Skip to content

Build1 publisher2 min readPublished

Footprint raises $25M for compliance agents that log a citation and a timestamp per step

QED led the round and MUFG joined as one of four new investors. Under Trust Fabric, a human reviewer has to approve a finding before later investigations can reuse it.

The Engineer · Build desk

Illustration accompanying Footprint raises $25M for compliance agents that log a citation and a timestamp per step

What happened

  • Eli Wachs and Alex Grinman announced on September 16th that Footprint raised a $25M Series B led by QED Investors, which had led the company's Series A two years earlier.
  • MUFG, Commerce Ventures, LightBank and Alumni Ventures joined the round, alongside existing investors including Index Ventures, Lerer Hippeau, BoxGroup, Operator Partners and Animal Capital.
  • The capital goes to Percy, Footprint's system of AI agents for financial-crime investigations, and to Trust Fabric, the data and governance layer underneath it.
  • Footprint says Percy can conduct anti-money-laundering reviews, enhanced due diligence, sanctions checks, know-your-customer and know-your-business workflows, and transaction-monitoring investigations.
  • The company plans to double its engineering and sales organizations and open a San Francisco office, and it has not provided a current headcount.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • constraint Because a human reviewer has to approve each addition to Trust Fabric's memory, the reviewer queue sets the ceiling on how much of the investigation load the agents can absorb.
  • exposure A bank that adopts Percy makes a vendor's step log and version history the artifact its examiners read when asking how a decision was reached.
  • precedent Bank corporate venture money is now an entry point for compliance-agent vendors before any institution has been shown running one on live cases.

An investigation that reuses an earlier finding costs less than one that starts cold, and reuse is also where this design can go wrong. Footprint says Trust Fabric retains the precedents its investigations produce, that a finding from one sanctions or due-diligence case can be applied to later work, and that human reviewers approve proposed additions to that organizational memory [7]. One approval therefore propagates into every later case that inherits the precedent.

Inside a single case, Footprint says Percy accepts plain-English instructions, connects to sources including LexisNexis, Experian, ComplyAdvantage and government registries, and records citations, timestamps and each step taken during an investigation [5]. The company also says agent changes are versioned and tested before deployment [8]. That version record is the harder requirement for a buyer. If a tool or a procedure changes, every case after the change was investigated differently, and the trail has to show which version ran.

Runtimewire, working from a PR Newswire release [20], wrote that MUFG's investment gives Footprint a path into institutions where governance and audit trails matter as much as model performance [14]. QED led the round and MUFG joined it [1][2]. The account names no bank customer and reports no accuracy or review-time figures for Percy [19].

The $25M is roughly 1.9 times the $13M Series A of May 2024 [17][9]. Product scope moved faster than that. Percy shipped in March 2025 as a beta assistant that let reviewers question fraud signals, inspect triggered rules and act on user classifications [11]. Nine months later, Grinman was describing Footprint as an AI-native system of record for onboarding, compliance and sensitive data [18][12].

Footprint started in 2022 as a portable identity and secure data-vault product [10], and in a 2023 Fast Company profile Wachs called an early version an "Apple Pay of identity" [13]. Investigations sit in a larger and more persistent operating budget than onboarding checks. Runtimewire notes the move exposes the product to much higher standards for consistency and auditability [15]. The publisher's argument for the architecture is that speed alone has limited value if a bank cannot reconstruct how an agent reached a conclusion, which data it used and whether it followed the bank's written procedures [16].

For the audit layer to transfer to another institution, a few things have to hold. Examiners have to accept a vendor's step log as the record of how a decision was made. Written procedures have to be expressible as the instructions Percy takes. And reviewers have to approve memory writes at the rate the agents generate them, the one step Footprint says a human performs [7].

What to watch

  • A named bank or fintech disclosing Percy on live AML or sanctions work, with review-time or false-positive figures attached.
  • Independent attestation of Footprint's claim that agent changes are versioned and tested before deployment.
  • Whether human approval of Trust Fabric memory additions stays mandatory as customers push for case throughput.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories