The threats that don't sound like threats — three patterns a QA scorecard isn't built to catch

A QA scorecard knows what a threat sounds like. Raised voice. Profanity. The explicit line — “we’ll send people to your house.” Those are the easy catches, and most floors catch them, because they are loud and they are obviously wrong.

The coercion that actually lands on an Indian collections call is quieter than that. It is polite. It stays inside the script’s word count. It would pass a sentiment model trained on English profanity without raising a single flag. And it is, very often, exactly the conduct the RBI Fair Practices Code and the recovery-agent Code of Conduct already prohibit — just not in the vocabulary a scorecard screens for.

Here are three patterns that recur on collections floors, and why the scorecard misses every one of them.

1. The family-implication threat

The agent never threatens the borrower. The agent asks about the borrower’s family.

“Does your husband know about this account?” “We may need to confirm a few things with your office.” “Your neighbour gave us this number — should we call them back?”

Not one of those sentences contains a threat. Every one of them carries the same payload: your default can be made visible to the people whose opinion you can’t afford to lose. That is social shame deployed as leverage, and the recovery-agent Code is explicit that intruding on the privacy of the borrower’s family, referees and friends — or contacting third parties to embarrass the borrower — is out of bounds. A scorecard built to flag abuse reads these as ordinary information-gathering. They are the opposite.

2. The asset-loss implication

“You understand you could lose the house over this.” “This will affect your job.” “Your CIBIL is finished.”

These work because they are frightening and because the borrower rarely knows enough to test them. They are also, in most cases, not true — an unsecured personal loan does not put the borrower’s house on the table, and a single delinquency is not the end of a credit file. Stating or implying a legal or financial consequence that isn’t real is misrepresentation, and the Code bars misleading the borrower about the consequences of non-payment. The scorecard hears a factual-sounding sentence and moves on. The compliance problem is precisely that it sounds factual.

3. Authority-impersonation drift

This one rarely starts as impersonation. It drifts there.

“This is now a legal matter.” “The lawyers are looking at it.” “There’s a case being filed.” “We’re required to inform the authorities.” No badge is claimed and no specific court is named — which is exactly why it slips the filter. But the cumulative impression the borrower is left with is that the police, a court, or a regulator is already in motion. When none of that is imminent, the call has crossed from collection into a false threat of legal action — and implying that an arm of the state is involved when it isn’t is not a tone problem. Impersonating a public servant is a criminal offence on its own.

Why this is the lender’s problem, not just the agency’s

Under the Fair Practices Code, the lender remains responsible for the conduct of the recovery agents acting on its behalf. “The agency said it, not us” is not a defence — it is, if anything, an admission that the calls weren’t being heard. Misrepresentation carries civil exposure; a credible threat or an impersonation of authority can carry criminal exposure. And the one record that settles whether any of this happened is the call recording the new recovery-agent framework (in force, as proposed, from 1 July) now expects lenders to keep. The conversation gets logged either way. The only question is whether anyone on the compliance side listened before the borrower’s lawyer did.

Three lines to add to your next 100 calls

None of this needs a new scorecard. It needs three flags added to the one you already run:

1. Third-party references. Any mention of the borrower’s family, employer, neighbour or referee in a pressure context — not “who else can we reach for a callback,” but “does he know.” Flag it and read the surrounding ninety seconds.
2. Consequence claims. Any sentence asserting a specific loss — house, job, salary, “CIBIL finished,” arrest. Check it against what the product can actually do. If the claim outruns the contract, it’s misrepresentation, regardless of tone.
3. The word “legal.” Track every call where “legal,” “case,” “court,” “lawyer,” or “authorities” appears. Most will be benign. The few that aren’t are the most expensive calls on your floor, and they are trivially findable once you decide to look.

A scorecard that screens for shouting finds the agents who lose their temper. It never finds the calm, fluent, scripted agent who frightens a borrower into paying with a sentence that never raises its voice. That agent is the compliance risk — and right now, on most floors, the only place that call exists is a recording nobody has opened.