the plan stops being a calendar and becomes a set of funded decisions, each with a baseline and an owner. Three things change immediately:

• Participant days become visible. A two-day course for 40 people consumes 80 working days. Add manager time for nomination, briefing and follow-up and the real cost is frequently double the invoice. If a request cannot survive Q4, it cannot justify 80 days either.
• Vendors get a specification, not a brief. You arrive with a behaviour, a sample of current performance, and a target. Quotes get comparable because the scope is.
• The next cycle starts with data. The Q3 work sample is a baseline. Re-run it in 90 days and you have an effect size instead of an attendance figure, which is what makes a pricing conversation about assessment and reporting concrete.

One discipline keeps it honest: the decision log. One line per request, with the four answers, the disposition, and the date. Six months later it is the only document that can settle an argument about why a programme was or was not funded, and it protects the person who said no.

Where the method fails

Three limits worth stating plainly, because a filter applied everywhere becomes bureaucracy.

Compliance and statutory training is exempt. If a regulation requires it, the four questions are answered by the regulator. Do not run the filter, run the calendar, and keep the certificate.

New capability has no historical performance to sample. If the organisation is adopting a new system or entering a new category, there is no observed gap, because the work does not exist yet. There the method is task analysis, not gap analysis: break the future job into tasks, define what good looks like, and build the training from that. Expect to get it partly wrong.

It needs a manager who can name a standard. Some managers cannot, and no amount of questioning will produce one. In that case the honest move is to write the standard with them, which is real work and often more valuable than the course they asked for. Budget for it.

And the political limit: when the request comes from the managing director, the filter is not a decision tool, it is a design tool. Run Q1 to Q4 anyway, privately, then use the answers to shape the programme so the money buys something measurable. You will not win the argument about whether to run it. You can win the argument about what it teaches.

Take three line items this week

Open your current training plan. Take the three largest line items. For each, write down the answer to Q1 and Q2 in a single sentence, and book a 45-minute work sample with three people from the target group.

Anything that cannot produce a dated instance and a standard moves to the want ledger with a review date 60 days out. The money that frees up funds the one request where Q3 showed a genuine skill deficiency and Q4 named a number.

That is the whole method. Not a new framework, just refusing to let intake do the work of analysis.