AI that cites its sources. Humans that make the calls.
Ask spend, contract, and pickup questions in plain language. Every answer cites the underlying records. Every financial action requires authorized approval.
Source-cited answers, not guesses
Drafts disputes for human review
No autonomous financial action
Confidence and assumptions disclosed
What the AI Spend Assistant can do
Useful, narrow, and source-cited. The assistant is built around the F2 record model — not generic spend chatter.
Contract Q&A
Answer questions like “What is our compressor PM coverage at site X under the active contract?” with the contract version cited.
Invoice explanation
Explain why a specific invoice line matched (or failed to match) a contract rate card, RFQ, or work order.
Evidence summary
Summarize the source documents behind a missing-credit or over-quote opportunity for the AP reviewer.
Dispute drafts
Draft a professional, source-cited dispute letter for human review. The user edits and approves before anything is sent.
Site or vendor questions
Answer questions like “Which sites are missing used-oil credits this month?” with the underlying pickup records cited.
Opportunity proposal
Propose a savings opportunity with cited evidence — never auto-create one without a human reviewer.
What the AI Spend Assistant will not do.
The AI doctrine is explicit: AI suggests; humans approve. The assistant has no authority to act on its suggestions. It cannot move money, hold an invoice, send a dispute, modify a contract, or change a vendor record.
This is not a constraint we will quietly remove later. Source-cited answers and human approval gates are the trust model that makes AI usable inside finance and operations.
Cannot place an invoice hold
Cannot approve a credit memo
Cannot send a dispute to a vendor
Cannot modify a contract or rate card
Cannot post to AP or any external system
AI Spend Assistant questions
It answers contract, invoice, pickup, and vendor questions with cited source records; explains why a specific invoice matched a specific rate card; summarizes the evidence behind a savings opportunity; and drafts professional dispute language for human review.
It cannot place a hold, approve a credit, send a dispute, post to AP, modify a contract, or take any other financial or vendor-facing action. Humans approve every action.
Every AI response must cite the underlying record IDs — contract version, invoice line, work order, pickup, manifest, prior credit. AI citation compliance is tracked at 100% as a launch metric.
The assistant must disclose its confidence and assumptions. If extraction or matching is uncertain, the response says so — and routes the question to a human reviewer.
SpendGuard scope and savings disclaimer
F2 SpendGuard surfaces invoice exceptions, contract compliance signals, and savings opportunities with source-cited evidence. It does not approve invoices, post payments, send vendor disputes, or replace AP / accounting systems — every financial action requires authorized human review. SpendGuard does not promise a fixed savings percentage; savings are tracked across five distinct states (potential, accepted, validated, realized, sustained) and reported separately.
See AI with the brakes that finance and operations actually need.
A SpendGuard demo includes a live AI assistant query against a sample dataset, with source citations and human-approval gates.