How We Automated 80% of Month-End Close with Claude + Kognitos
Most finance teams using Claude for the close have a Governance Gap: Claude interprets instructions, but there is no deterministic layer beneath it to guarantee execution, enforce approval chains, or produce a defensible audit trail. This session, with George Williams (VP of Solutions Engineering), closes that gap live, three close workflows running through Claude, governed end-to-end by Kognitos MCP. Watch above, then book a demo to map it to your close process.
See how Kognitos works for you.
More Webinars
Session FAQ
Binny Gill (Founder & CEO, Kognitos), Niraj Mathur (Chief AI Officer) and George Williams (VP Solutions Engineering) walk through how a Claude-powered Kognitos agent can compress a finance team's month-end close from 8-12 days to as little as 1-2 days. The session covers a live Slack-driven demo, the architecture of deterministic agents on top of Claude, how exceptions are handled in plain English, and a live Q&A on tooling, integrations and security.
Controllers, accounting managers, R2R / close leaders, FP&A heads who consume the close, and CFOs evaluating AI for finance. It is also useful for IT and CISO teams who need to understand how a Claude-backed agent is deployed safely against ERP and reconciliation data.
The session cites the industry average at roughly 8 days, with accounting teams spending about 60% of their time on close-related work and one in four closes requiring post-close adjustments. In a live poll only 5% of attendees said they close in under 3 days. Kognitos customers have compressed their close to 1-2 days for the targeted workstreams by automating reconciliations, journal entries and document gathering with a Claude-backed agent.
Kognitos is LLM-agnostic. The Slack experience in the demo runs on Claude, but the same orchestrator can run on OpenAI, Gemini or another foundation model. A deterministic top-level orchestrator owns the dependency graph between automations; the LLM only handles language interpretation. You can change the underlying model without rewriting your processes.
No. The presenters address this concern directly: Kognitos does not train any LLM on customer data. Personalisation comes through teaching, not training. You write SOPs and troubleshooting playbooks in plain English, the platform follows those rules deterministically, and the underlying model can change without retraining. Your data stays your data.
The demo shows agents pulling source documents, matching balances across systems, computing journal entries (often 100% deterministic math with no LLM call), posting to the ERP, and escalating only the items that need a human decision. Critically, the user interacts with the agent through Slack or email in English; the workflow itself is policy-driven and fully auditable.
Yes. The webinar's audience Q&A includes a specific question about Oracle EBS, and the presenters confirm support across the major ERPs including SAP, Oracle, NetSuite and Workday. Where a connector does not exist, the platform can be authored against any system with an API or even via UI automation, with the same audit and policy guarantees.
When the agent is not confident or hits a scenario the SOP does not cover, it raises a plain-English question to the right business user instead of guessing. The user responds, the answer is captured into the SOP, and the agent handles the same situation autonomously next time. This is how the platform learns without ever training the underlying model on your data.
Yes. Every registrant receives a recording link after the session, and the recording plus supporting resources are available on demand on this page. You can also book a personalised demo to see the same workflows running against your own ERP and close calendar.
Start with the heaviest, most repetitive close steps: bank and account reconciliations, intercompany matching, and recurring journal entries. Add prepaid amortization, accrual proposals and balance-sheet flux next. These give a measurable close-day reduction in 30-60 days and create the policy patterns you'll reuse for harder areas.