InDebted

The removal test leaves no recovery operation behind, which is what separates a digital first agency from a traditional one that bought analytics. Headcount is deliberately minimal and there is no calling floor: the models decide when each consumer is contacted, through which channel, with which message, and every interaction feeds the next decision, while an inbound conversational agent handles queries.

People are the exception path, reserved for escalations and vulnerable customers, rather than the workforce the models assist. Strip the modelling and there is no one left to collect, which is the same reading this index applied to the other machine learning driven recovery agency already indexed.

A named mechanism and a named human off ramp, without the specification that would lift this higher. Compliance is described as built into the product code with a multi layered line of defence covering collections and consumer protection law in each operating market, which is a control positioned before an outreach is sent rather than a policy asserted around it.

More unusually, the escalation path is defined by who rather than only by what: a human customer experience team handles escalations and customers identified as vulnerable, and vulnerability identification is a specific supervisory expectation in at least two of the markets served. What is missing is every threshold: nothing states what triggers vulnerability identification, what contact frequency or timing caps the code enforces, or what a consumer must do to reach a person.

The models are described entirely in terms of what they achieve and never in terms of how they behave. Published material states that machine learning predicts the moment a consumer is most likely to act, selects the message, and learns from each interaction, which is an account of purpose rather than of method.

No accuracy, uplift or calibration figure is published, no back testing appears, no retraining cadence or drift monitoring is described, no artificial intelligence management system certification is held, and no validation documentation is offered. A creditor placing accounts is accountable to its own regulator for the treatment of its customers and would have nothing to show for how the recovery decisions were made.

Percentage claims without a single named reference. The published figures are outperformance of traditional collections by up to 40 percent, a large share of inbound requests resolving through conversational artificial intelligence in some markets, and improved conversion from machine written messages, and every one is company reported with the market and the client unspecified. No creditor is named, no placement volume or recovery total is published, and no customer executive is quoted.

Independent trade commentary describes the model accurately but validates none of the numbers. Under this index's bar that is self reported figures with no named customer, which is a grade below the comparable agency already indexed, and the difference is evidence rather than capability.

One safety relevant control is named and the data boundaries are untouched. Identifying vulnerable customers and routing them to people is a real protective mechanism and is credited on the autonomy axis. Beyond it, nothing describes what the conversational agent is prevented from saying to a consumer in distress or disputing a debt, and no red teaming or output screening practice appears.

On stewardship the multi client structure makes the unasked question sizeable: the firm collects for many creditors and states that every interaction informs subsequent communication, and nothing states whether a consumer's behaviour recorded while recovering one creditor's debt informs how the models treat that same consumer when a different creditor places their account.

No data processing agreement, subprocessor list, retention schedule or consumer facing privacy detail was located. The exposure is larger than for a software supplier in this pocket, because the agency receives the creditor's full delinquent account file and then builds its own behavioural record of each consumer across the recovery, holding identity, balance, contact history and inferred responsiveness on people who never chose to deal with it. Operating across four or more jurisdictions with different data regimes sharpens the question of what is held where and for how long, and none of it is addressed.

Three credentials are claimed and each is stated imprecisely. The published position is that the infrastructure is certified to an international standard, and holds a service organisation control attestation and payment card compliance, supported by a dedicated compliance page. Naming three separate frameworks and maintaining a standing compliance surface is more than most of this pocket offers and it carries the grade.

The imprecision is worth recording because this index now tracks several forms of it: the international standard is named only as a family and not as a numbered standard, the service organisation control claim does not state whether it is a point in time or a period examination, and payment card handling is described as compliance rather than as an assessed level. Each of those distinctions changes what a buyer is being told.

This is genuine supervisory standing rather than a technical conformity claim, and it is the structural advantage of the agency model. Operating as a third party collector in the United States, United Kingdom, Australia and New Zealand requires authorisation or licensing under each market's collections regime, and the firm therefore sits inside the regulatory perimeter rather than beside it as a supplier to someone who does.

It is directly supervised, directly examinable, and directly sanctionable for its own conduct, which is a materially different position from every software vendor in this pocket, all of which correctly hold no licence at all. The company's own framing is consistent, describing compliance with collections and consumer protection law across each market as built into the product rather than delegated to a client.

No fairness testing, differential outcome analysis or model governance disclosure is published, and the decision the models make is precisely the kind this axis exists for. The system decides which consumers are contacted, when, how often and with what message, and those choices determine who is pushed toward repayment hardest and who is left alone.

Vulnerability identification is named as a capability and never described, so nothing states how the models decide a consumer is vulnerable, how often that determination is wrong in either direction, or whether recovery pressure and vulnerability outcomes differ systematically across demographic or linguistic groups in the four markets served.

The agency model inverts the liability picture that applies to every software vendor in this pocket, and it does so in the consumer's favour. Because the firm is engaged as the collector rather than as a supplier, it is itself the regulated party in each jurisdiction and answers directly for its own conduct under the applicable collections and consumer protection regimes, with statutory complaint routes and supervisory action available against it rather than against a client who bought a tool.

A named human team also exists to escalate to, which is a practical route where most peers offer none. Held off the top grade because nothing is published in the vendor's own right: no error rate, no remediation commitment, no described dispute or correction process, and no statement of what happens when a model driven outreach is wrong. The recourse here comes from the regulatory structure the business sits inside, not from anything the company has committed to.

No model provider, family, version or hosting arrangement is disclosed for either the outreach optimisation models or the conversational inbound agent. Nothing states whether the models are built in house or assembled from external providers, and nothing indicates whether consumer conversations handled by the inbound agent pass through infrastructure the company does not control. For a firm operating across several jurisdictions with differing rules on cross border processing, the absence leaves an unanswerable question at the centre of its own compliance claim.

Integration depth is structurally shallow here and that is a property of the business model rather than a failing. Creditors place accounts with the agency and monitor progress through a portal; they do not license the technology or run it inside their own stack, and the company states the technology is not available for in house use.

No integration with a loan management, servicing or core banking system is named, no application interface is documented, and no integration count is published. A creditor evaluating this option is choosing to move the collections capability, the data and the consumer relationship outside its own systems entirely.

There is no deployment model to disclose, since the service is delivered as an outsourced engagement rather than as software a creditor installs or configures, and the company states the technology cannot be licensed for in house use.

That leaves residency as the live question and it is unaddressed: nothing states where consumer account data from each market is processed or stored, whether records from one jurisdiction remain within it, or how data flows between the operating entities in four or more countries. A creditor in a market with data localisation obligations would find nothing to rely on.

No pricing, fee structure or commercial basis is published. The omission is specific in an agency engagement, because recovery work is conventionally paid as a share of amounts collected and the commission rate is the single number that determines what a creditor nets, whether the rate varies by portfolio age or debt type, and whether the incentive structure aligns with treating consumers well. None of that is stated, and the published material describes how the firm incentivises its own team without ever stating how it charges the creditor.

Genuine multi market reach and no named client anywhere. Operations span the United States, United Kingdom, Australia and New Zealand with expansion into further markets, which is broader geographic coverage than the comparable agency already in this index and materially harder to achieve, because each jurisdiction carries its own collections regime that the product must encode. Creditor types are described across consumer lending and adjacent billing sectors.

Held off the top grade because nothing is evidenced by name: no creditor, no portfolio size, no placement volume and no market share figure appears, so the breadth is asserted rather than demonstrated.