Adapting a borrowed competency framework without its arguments

The group HR head in Gurugram had a 42-page document and a deadline. The parent company in Stuttgart had issued a competency framework and asked every operating company to adopt it by the next appraisal cycle. She read it on a Sunday. By page eleven she had found the problem.

One competency was "manages supplier cost pressure through specification discipline", sitting at level three with four behavioural anchors. The anchors described a procurement engineer negotiating down a bill of materials on a product designed in-house. Her company did not design anything. It assembled to a customer's drawing, sourced from approved vendors, and won contracts on delivery reliability, not engineering cost.

She could keep the competency and rewrite the anchors. That is the safe move. It would also have kept a level her business had no evidence for, and no way to rate.

The framework carries the arguments that made it

A competency framework is not a list of words. Child and Shaw, writing for Cambridge University Press and Assessment, define one as three things together: a definitional system, a rationale for why the definitions are organised that way rather than another, and a proposed use. They add the part most people skip. Competence frameworks reflect the values of the people who built them.

That has a practical consequence. When a framework crosses an organisation boundary, the definitions travel. The rationale travels with them whether you notice it or not, and the anchors are where it hides, because each anchor describes a good outcome in the originator's business.

So the Stuttgart framework was not neutral shorthand for "good at procurement". It encoded a strategy: engineer the cost down at design stage, own the specification, compete on the product. Put that anchor in front of an assembly business and you have not adopted a vocabulary. You have imported a competitive posture your managers cannot execute.

Four tests before anything gets transplanted

The instinct is to review the framework competency by competency, which is slow and produces arguments about wording. A filter is faster. Run every competency and every level through four tests and let the failures fall out with a recorded reason.

Each is answerable in a meeting with the document open, in about twenty minutes per competency. Outcome and ladder are quick. Context and evidence are where the work sits, and they are the two that get skipped.

Test one: the outcome test

Take the anchor and ask what result it names. If your business produces that result, the competency probably survives with an edited sentence. If the result belongs to a different operating model, the competency is not yours.

Procurement cost pressure failed. So did "optimises plant utilisation across a multi-site network", because there was one site. But "holds a delivery commitment when the schedule slips" passed at every level, because late delivery was the thing that lost contracts. Nothing in that anchor referred to design authority or product ownership. It described a commitment and the conditions under which it holds.

Write the result in a column next to the anchor. Not "understands" or "demonstrates awareness": those are not results, and an anchor built on them fails the evidence test later.

Test two: the ladder test

Level descriptors are written against an authority structure. The UK Civil Service is unusually explicit. Its Success Profiles behaviours are defined at grade; the published guidance says they are specific to the grade level of the role, and that recruiting managers choose a selection rather than assessing all nine. The examples run from Administrative Assistant to Director General, and the same behaviour reads differently at each rung.

The Director General version of "working together" reads "confront issues and challenge assumptions at the highest levels". The AA version reads "get to know your colleagues and build supportive relationships".

Same label, different job. That is the design, not a flaw. A level descriptor is portable only if your grade structure puts comparable authority at that rung.

Test it directly. For each level, name the person who holds that authority. If you cannot name anyone, the level is decorative. In the Gurugram case, level four assumed a manager with budget authority over a supplier relationship. Two people had that authority and neither sat at the grade the framework expected. Level four became an escalation behaviour with a named approval threshold, and it started being rateable.

Test three: the context test

This has the strongest evidence base, and it argues against treating competencies as universal.

The Cedefop typology study by Winterton, Delamare-Le Deist and Stringfellow traces how differently the same word behaves across countries: the UK built a predominantly functional model around demonstrable workplace performance, France a multidimensional one, Germany the unity of the craft. The study is blunt that competence is ambiguous and that each occupational family has its own interpretation.

It goes further, citing Fischer: skill level is a characteristic not only of a person but also of a context, because people do not have competences independent of context. Stasz's work on the SCANS competences makes a related point. What employers named as essential was common rather than generic, meaning it had to be embedded in a setting to be understood at all.

That is the test. An anchor passes only if a rater in your setting could observe it this month. "Influences senior stakeholders" fails, because nobody can point at it. "Gets a quality hold released by the plant head within one shift" passes, because it happened or it did not.

US employment regulation makes the same demand in harder language. Under 29 CFR 1607.5, a content validity study has to show that the content of the procedure is representative of important aspects of performance on the job for which candidates are evaluated. If the framework is used for selection or promotion in a way that produces adverse impact, that documentation obligation attaches to you, not to the organisation that wrote the framework. A borrowed anchor with no local performance link is not a smaller version of evidence. It is a gap.

Test four: what evidence do you have for a level you did not write?

Now the awkward part. If you keep a level, you are asserting that people at that level perform differently in a way that matters. The originator may have had data for that. You do not inherit the data.

The Cambridge paper sets out why this is a real question rather than paperwork. It groups the claims a framework makes into construct claims (the definitions and their organisation), audience claims (who it is for), utility claims (predictive, measurement, contextual, pedagogical) and impact claims. Most frameworks in their review carried construct or audience claims. Utility and impact claims were rarer, and the authors suggest the reason is the difficulty of collecting evidence for them.

Which means the document you were handed may never have proven that its levels discriminate. It may only assert it.

So ask for three things before keeping a level, and accept "we don't have that" as an answer you can work with:

1. The sample the levels were calibrated on: how many people, which roles, rated by whom.
2. The distribution. If every incumbent scored at two and three, the level four descriptor was never tested against a real population.
3. Any link between level and an outcome you also measure: rework rate, on-time delivery, first-pass yield, first-year attrition.

If none of the three exists, keep the level only as a development ladder with no decision attached, and put a review date on it. The regulation is explicit that validity is not permanent. Section 1607.5 requires review of currency, and says there are no absolutes in deciding when a study is outdated, only circumstances to consider, including changes in the labour market and the job.

A worked example, with illustrative numbers

An auto-components supplier with 1,400 employees adopted a group framework of 34 competencies. The adaptation ran six weeks.

Week one, the outcome test removed 9 competencies outright, each with a one-line reason. Week two, the ladder test found 11 levels assuming authority the company did not have at that grade. Seven were rewritten against named approval thresholds; four were deleted because a different function did the work.

Weeks three and four, the context test rewrote anchors for the remaining 25 competencies using incidents from the last two quarters of the company's own quality and delivery records. This is the step that costs the time. Every anchor needed a real example a rater could recognise, and the examples came from records, not from a workshop of willing managers.

Weeks five and six, the framework went to 40 managers as a pilot on 120 employees. The pilot was not run to rate anyone. It was run to find anchors where two managers disagreed about whether the behaviour had occurred. Fourteen anchors failed that check and were rewritten again. Two competencies were merged because raters could not separate them.

At the end, 21 of 34 competencies survived in recognisable form, 7 were rewritten beyond recognition, 6 were dropped. The figure nobody puts in the slide: 38 percent of a framework issued as a group standard did not describe this business, and finding out took six weeks and no consultants.

The honest limit

This filter will not tell you whether the surviving competencies predict performance. It tells you whether they describe the job accurately enough that a rater can answer. Those are different claims, and only the second one is available to you in six weeks.

There is also a political limit. Saying "we dropped nine of the group's competencies" requires a sponsor who will take that meeting. If the framework is being adopted for a compliance reason rather than a development reason, the filter still works, but run it internally and publish the mapping instead of the deletions. You keep the standard, you show your work, and the arguments stay visible in a document rather than in a corridor conversation.

And if the group framework is genuinely good, the filter will say so. Most of it usually survives. The point is not to reject borrowed material, it is to know which sentences you are signing.

Do this before the next cycle

Pick one competency from the borrowed framework, ideally the one that made someone in the room frown. Run the four tests on it in a single sitting, with the document and last quarter's records open. Write the outcome, name the person who holds that level's authority, find one observable instance from the records, and ask what evidence exists that the level discriminates.

One competency, one hour. Then decide whether the other 33 deserve the same hour, and say out loud how many you expect to survive. Building your own instead? The same four tests apply, and the build sequence that survives a retranslation test is where the anchors should come from. If the framework is already live and drifting, the year-two maintenance routine covers the review that keeps borrowed levels honest.

Either way the anchors end up in a rating tool, and what the product does with them matters more than the document they arrived in.