Flank

What it is

Flank builds autonomous agents for in-house legal teams that work where requests already land: an Outlook inbox, Microsoft Teams, Slack, an intranet portal, or a custom URL. An agent takes a request end to end. It triages it, drafts or redlines the NDA, DPA, or vendor contract, handles the counterparty’s objections, and answers the policy question. A lawyer approves finished work and anything flagged as high risk through Flank’s supervision layer. The second product, Flank Record, launched on 23 July 2026. It is an agent-maintained contract system of record: agents capture metadata from each executed contract at signature, trace every field to its source clause with a confidence score and the name of the person who verified it, flag amendments and duplicates, and answer questions like “which contracts expire this month?” in Teams, Slack, or email. Flank started in Berlin as Legal OS, a no-code legal automation platform, before it turned to agents. It is led by co-founders Lili Breidenbach (CEO) and Jake Jones (COO). In June 2025 it raised a $10M round led by Insight Partners, with Gradient Ventures, 10x Founders, and HV Capital following on, which brought total reported funding to about $18M. Named customers include DeepL, SumUp, Perk (formerly TravelPerk), Lusha, PROS, Axel Springer, the Financial Times, Bolt, and Accor. Simmons & Simmons runs Flank agents in production for its clients.

Why it shows up in Legal Ops stacks

• The agent replaces the request queue instead of feeding it. Checkbox and Streamline AI are front doors: they capture, route, and track a request until a lawyer picks it up. Flank’s agent does the work itself and sends the result back in the same email thread or Teams chat, with no ticket in between. Perk reports the agent handles 5,000 requests a month. That figure is vendor-reported, published in Flank’s June 2025 funding announcement.
• Record puts agents under the data layer, not on top of it. Most legal AI answers questions from whatever sits in the existing repository, so a wrong renewal date in the CLM comes back as a confident wrong answer. Record extracts metadata from the signed version rather than the draft, reconciles duplicates daily, and needs a human to confirm a field before it enters the record. The DocuSign and Ironclad connectors are the two Flank names publicly. Flank says plainly that Record is “not a new CLM”: it has no CLM interface for your team to operate.
• It is an EU-native vendor with enterprise controls on the main site. Flank lists SOC 2 Type II, ISO 42001, GDPR, CCPA, and Saudi PDPL. It also advertises regional tenancy, zero data retention, no training on customer data, SSO, role-based access, and full audit logs. For a legal team in Europe or the Middle East that has to answer residency questions before a pilot starts, that list is the shortest route through procurement in this segment.

Pricing reality

Flank publishes no price, offers no self-serve tier, and has no free trial. Every route ends at a demo, and deployment runs on a forward-deployed engineer model in which Flank engineers build your first agent with you in a stated 2–4 weeks. The only public pricing signal is the line on the Record page: customers “pay for answers, not seats”. That points to volume-based billing, but no unit or rate is published. Record adds roughly three months of back-catalogue ingestion and a parallel run. Flank says it does the migration at no charge.

For a budget anchor, use the segment. Streamline AI publishes a $22,900/year floor for four core users, and GC AI charges $500/seat/month, or $6,000/seat/year. A vendor that ships engineers to build your agents sits above the self-serve end of that range, not below it. Plan for an enterprise annual contract, and do not build a business case until you have a written quote that names the billing unit.

Best for

Legal ops leads and GCs at enterprises with roughly 1,000 or more employees, in Europe, the Middle East, or the US. The fit is a business that sends a steady volume of repeat documents (NDAs, DPAs, vendor and franchise agreements) through Outlook and Teams, and a legal team that wants that work done for them rather than routed to them. The specific trigger is a department that already knows its request volume, has written playbooks for its top three document types, and is still losing lawyer hours to first drafts and first-round redlines. It is also the pick for a law firm that wants to offer agent-handled routine work as a client service, which is the Simmons & Simmons model.

Versus the alternatives

Ironclad is the incumbent when the contract record is the purchase. Pick it when you need a full CLM with workflow designer, obligations, and renewals that your team operates itself. Pick Flank Record when you want the record kept accurate by agents and nobody on your team wants to run a CLM. Harvey is the category leader in legal AI. Pick it when lawyers want a research-and-drafting assistant they drive themselves; pick Flank when the business, not the lawyers, is the one asking. Wordsmith is the closest head-to-head. It also resolves requests in Slack, Teams, and email, but it adds an official MCP server and a metered self-serve tier with a $200/month default budget cap. Pick Wordsmith if you want to pilot without a contract or call legal review from your own agents; pick Flank if you want Flank’s engineers to build the agents and you need Record. Checkbox is the fastest-growing entrant in the front-door segment ($23M Series A in January 2026). Pick it when legal ops wants a no-code canvas it configures itself. GC AI at $500/seat is the pick when the lawyers need a better tool and there is no business-side intake problem to solve. LegalOn and SpotDraft fit when playbook-driven contract review is the whole job. If none fit, start with a shared legal inbox, three approved templates, and a routing rota for one quarter. That costs nothing and gives you the request-volume baseline every vendor in this list will ask for.

Watch-outs

• No price, no trial, and an unpublished billing unit. “Pay for answers” is attractive until you find out what counts as an answer. Guard: before the scoping call ends, get in writing what is metered, the price per unit, what happens when volume doubles, and whether a contractual cap exists.
• Record is waitlist-only and three months from useful. It launched on 23 July 2026 through a design-partner programme. The connector list on the public page names two systems, and no independent accuracy data exists yet. Guard: keep your current repository live through the parallel run, and switch off the old tool only after you have sampled at least 100 Record entries against the signed PDFs yourself.
• Every efficiency number is Flank’s. The 5,000 requests a month and the customer testimonials are vendor-published, and none is independently audited. Guard: scope the first agent to one document type, and gate expansion on a deflection rate and a lawyer-hours delta you measured over 60 days.
• The forward-deployed model means Flank engineers build and change your agents. Deployment is fast, but you depend on them for every change. Guard: contract for who can edit playbooks and agent instructions without a ticket to Flank, and for export of your playbooks, request history, and Record metadata on exit.
• Autonomy needs explicit approval thresholds. An agent that replies directly to the business is a risk if nobody has written down what it can send without a lawyer. Guard: launch with lawyer approval required on every output, then remove it per request type only after a clean run of at least 50 items.