Merging four site TNAs without producing a course list
Four sites send four training lists and the plan becomes their union. Merge on the task, build a divergence matrix, and defend every exclusion.
Rajendra Kumar · 17 November 2026 · 9 min read

Representational image.
Four plants send their training needs analysis in one fortnight. Coimbatore asks for 14 courses, Pune 11, Hosur 16, and the new Rajpura line 9, four of them for machines still in crates. Central L&D has one budget line and three weeks, so it does the thing that requires no decision from anybody: it takes the union. Thirty-eight courses, one plan document. By June, eleven are marked deferred and nobody can say which of the other twenty-seven mattered.
The failure is not the duplication. It is that a union has no room for the question the four lists were supposed to answer: which gaps are real, and which of those does training close?
Why the union always wins
A course name is a container. It belongs to a supplier, it has a duration and a price, and the site team with the plant head's signature stands behind it. Dropping it means telling a plant head his team's judgement is unsound, in writing, with no shared evidence. Adding it costs an email. Every consolidation I have watched optimised for the email.
There is a second reason, and it is the expensive one. Courses are not comparable objects. "Advanced Hydraulics, 3 days" at one site and "Hydraulic System Maintenance" at another can be the same task under two names, or two different tasks under one name, and the list cannot tell you which. It has thrown away the information you need to merge it.
ISO 10015, the quality guideline for training, states the sequence in clause 4.2.1: training starts after a needs analysis, and the purpose is to define the gap between existing and required competence, then define what training is needed by the people whose competence falls short. Clause 4.2.6 adds what a course list erases: gaps may be closed by training, or by process redesign, recruitment, outsourcing, job rotation or a changed work procedure. A list of 38 courses is 38 decisions already made. That is what you are merging.
The unit is the task, not the course
Merge on the task statement and the duplicates fall out on their own. A task statement has a verb, an object and a standard: purge the extruder barrel before a grade change so that no resin remains in the die. Everything else on the row is evidence or delivery, and both can wait.
The US Office of Personnel Management's job analysis guidance gives the formula: perform what, to whom or what, to produce what or why. Their worked example is "sort incoming mail into groups for distribution". The traps matter more: double-barrelled items hiding two tasks in one line, items so specific they name the software rather than the work, and vague qualifiers such as thorough knowledge or adequate, which describe nothing observable. Merging four sites' rows, you will find all three already in the file.
Once the rows are task statements, sort them into three buckets and agree the bucket before you look at a single row.
| Bucket | Test | What happens |
|---|---|---|
| Exact | Same verb, same object, same standard | One row, one delivery, one owner |
| Family | Same verb and object, different local standard, equipment or scale | One row plus a site annex |
| Distinct | Different verb or different object | Separate rows, no merge attempt |
The family bucket is where the money is, and it is also where the merge must stop. Two plants both "respond to a filter press leak", but one is running a 24-inch plate and frame press and the other a 12-inch recessed plate press. That is one task with two annexes, not two courses and not one course either.
The merge, in the order the work happens
1. One row per request, with six columns
Site, role, task statement, current standard, evidence, requested course. The first five carry the argument, the sixth is reference only. A row with no evidence is a request, not a need: move it to a separate sheet rather than deleting it, because deleting it invites it back next cycle.
2. Normalise the vocabulary before merging anything
The step teams skip because it looks like admin. Take two people, two sites at a time, and rewrite every row to one house style: one verb, one object, one standard, no supplier names, no local job titles. A task that cannot be rewritten as one sentence is two tasks. Budget a day per pair of sites, and expect the first hour to be an argument about whether "line operator" in Pune means the same as in Hosur. That argument is the work.
3. Collapse with the three buckets, out loud
Read the collapsed list back to the four site leads in one sitting. Not for approval, for correction. A site lead accepts a merge he watched happen and rejects the same merge by email, because the second looks like head office deleting his plant's judgement.
4. Rank what is left, using ratings and not trust
OPM's task rating approach is the cheapest instrument available: ask incumbents and supervisors to rate each task on importance and frequency, 0 to 5, then apply cut-offs to the averages. Their recommended cut-off on both scales is 3.0 or above, after removing tasks rated "not performed" by at least half the subject matter experts. The Department of Energy's training handbook adds difficulty to those two and turns the three into a train, no-train or over-train decision per task.
Use the ratings to rank, not to prove. A 4.5 average importance means four people believe the task matters. It does not mean it fails. What it buys you is a defensible order of work that did not come from the loudest plant.
5. Build the divergence matrix
This is what the exercise exists to produce. For every task at two or more sites, record what fails at each site and what the site thinks causes it.
| Task | Site A reported failure | Site B reported failure | Same intervention? |
|---|---|---|---|
| Change over on Line 3 within cycle time | Setup takes 95 minutes against a 60 minute standard | Setup is fast but first 40 bottles are out of spec | No |
| Close a customer complaint in the QMS | Closure averages 11 days against a 5 day target | Closure is fast, root cause section left blank | No |
| Isolate and lock out a pump before maintenance | No reported failure | Two near misses in the last quarter | Investigate |
Read across the third row and the merge has already told you something a course list never could: the same task is fine at three plants and failing at one, which points at that plant's permit practice, not at a training gap shared across the group.
6. Apply the tie-break before funding anything site-specific
When two sites want different things for the same task, decide in this order and record which test you used.
- External obligation. A statutory requirement, a customer audit finding or an OEM condition precedent. Fund it, no debate, and mark it as compliance rather than development in the plan.
- Spread and criticality. The task appears at three or more sites with importance at or above 3.0. Fund it centrally and deliver it once.
- Cost per head of closing the gap. If one site's version costs three times per head what the others do for the same task, that site is solving a different problem. Go back to the matrix rather than funding the expensive version.
- Is it a knowledge gap at all. If the site's evidence is a repeating defect, check the failure record before you design anything. ISO 10015 lists the alternatives, and most of the expensive rows on a site TNA are one of them.
What survives these four tests gets funded. The rest goes into a dated exclusion table with a one-line reason, and that table is the most useful page you will produce all year.
7. Publish the exclusions
A merged plan without an exclusion table returns next February as the same union, because nobody remembers why a row was dropped. Give each exclusion a reason code: duplicate of a funded row, no evidence, non-training cause, site-specific and deferred to the plant, or below criticality. Nine lines. Send it to the four plant heads with the plan.
A worked example
The numbers below are illustrative, built to show the shape of the arithmetic rather than any real plant.
Four sites submit 112 rows between them. Step 2 rewrites all 112 as task statements and finds that 19 of them were never tasks: eight were supplier course titles with no task attached, six were compliance obligations already covered by an existing module, five were requests for a person rather than a skill. That leaves 93 rows covering 41 distinct tasks.
Of the 41, twelve appear at two or more sites. Nine of those twelve report the same failure with the same cause, so they become one delivery each with a site annex. Three report different causes and go down the tie-break list. One is an OEM condition on the new Rajpura line and is funded as compliance regardless of criticality.
The 29 single-site tasks are not automatically cut. Six are statutory. The remaining 23 are ranked by importance and frequency, and the plan funds 4 of them, on the grounds that a task present at one site with high importance and high frequency is still a real gap. Total funded: 9 shared deliveries, 4 site-specific, 1 compliance item, 14 rows. The other 79 rows are on the exclusion sheet with codes.
That is 112 rows to 14, and every one of the 14 defends itself with a task statement, a site and a rating. The other 79 defend themselves too, which is what stops the argument.
The limits, honestly
This only works if two sites can agree a verb and an object inside a day. Sometimes they cannot, because the work genuinely differs and the job titles hide it. Then the disagreement is the finding: the group has two jobs wearing one title, and that belongs in the report before any training plan does.
It also costs the plant head something. He loses his list and gains an annex. Take the list without giving him something to own and he will rebuild it quietly in his own budget, and your merged plan will describe a programme nobody runs.
And the ratings are perception. Supervisors rate what they remember, which skews towards recent incidents and towards tasks they personally supervise. Pair the ratings with a failure record or with a work sample before you let them decide a budget. The post on checking a training need against a failure record is the version of that check I would run on the twelve shared tasks first.
One more, plainly: a sampled task inventory never covers every task, and merging four samples gives you four partial views. Publish which roles each site sampled. The method is in the two-stage task inventory, and its coverage rule applies per site.
What to do on Monday
Pick the two sites with the largest headcount. Book one room, one day, four people: the site training coordinator from each plant, one supervisor who does the work, and whoever owns the training budget. Bring the raw rows and nothing else. Do step 1 and step 2 in the room, and leave with a collapsed task list that two sites have already argued over.
Take that list, not the course list, to the next budget review. The gap between the two numbers is the slide. If it is 38 against 9, the room will ask what happened to the other 29, and you will have an answer for each. That is the exercise, and it is why the capability data behind the plan has to come from assessment: a task list with no evidence behind it is just a shorter wish list.
Next step
Run the same measurement on your own team
Bring a real team and a real competency. We set up the assessment, run one cycle, and walk you through the report so you can judge the output yourself.
Written by
Rajendra Kumar
Head of Enterprise Business, Conceptial
Runs enterprise work at Conceptial, the training and consulting firm behind BolderBrain. Two decades of building assessment and development programmes for Indian enterprises.