Who scores the agencies?

Ask a lender how it decides which accounts go to which agency, and the answer is usually some combination of historical relationship, stated capacity, and a spreadsheet. Slices get cut. Each agency or vendor receives its allocation and gets to work.

What happens next is often genuinely sophisticated. A vendor receiving an allocation will score it, segment it, decide channel and timing, design campaigns, and iterate on what works. Some of the best applied modelling in Indian financial services happens inside those operations.

But all of it is bounded by the allocation. A vendor optimises within the slice it was handed. It cannot compare its own performance against the agency working the desk next door on a similar cohort, because it never sees that cohort. It cannot tell the lender that a particular set of accounts would recover better somewhere else entirely. It is not being withholding — it simply has no line of sight beyond its own portion of the book.

The decision nobody is positioned to make

Which leaves the most consequential decision in the whole arrangement — who gets which accounts — sitting with the lender, made at a level of resolution far below what any individual vendor has about its own slice.

The lender knows aggregate recovery by agency. It rarely knows whether agency A outperformed agency B on comparable accounts, because the two were never given comparable accounts. It cannot see that a cohort underperforming with one partner has a profile that another partner consistently resolves. Those comparisons require a view across the whole book, held by someone with no stake in the answer.

Allocation is the highest-leverage decision in collections, and it is routinely made with the least information.

Why this layer doesn't already exist

Part of the reason is technical: comparing partners fairly needs one taxonomy, one set of outcome definitions, and outcome data flowing back from every channel in a consistent shape. That is real work and most institutions have not done it.

But the larger reason is structural. The organisations with the data, modelling capability and domain knowledge to do this well are, almost without exception, also organisations that receive allocation. And the moment a party that receives allocation starts recommending allocation, its recommendations acquire a second possible reading that nobody can rule out by looking at them.

If the recommendation says shift volume toward a particular channel, is that because the channel performed, or because the channel belongs to whoever produced the recommendation? Both hypotheses generate the same output. The lender cannot separate them by inspecting the result.

This is not an accusation about how anyone operates today. It is a constraint on who can credibly hold a role that is mostly still vacant.

Why "we would never do that" isn't sufficient

The natural response is assurance: we would never weight our own channels. In most cases that would be entirely sincere. It is also unfalsifiable, and unfalsifiable assurances carry little weight inside a regulated institution.

A bank's risk function accepts "trust us" nowhere else — not for model validation, not for access control, not for data handling, not for outsourcing. Each of those must be evidenced rather than promised. There is no principled reason allocation should be the exception, least of all when allocation sits closer to the money than any of them.

The party nobody asks

There is a second constituency here, and it is routinely overlooked: the agency being measured.

Cross-agency allocation only works if outcome data flows back from every partner in a consistent, timely, structured form. Which means asking each agency to submit its contact attempts, dispositions, outcomes and costs into a shared system.

If that system is operated by a party the agency competes with for allocation, the agency has an obvious incentive to submit the minimum, as late as it can, in the least usable format it can defend. That is not obstruction; it is a rational response to being asked to arm a competitor.

So the conflict does not only risk distorted allocation. It starves the layer of the data it needs to function at all. A scorecard the agencies will not feed honestly measures nothing — and the institution ends up allocating on partial information while believing it is allocating on evidence.

Structural, not aspirational

Neutrality asserted in a sales meeting is a statement of intent. Intent moves with quarterly targets, with a new head of revenue, with an acquisition. What an institution needs is neutrality it does not have to take on faith — a constraint the vendor cannot quietly exit.

That means committing to things that are costly and checkable:

• Never operating human collections. No tele-calling floor, no field force, no collection agents — for any client, at any scale, permanently. A party that cannot receive allocation has nothing to bias allocation toward.
• Scoring every channel on one methodology, including our own, in the same tables, through the same code path, with the same outcome definitions — not a parallel process that happens to produce similar numbers.
• Leaving the objective function with the institution. The platform reports performance and recommends against the priorities the institution states. It does not move volume on its own initiative.

Each of those is checkable. An institution can inspect whether a separate code path exists for our own channels. It can ask whether volume has ever moved without its approval. It can require, in a contract, that we have not reserved the right to enter human collections later.

What the commitment costs

It would be dishonest to present this as costless positioning. Human collections is the largest revenue line in this industry by a wide margin. Committing never to operate it means permanently forgoing the biggest single business available in the category we work in.

That is what makes it a commitment. A promise that costs nothing to keep tells an institution nothing about what a vendor will do under pressure. The question worth putting to any collections technology partner is not whether they intend to be fair. It is what their neutrality would cost them to abandon — and whether they have written that answer down somewhere an institution can hold them to it.

I spent eighteen years in this industry, including at companies whose primary business was collections, before building ShieldX. Nothing here is aimed at them. The allocation layer simply wasn't a job anyone was positioned to take — and working inside those operations is where it became obvious both that the gap was real and what would have to be true of whoever filled it.