Taktile

The foundation is a decision engine rather than a model. Risk teams assemble underwriting and onboarding logic from low code building blocks, connect data sources, and deploy policies they own, and that product works without any machine learning in it.

AI sits on top in three places that do real work: a copilot that drafts and debugs decision logic from plain language and generates Python, an agent manager with templates for financial services tasks, and agents that extract data from documents or read financial statements. Apply the removal test and a capable decisioning platform survives, which places this alongside the orchestration vendors rather than the model native ones.

Oversight is stated repeatedly and, more importantly, built into the workflow. The company commits to maintaining transparency, human oversight and control across automated workflows, describes its approach as blending machine speed with human oversight, and positions agents as operating alongside underwriters with each handling a designated task rather than owning the decision.

The mechanism that earns the top grade is validation before deployment: changes can be tested in seconds with expected outputs shown, backed by backtesting, split testing and performance monitoring, so a policy change is proven against historical cases before it touches a live applicant. Risk teams own and can read the logic they are deploying.

The strongest ongoing monitoring toolkit in the index so far, and it comes as product rather than paperwork. Backtesting against historical cases, split testing between competing policies, continuous performance monitoring and real time analytics showing how strategies behave across the customer lifecycle together give a validator the evidence base that model risk guidance actually asks for, and because the institution authors and can read the logic, the decision path is inspectable end to end rather than inferred from a score.

Two things hold it below the top grade: no formal model documentation, validation summary or stated position on supporting customer validation, and no account of how copilot generated logic is reviewed before it becomes part of the model inventory.

Customers are named and their results are specific, which is the standard this axis is meant to reward. A working capital lender reports a 95 percent reduction in underwriting time and its risk team handling three to five times more applications, a Latin American lender reports 67 percent faster deployment of policy logic and double the experimentation volume across fraud, credit and portfolio workflows, and a buy now pay later provider's head of credit risk is quoted by name and title on cost savings and independence from engineering.

Third party standing comes from four consecutive quarters as a category leader in a decision management platform ranking. A leading investment bank led the most recent funding round, which means an institution with its own risk expertise diligenced the company.

One capability here deserves more scrutiny than it currently receives anywhere. The copilot generates Python that becomes decision logic executing against live credit applications, which means machine written code is running inside a regulated decision path where an error produces wrongly denied applicants or unpriced credit risk.

Nothing public describes what review or approval that generated code passes through before deployment, who is accountable for it, or how it is distinguished in an audit from logic a person wrote. Separately there is no disclosure of which models are used or supplied by whom, and no statement on whether customer decision data informs anything shared.

The platform sits at the point where the most sensitive data a lender handles converges, pulling identity verification, bank account ownership and real time cash flow data through connected providers and combining it with uploaded financial statements and documents to reach a credit decision.

No published privacy framework, retention schedule, subprocessor list or statement of financial privacy service provider obligations was located in this pass, and none of the connected data flows is documented from a privacy standpoint on the vendor's own material.

This pass surfaced no trust centre, certifications page, attestation list or scope statement on the public site. Customers include regulated banks, credit unions and insurers whose vendor risk programmes would require attestations before allowing a third party into a live credit decision path, so the actual control environment is very likely stronger than the published record shows. The grade records what a buyer can verify without entering a sales process and should be revisited if a trust surface is published or located.

Taktile supplies software and holds no financial licence, the expected posture. Its product scope crosses several regulated domains at once, covering credit underwriting subject to lending and fair credit rules, customer due diligence and transaction monitoring subject to anti money laundering obligations, and insurance claims decisioning, across United States, United Kingdom and German operating bases with their differing regimes. The material is clear that the institution owns the decision and therefore the obligation, which is the correct framing, though no specific authorisation or formal admission is published.

Credit underwriting is the headline use case, which places these decisions inside equal credit opportunity rules where adverse action reasons must be specific and disparate impact is an active supervisory concern, and no fair lending testing, demographic performance analysis, adverse action reason code documentation or independent audit is published. Two features sharpen the problem rather than easing it.

The copilot writes decision logic, so a lender may deploy criteria no person composed and cannot fully explain to an examiner. And agents reading financial statements introduce inference steps between the applicant's documents and the decision. The backtesting and split testing infrastructure is exactly what a disparate impact analysis would run on, which makes the silence about whether anyone uses it that way more conspicuous, not less.

Taktile gives an institution genuine means to verify before harm occurs, since a policy can be backtested against historical cases and split tested before it touches a live applicant, which shifts responsibility to a customer who has been handed the tools to discharge it. Nothing binds the vendor. No accuracy guarantee, no remediation term, no stated position on copilot generated logic that produces a wrong decline, and no correction route for a declined borrower.

The data and detection chain is named at the points that matter, including a loan origination platform, an identity and cash flow data provider, a synthetic identity and first party fraud detector resold directly, and an agentic document fraud service surfaced inside decision workflows. A buyer can therefore enumerate most third parties touching a decision. The model layer is the gap: nothing states which provider powers the copilot that writes production credit logic, or where generated code is processed.

Integration reaches in both directions and into the systems that matter. Upstream a loan origination partnership plugs the decision platform into the workflow banks and credit unions already run for verification, document collection, underwriting and core booking.

Sideways, data and detection providers connect natively, including identity verification and cash flow data, synthetic identity and first party fraud detection through a reseller arrangement, and agentic document fraud detection surfaced directly inside decision workflows. The build surface itself is dual, with pre built nodes for non technical users and generated Python underneath for full code level control, so a team is never blocked by the abstraction.

Delivery is cloud hosted software as a service operated from bases in the United States, Germany and the United Kingdom, a footprint that implies European infrastructure and brings European data protection requirements directly into scope for customers on that side.

None of it is described publicly: no hosting regions, no residency options, no transfer mechanism and no subprocessor list, which matters for a platform routing applicant identity, bank account and cash flow data between providers as part of every decision.

No pricing, tier structure, billing unit or minimum is published and paths lead to a demo request. The gap is worth noting for this product specifically because the platform is sold on the promise of independence from engineering resources, which is a cost argument, and a buyer cannot weigh the licence against the internal engineering time it replaces without a number to compare.

Coverage runs across fintechs, banks, credit unions and insurers, with a loan origination partnership aimed specifically at banks and credit unions, and use cases spanning onboarding, credit underwriting, fraud, transaction monitoring, claims and collections, so the platform reaches into insurance decisioning as well as lending. Operations span New York, Berlin and London, giving genuine presence in both the United States and Europe. What is absent is the segment by segment depth the strongest vendors in this index publish, with no dedicated material per institution type and no stated customer counts by segment.