Bits Technology

A C here is unusually well founded because the vendor scopes its own AI and does so honestly. Its dedicated AI page states that the company is deliberate about where AI fits and that it handles the sorting, summarising and surfacing so the compliance team can spend time on the right things rather than on data gathering. That is an explicit statement that the models assist rather than decide.

The removal test agrees: strip them and the orchestration layer, the connections to registries, sanctions lists and electronic identity providers, the no code workflow builder, the risk scoring rules, transaction monitoring and case management all remain, which is the entire platform as it existed before the AI product was added. This sits a grade below the other orchestration vendors in this pocket, where the agents perform document analysis, ownership resolution or case preparation rather than summarisation.

The path is described step by step, which is more than most: the agent summarises the raw data, the agent drafts an actionable decision, and a human agent then verifies or edits the agent's work. Naming an edit right rather than only an approval is a meaningful detail, because it gives the reviewer something to change rather than a binary to rubber stamp. Risk based routing sits alongside it, with human review described as concentrating on higher risk customers.

Held at B rather than A on modality: the human step is permissive rather than mandatory, described as something reviewers can do, where the A in this pocket goes to a vendor stating that human sign off is enforced at every point. No threshold is published for what counts as higher risk, and nothing describes what happens if a drafted decision is accepted without being read.

Every published number measures work removed rather than decisions made correctly: manual case handling down 50 to 70 percent, onboarding four to six times faster, decision time reduced from hours to seconds. None of it speaks to whether the summaries are faithful or the drafted decisions correct.

That matters more than usual given the design, because if the agent's summary is what the human reads before verifying, the quality of the summary is doing the analytical work, and no accuracy rate, evaluation method, error taxonomy, benchmark or validation material exists for it.

The strongest evidence position found in this pocket and one of the strongest in the index. Around twenty two customers named, and ten executives quoted on the record by name and title, among them the chief executives of Qliro, Walley, Coinmotion, Lesslie, Fondo and Lawster, the Sweden country manager of Alisa Bank, the head of anti money laundering at Brite Payments, and chief operating officers at CleverCards and 0to9.

Outcomes are attached to named clients rather than floated separately: Qliro, Alisa Bank and Walley are cited as having cut manual case work by up to 70 percent and achieved onboarding up to six times faster, alongside a stated 3.8 times return over five years and onboarding measured in minutes rather than days. That is the A bar, a named customer carrying a number, met repeatedly rather than once.

Nothing states whether customer case data, uploaded documents, analyst edits or verification outcomes are retained, reused or used to train or improve the agent or the co pilot, and the question is live because the co pilot is described as analysing millions of lines of data and because analyst corrections to drafted decisions are exactly the signal a system would learn from.

One genuine positive is recorded rather than credited as a control: the company scopes its own AI honestly, stating it is deliberate about where AI fits and confining it to sorting, summarising and surfacing rather than deciding. That restraint is the same family as the honest disclosure this index has recorded elsewhere in the pocket, and it costs a vendor nothing.

A single combined legal page covering terms, privacy and security, and a GDPR compliant marker in a marketing strip. No subprocessor list, no retention schedule, no deletion commitment and no data processing agreement terms were located.

The subprocessor gap is the pointed one here, because the platform's entire design routes customer and beneficial owner data through an orchestrated ecosystem of third party verification and data providers, so the processing chain is longer than for a single vendor product and the parties in it are not enumerated. One small transparency point recorded in its favour: the Swedish corporate registration number is published in the footer.

No certification, attestation or trust portal was located. What stands in their place is a marketing strip reading security first, verified providers, encrypted data and GDPR compliant, which asserts qualities without naming anything a buyer could check, and is weaker than silence for the reason this index has recorded before: it invites an assumption that cannot be tested.

The navigation compounds it, since the Security, Privacy and Legal links all resolve to the same single legal page, so there is no security document as such. That is a real gap for a vendor whose named customers include a supervised listed bank and a regulated crypto platform, both of which would have run third party risk assessments to buy.

A software supplier outside the regulatory perimeter, incorporated in Sweden as Finbits AB. No licence, registration, supervised programme or government trust framework certification was located. Its customers hold the obligations and face the supervisors, including a bank licensed by the Finnish Financial Supervisory Authority. The company publishes substantial material on forthcoming European anti money laundering regulation, which is subject matter expertise rather than standing.

No fairness disclosure, no evaluation across jurisdictions or business types, no governance framework. The exposure specific to this design is automation bias rather than model bias: the agent drafts the decision and the human verifies, so the framing of the summary shapes what the reviewer concludes before they begin, and a reviewer working at speed against a productivity target measured in cases cleared is being nudged toward confirmation.

Nothing published examines whether drafted recommendations are accepted at different rates for different customer types, or whether sanctions and politically exposed person matching performs evenly across the naming conventions of the many jurisdictions covered.

Nothing states who bears the cost of a wrong decision or how one is contested. The orchestration architecture adds a specific complication that no published material addresses: when an outcome is produced from a chain of third party providers assembled by this platform, and then summarised and drafted by its agent, responsibility for an error is divided between the underlying data provider, the orchestrator and the institution that acted, and nothing describes how it is allocated. A business wrongly refused has no relationship with any of them, no notice, and no route to correct a record that may sit upstream.

The same split this index recorded on another orchestration vendor in this pocket, and now seen twice: the verification and data supply chain is disclosed at least partly, with Veriff named and the categories of registry, sanctions and electronic identity providers described, while the model supply chain is not disclosed at all.

No provider, family or version is named for the agent or the co pilot, no country of processing is stated, and nothing says whether case data, which includes beneficial ownership details and identity records, is routed to an external model provider or what that provider may retain.

The whole commercial proposition is integration depth: one connection replacing an entire local compliance stack, reaching European company registries, beneficial ownership data, sanctions and politically exposed person lists, fraud signals and electronic identity providers. A dedicated integrations page, public developer documentation, an API reference and a changelog support it, so the surface is inspectable before contracting.

Held below an A because only one orchestrated provider is actually named, Veriff, with the rest described by category. A platform whose value is the breadth of what it connects to should be able to list what it connects to, and a buyer replacing a known stack needs to know which of their existing providers survive the migration.

A hosted multi tenant platform with no published deployment or residency information: no cloud provider, region or jurisdiction named, no single tenant or private option, and no statement of where case files, identity records or beneficial ownership data rest.

The absence is notable for a vendor whose entire positioning is European, whose customers are supervised in the Nordics and Finland, and who publishes extensively on the incoming European anti money laundering regime, since data location is one of the questions that regime and the institutions inside it will ask.

No pricing published at any level and no statement of the charging basis, which matters for an orchestration platform because the underlying providers are separately metered and a buyer cannot tell whether cost follows checks, entities, providers enabled or seats. A small inconsistency worth recording: the primary call to action is labelled as starting a free trial, and the page it leads to presents a demo booking, so what looks like self service is a sales conversation.

Genuinely broad and, unusually for this pocket, demonstrated rather than claimed. Coverage spans more than 100 jurisdictions with expansion into the DACH region and the United Kingdom, and roughly twenty two customers are named publicly across six distinct financial segments: a licensed and Nasdaq Helsinki listed bank, six payments and buy now pay later firms, a crypto exchange, three investment and fund platforms, several lending and credit businesses, and adjacent legal and software companies.

Reaching a supervised listed bank, a regulated crypto platform and a fund launch on the same platform is real institutional breadth for a company founded in 2022, and the segment spread is evidenced by named logos rather than by an industries menu.