Kognitos

• Taking supplier invoices from email intake through extraction, coding, PO and receipt matching and ERP posting, with mismatches routed to a person
• Screening each incoming invoice for exact and fuzzy duplicates and blocking or holding a suspect payment for approval
• Running bank, sub-ledger and balance sheet reconciliations each period and certifying them with a readable record of every step
• Turning a control question such as "did any supplier's bank details change without verification and independent approval?" into a daily monitor
• Running KYC and customer due diligence checks during client onboarding at a bank, with exceptions escalated to an analyst
• Letting an internal agent in Claude or Copilot Studio query vendor and payment context through the Context Graph's MCP or REST access

Kognitos is worth a close look if your obstacle to AI in finance is not capability but defensibility: approved SOPs, a deterministic engine, human sign-off on anything new and a journal per run map directly onto how controllers and auditors already think about controls. The AI Employees catalog is the clearest way in, because each role shows which SOPs are ready and which still need building, which tells you how much of the work is Kognitos's and how much is yours. Treat it as an enterprise rollout rather than a tool: AI Employees are in preview for Kognitos customers, implementation runs through Kognitos's solutions architects, and pricing is consumption-based with no public rates. Start with one high-volume, rule-heavy process such as invoice intake or duplicate payment screening, measure how often runs pause for a person over the first weeks, and ask for a written estimate of transaction-based cost before expanding.

About Kognitos

Kognitos calls its approach English-as-Code: you describe a process the way you would explain it to a colleague, a builder agent turns that description into a written procedure (an SOP), and once a person approves and publishes it, the English is what runs. The design is neurosymbolic. Language models read documents and interpret requests, but execution is handled by a deterministic engine that, according to Kognitos, never adds a step that is not in the approved SOP. When a run hits something the SOP does not cover, it pauses at that step, explains the problem in plain English and waits for a person; an answer becomes reusable guidance only after an approver signs it off. Every run keeps a journal of each step and each call to an ERP, bank or inbox, tied to the SOP version that ran, and runs can be replayed. On top of that core, Kognitos has added three finance-focused layers in 2026. Context Graphs (launched in August) map records such as bills, vendors and payments from your connected apps, plus the knowledge your team adds, without writing back to source systems; ready-made monitors turn questions like "are we paying any vendor twice?" into scheduled, deterministic controls, and the graph is open to your own agents over REST and MCP. Apps provide 65 finance and 39 banking applications for the close, reconciliation, cash application, treasury, KYC and more. AI Employees (announced October 21, 2026) package the platform as 61 finance and banking roles, such as AP Specialist, Reconciliation Accountant, Close Manager and KYC & Onboarding Analyst, backed by a library of 515 SOPs, with Quill, the builder agent, drafting any SOP a role is missing. The platform itself is not finance-only (Kognitos also sells into healthcare, supply chain and IT operations, and its docs cover HR and sales workflows), but its product, roles and controls are now built around the office of the CFO. The tradeoffs: AI Employees and Context Graphs are in preview for Kognitos customers, there is no published price beyond a free workspace and consumption-based enterprise contracts, and production rollouts are typically done with Kognitos's solutions architects rather than fully self-serve.