Taktile

The foundation is a decision engine, not a model. Risk teams assemble onboarding, underwriting and claims logic from rules, decision tables, scorecards and Python nodes, and that engine runs without any machine learning. AI does real work on top of it. Eleven prebuilt agents cover tasks such as financial spreading, payslip reading, ownership tracing, adverse media checks and credit memo drafts. Taktile AI, the in product assistant, helps build, query and debug flows. Strip the AI out and a capable decisioning platform still stands, so the models drive key features rather than the whole product.

Taktile publishes how automation and people share the work. Each customer builds its own flows from rules, decision tables, scorecards and models, so the customer sets what goes straight through and what is escalated. Complex cases go to a case manager, with queues configured by escalation reason, and AI drafts the case summary while a reviewer makes the call. Agents each handle a set task inside the flow rather than owning the decision.

Every decision is kept in an immutable history that replays each step, including human actions, and approvals are tied to the exact version that reached production. Policy changes are backtested on historical cases and A/B tested before full release.

Validation tools are built into the product. Policies are backtested on historical cases and A/B tested live before full release. Every decision is stored in an immutable history that replays inputs, rules, models, agents and human actions, with a plain language explanation. Each published version is kept, rollbacks keep the audit trail, and sign offs are tied to the version that reached production.

Taktile states that these controls align with SR 26-2, the EU AI Act and DORA, though it publishes no detailed mapping. Taktile Labs publishes benchmarks with methods and results, such as KYBench, which tests 25 frontier models on adverse media checks. Sign offs apply to every version, but no separate check is described for logic written with help from Taktile AI.

Named customers report specific results. Zilch moved underwriting for more than 4 million customers onto Taktile in four weeks and cut decisioning costs by more than half. Breakout Finance cut underwriting decisions from over an hour to as little as three minutes and handles three to five times more applications. Allianz Partners reports 30 percent more claims paid straight through and 4 to 6 percent better adjudication quality, with Pieter Viljoen, its chief data and AI officer, quoted. Willow Wealth cut case review time by 83 percent. Growth Equity at Goldman Sachs Alternatives led Taktile's $110 million Series C in June 2026.

Each customer workspace runs on its own compute and data plane. Taktile says it uses shared infrastructure only for operations that do not involve customer decision data. Its data processing agreement bars use of customer personal data beyond the contract's purposes. AI features are optional, and language model processing happens only when a customer turns them on, in the regions the subprocessor list sets out. Agents can learn from a customer's own outcomes. What is missing is an explicit statement that customer data never trains a model shared with others, Taktile's or a provider's, and any retention terms at the model providers.

The privacy terms for customer data are public in detail. Taktile's data processing agreement sits on its website. It sets AES-256 encryption at rest, TLS 1.2 or higher in transit and a 48 hour breach notice. It sets default hosting by contracting entity, in the United States, Germany or the United Kingdom. On termination, Taktile deletes customer personal data within 90 days unless the customer asks for its return, and lenders can instruct longer retention.

A public subprocessor list, updated 27 February 2026, names every processor, and Taktile gives 30 days notice of changes. Taktile LLC is certified under the Data Privacy Framework for transfers from the EU, the UK and Switzerland. The privacy policy itself covers only website and business contact data.

Taktile states that it is certified under ISO 27001 and SOC 2 Type II, with regular independent audits and annual penetration tests by outside experts. Its data processing agreement lets customers rely on those reports in place of their own audit. The agreement refers to a Trust Center, but none is linked from the public site, and neither report can be downloaded there.

Taktile supplies software and holds no financial license, which fits the product. Its scope crosses credit underwriting, onboarding checks, anti money laundering screening and insurance claims, across the United States, the European Union, the United Kingdom and Brazil. Taktile says its controls are aligned with requirements such as SR 26-2, the EU AI Act and DORA, and its material leaves the decision and the obligation with the institution. No register entry, supervisory designation or formal admission is published.

Credit underwriting is a headline use case, which brings ECOA, adverse action reasons and disparate impact into play. Taktile's own fair lending testing is not published. The fairness route it offers runs through a partner. Since July 2025, lenders can place FairPlay's tools in their Taktile credit policy logic to detect bias across demographic groups, take a second look at declined applicants and produce fairness reports for regulators.

No results from that integration, and no demographic analysis, are published. Agents that read financial statements and AI that helps write rules add steps between the applicant's documents and the decision, and no fairness review of those steps is described.

An institution can test before harm occurs. A policy can be backtested on historical cases and A/B tested before it reaches a live applicant, and every decision can be replayed. That puts the means of challenge in the customer's hands. Nothing binds Taktile itself. There is no accuracy guarantee or remediation term, no stated position on logic written with AI help that produces a wrong decline, and no correction route for a declined borrower.

The chain is named in a public subprocessor list, last updated 27 February 2026. Amazon Web Services hosts every customer, region by region. When a customer turns on language model features, Anthropic models served through AWS are the default, and Google and OpenAI are optional, each with its processing region shown. Parallel Web Systems orchestrates AI agents when they are enabled. Resold data and detection services are named too, including Socure, Middesk, Equifax, Inscribe, Ocrolus, Sumsub and ComplyAdvantage.

Integration reaches the systems lenders run on. MANTL Loan Origination, an Alkami product, embeds Taktile's decisioning for community banks and credit unions, and Alkami's own release of 5 September 2025 documents it. LoanPro is among the named integrations, and the connector library adds data warehouses such as Snowflake, Databricks and BigQuery. Data partners include Equifax, Experian, TransUnion, Plaid and Prism Data, whose CashScore and cash flow attributes run inside Taktile flows. Product documentation sits behind a customer login.

Taktile runs on Amazon Web Services and hosts each customer in its preferred AWS region. The subprocessor list names 11 hosting regions across the United States, Europe, Canada, Brazil, India, Singapore, Australia and Japan. By default, US contracts host in the United States, EU contracts in Germany and UK contracts in the United Kingdom, with Canada or Brazil on request. Each workspace gets its own compute and data plane. Some model providers process data only in the United States or the EU, as the subprocessor list shows. No on premise or private cloud option is published.

No price, tier, billing unit or minimum is published, and every route leads to a demo request. That gap matters for this product. Taktile sells independence from engineering teams, which is a cost argument, and without a number a buyer cannot weigh the license against the engineering time it replaces.

Taktile names banking and insurance as its markets and sells to banks, credit unions, fintechs and insurers. Dedicated solution pages cover business onboarding, business underwriting, consumer onboarding, consumer underwriting and claims management, with motor, property and casualty, and health claims named. Community banks and credit unions reach it through MANTL Loan Origination, an Alkami product. Named customers range from Zilch and Monzo to Allianz Partners and Faire. Offices sit in New York, Berlin, London, São Paulo and Iași. No page is dedicated to a single institution type, and no customer counts are given by segment.