Cogent is PredictX’s agentic AI framework for travel and expense management: specialist AI agents plan, execute and validate complete T&E work for you, and every answer arrives cited, traced to source and auditable.
You read one answer. A team of specialists did the work.
PROOF · LIVE DEPLOYMENTS
01 · SCENARIO MODELLING
£600K–£800K
Global financial services · 20,000+ travellers
A business-class threshold change was modelled across roughly 450 affected flights, projecting £600K to £800K in annual savings. Manually: two to three weeks of analyst work. With Cogent: one brief, seconds, a finance-ready export with country-level breakdown.
02 · OFF-CHANNEL DETECTION
145 instances
Global pharmaceutical · 10,000+ travellers
Expense fraud detection in one brief: 145 instances of off-channel spend surfaced, 80% traced to one department at one conference, protecting an estimated £45K to £55K in a single quarter.
03 · GROUND TRANSPORT BLIND SPOT
$4.44M
Global pharmaceutical · enterprise programme
One query exposed that 99.99% of rideshare spend carried no location data: a category most programmes cannot see at all. Cogent surfaced the blind spot in a single brief.
04 · ONE SOURCE OF TRUTH
3 → 1
Global biopharmaceutical · executive reporting
Three systems reported the same YTD air spend three ways. Cogent reconciled all three to one canonical, defensible figure, before the board meeting rather than during it.
05 · HIDDEN HIGH-VALUE TICKETS
17 tickets
Global biopharmaceutical · premium cabin audit
A search for premium tickets on one transatlantic route returned 2 results on the initial filter, then 15 more once the return leg was included: 17 tickets over $12,000 missed entirely.
“Every travel technology vendor promised us intelligence. PredictX is the only one that delivered autonomy. There is a significant difference between the two.”
HT
Head of Global Travel
Fortune 500 Manufacturing Group
The Router Agent receives every query, classifies the intent, and creates a detailed plan across sub-agents. This is multi agent orchestration in practice.
Every brief is understood before anything runs, so compliance and sourcing questions take different paths.
Complex T&E work is broken into small, focused jobs with a clear owner, input and output.
Specialists run simultaneously, not in sequence; answers arrive in seconds rather than days.
Results return as one coherent, cited answer.
Specialist AI agents work in collaboration, each modelled after a real T&E role, not a generic model dressed up. Higher accuracy. Faster answers. Workflows that grow without adding headcount.
Interrogate consolidated booking, card, expense, HR and finance records at transaction level.
Read contracts, proposals and policies, with 100+ page PDF document comprehension.
Bring external fare, rate and supplier signal into the same analysis.
Continuous compliance scrutiny alongside DetectX, catching issues before reimbursement.
Agents do not just answer from your data; they work with it: a governed semantic layer over your whole estate, with retrieval augmented generation across your own documents, plus a toolkit including live web research. No rip and replace.
200+ maintained connectors: TMC, expense, card, booking tools, GDS, supplier, HR and General Ledger, unified.
Agents reach beyond your estate for live fares, capacity, routes and supplier signal.
When records disagree, a live booking record outranks a six-month-old PDF.
Every agent respects the requester's role-based access, to row and field level.
A confident-sounding answer is not the same as a correct one. The framework eliminates hallucinated answers through multi-step validation: every answer is checked, cross-referenced and grounded in your data before it surfaces, scored against your own rules rather than model confidence.
Data is cleaned and reconciled during ingestion, before the AI sees it.
A separate agent audits every output before release.
Surprising results pause for human review before a report is written.
Every answer records what was asked, what answered, traced to source.
Every answer shows why it was reached, not just what it concluded.
“The interface was never the hard part. The hard part sits underneath: an independent validation layer between the model and the answer. That is what lets an enterprise delegate whole T&E workflows, not just questions.”
.png)
Agentic AI is already running these workflows on this framework, from expense management automation to full sourcing cycles.
Automated expense reporting: compose, format and deliver auditable reports on demand, every figure traceable to its source query.
Build data-backed negotiation briefs from contract terms, historical spend and live market signal.
Q3 travel spend, board-ready.
Report compiled: 4 regions, every figure cited to source.
Continuous auditing: agents autonomously scan live data and surface opportunities with action plans, including T&E leakage detection.
Natural-language interrogation of every spend dimension, including forecasting and budget variance, down to entity-level travel spend analytics, with the working shown.
Procurement analytics, sourcing, supplier evaluation, RFP automation and contract execution run as delegated agent jobs.
Chain agents into repeatable jobs, triggered on schedule, on event, or on demand from any connected interface.
BUILT ON THIS FRAMEWORK
The framework ships inside two products and one free library.
AI T&E reporting and analysis
Ask in plain language
Every figure cited to source
Suggested next steps with every answer
The AI agent team for travel and expense
An AI agent team, not a chatbot
Human approval gates on every plan
Consultant-grade deliverables, exportable
50 AI prompts for travel and expense
Ten categories, copy and run
Starter pack for your first week
Free, no gate
WHY WE BUILT THIS
I have watched travel and expense teams live inside the same contradiction for twenty years: more data than almost any function in the company, and less capacity to act on it. The problem was never the data. It was the speed at which decisions could be made from it.
And the real potential of AI is not in taking over the jobs people already do. It is in doing the work that is not being done: the work that was too expensive, that needed too many people, that never made it off the list. That is what we built the framework for. Not a faster dashboard. A workforce.
Keesup Choe · CEO, PredictX
Every term below describes something the Cogent AI Framework already does. They are defined here because the words matter as much as the capability when a team is deciding what to buy.
One brief is planned, split, executed and reassembled into a single answer.
Knowing where employees travelling on business are at a given moment.
An employer's obligation to protect employees from foreseeable harm.
Converts a plain-language question into a database query.
