How to put an AI agent on a real task
Each playbook takes one job — matching invoices, comparing supplier quotes, answering an RFP — and sets out the inputs to supply, the instructions to give the agent, what the output looks like, and what a reviewer should check before anyone relies on it.
What’s on your team’s list?
Find a task by keyword or explore a category.
Match invoices against purchase orders
Compare invoice lines with purchase orders and goods receipts. Prepare a cited exception report for accounts payable to review.
Invoices, POs and receipts
Line-level exception report
Compare supplier quotations side by side
Turn supplier quotations into a comparable table of price, lead time, scope and missing terms, with references to each source.
RFQ and supplier quotations
Comparable terms and open questions
Find contract renewals before notice is due
Extract renewal dates, notice periods and amendment references from agreements, then prepare a review queue for contract owners.
Agreements and amendments
Renewal and notice-date register
Draft RFP responses from approved knowledge
Map RFP requirements to approved company documents, prepare cited response drafts and flag answers that need a subject-matter expert.
RFP and approved company documents
Cited draft with a review queue
Answer employee policy questions with sources
Help employees find policy answers grounded in current handbooks, with the relevant source and a clear path to HR for exceptions.
Handbooks and an employee question
Policy answer with a source
Prepare security questionnaire responses
Draft security-review answers from approved policies and evidence. Flag missing proof, product-scope differences and claims requiring review.
Questionnaire, policies and evidence
Evidence-backed answer draft
Turn a support thread into a clear handoff
Summarize customer history, attempted fixes and unresolved questions into a source-backed escalation brief for the next support team.
Ticket history and approved knowledge
A concise escalation brief
Turn mixed documents into structured data
Extract defined fields from PDFs and scans into a reviewable table, retaining source references and flags for missing or conflicting data.
PDFs, scans and a field definition
Structured rows with source references
Review expense claims against your policy
Check submitted expense claims and their receipts against your travel and expense policy, and prepare a queue of claims that need a human decision.
Claims, receipts and the policy
Claims flagged against a rule
Reconcile a supplier statement against your open items
Compare a supplier statement of account with your own open items and produce an explained list of differences for accounts payable to work through.
Supplier statement and your ledger extract
Explained differences list
Gather the evidence behind month-end accruals
Assemble the supporting documents for each proposed accrual at period close, and show clearly which ones are not yet supported.
Accrual schedule and supporting documents
Accruals with evidence attached
Check a new supplier’s onboarding pack
Read the documents a new supplier submits, check the pack against your onboarding requirements and return a completeness report with the gaps named.
Supplier pack and your requirements
Completeness report with gaps
Triage incoming purchase requests
Classify each purchase request, check it against existing agreements and catalogue items, and return the requests that need a buyer rather than a rubber stamp.
Requests, agreements and catalogue
Routed requests with reasons
Review an incoming contract against your standard positions
Compare a counterparty’s draft against your standard clause positions and return the deviations, each quoted from the draft with your fallback beside it.
Draft contract and your clause positions
Deviations with your fallback
Triage inbound non-disclosure agreements
Read each inbound NDA, pull the terms your team decides on and route the ones that fall outside your standard so routine agreements stop queuing behind them.
Inbound NDAs and your standard terms
Intake register with routing
Build a register of what a contract commits you to
Pull the reporting, service and payment obligations out of a signed contract into a dated register with an owner against each one.
Signed contracts and your owner list
Dated obligation register
Prepare a renewal brief for an account
Pull contract terms, support history and approved account notes into one cited brief before a renewal conversation.
Contract, tickets and account notes
Cited renewal brief
Check a new joiner’s onboarding paperwork
Check the documents a new joiner submits against your onboarding requirements and return the missing and expiring items before the start date.
Submitted documents and your checklist
Outstanding items before day one
Find what a policy change affects
When a policy is updated, find the guidance, templates and answers that still describe the old rule, and list them for their owners.
Old and new policy, plus your guidance
Documents that contradict the change
Review a vendor’s security documentation
Read a vendor’s audit reports, test summaries and data terms against your own requirements, and return the gaps with the date of every piece of evidence.
Vendor evidence and your requirements
Gap list with evidence dates
Assemble evidence for a user access review
Pull together the account lists, approvals and leaver records a periodic access review needs, and show plainly which accounts have no approval behind them.
Account exports and approval records
Review pack with unexplained accounts
Draft a cited answer to a support ticket
Draft a reply to an incoming support ticket using only approved help content, with the source beside every claim and an honest gap where none exists.
Ticket and approved help content
Draft reply with sources
Classify and route inbound documents
Sort an incoming document stream into your own types, split files that hold several documents and route each one to the team that handles it.
Mixed inbound documents
Typed and routed queue
Review likely duplicate records
Surface likely duplicate records with the evidence for and against each pair, so a data owner can merge deliberately instead of trusting a similarity score.
Record extract and your merge rules
Candidate pairs with evidence
There’s more to explore.
Your knowledge. Your rules. Your infrastructure.
Make it work for your team.
Bring a task, a few representative documents and your deployment requirements. We’ll help map the configuration and the checks that make the result useful.
Explore sovereign deployment