What is the Cogent AI Framework?
PredictX’s agentic AI framework for travel and expense. It splits complex T&E work across specialist agents and orchestrates it end to end: a Router Agent plans every request, specialists execute in parallel across 200+ governed connectors, and an accuracy and reliability layer validates every outcome before release. Two products run on it: Cogent, the workspace, and Orchestra, the agent team. Every answer is cited, permission-governed and auditable.
Is Cogent a chatbot?
No. A chatbot waits for a message and replies from one system’s data. Cogent plans the work, delegates it to specialist agents, executes complete T&E workflows, validates the result against your business rules, and can start work from autonomous triggers rather than prompts. A chat surface is one way to brief it. The framework doesn’t assist with the work. It does it.
How does agentic AI improve T&E policy compliance?
It moves compliance from after-the-fact reporting to real-time intervention. The framework audits every booking, card and expense record continuously against your live policy, surfaces out-of-policy spend by named traveller, and catches fraud before reimbursement rather than after. 24% of travel buyers cite non-compliant bookings as their top risk priority; continuous agentic auditing addresses it as it happens. Source: GBTA.
How does the framework guarantee accuracy and reliability?
Through layered validation rather than model confidence. Findings are scored against your business rules, an independent validation agent audits every output before anyone sees it, source authority resolves conflicts between records, and surprising results pause at a discovery checkpoint for human review. Every answer is traced to its source query, and it targets a 99%+ acceptable response rate on relevant programme queries.
Does it replace our existing booking and expense platforms?
No. The framework is the intelligence layer that sits across your existing stack. It consolidates data from your corporate travel management software, expense platforms and TMCs, alongside card, GDS, HR and General Ledger feeds, through 200+ maintained connectors, then runs the investigations those systems’ reporting modules cannot. Booking and expense submission stay exactly where they are. No rip and replace.
How long does deployment take?
Minutes, not months. The framework runs on your existing estate through 200+ maintained connectors, so there is no rip and replace and no new point integrations to build, and permissions carry over from your current PredictX deployment. Technical teams typically stand up their first deployed agent in around 10 minutes; enterprise onboarding then configures workspaces, policies and schedules.
What data sources does it connect to?
200+ maintained connectors across TMCs, expense systems, corporate cards, OBTs, GDS, suppliers, HR and General Ledger, consolidated into one governed estate across every entity and geography. Source authority resolves conflicts between records, and every agent inherits the requester’s role-based access down to row and field level.
How do humans stay in control?
Control is configured, not assumed. You define which decisions agents take autonomously and which route to a named human, reading and writing carry separate permissions, surprising findings pause at a discovery checkpoint, and the framework never commits spend without a final human Go or No-Go. Every agent action is logged in full and reversible by design.
