From a sentence to a purchase order, without the catalog hunt.
Employees describe what they need in everyday language. The Intake Agent parses quantities, dates, and budget hints, matches contracted catalog items, runs policy checks (supplier risk, split orders, lead time), and previews the approval chain before anyone submits.
How it works
The same pipeline you see in the live module, step by step.
- 1
Parse the request
Extract units, needed-by dates, and budget phrases from free text — including informal quantities like “a pair of” or “around $25k”.
- 2
Match catalog and supplier
Score catalog keywords and names against each line need. Prefer contracted items and infer category and supplier from the best matches.
- 3
Run policy checks
Evaluate contracted pricing, preferred supplier tier, supplier risk score, budget overrun, split-order patterns, and lead time vs the requested date.
- 4
Build the approval chain
Add the requester’s manager, category manager for IT spend, Finance above $25,000, and the CPO above $250,000 — skipping redundant steps when limits already cover the total.
- 5
Submit and issue the PO
After approvers sign off in sequence, the agent creates a purchase order with supplier payment terms and links it back to the request.
Checks and rules
Deterministic logic runs on every action; optional AI enriches text when configured.
- Catalog keyword match with contracted price preference
- Blocked supplier → policy block
- Supplier risk score above 60 → warning
- Stated budget vs line total
- Split order: same requester + supplier within 14 days crossing threshold
- Lead time vs needed-by date
What you get
Artifacts land in the shared procurement record and activity log.
- Draft purchase request with editable lines and justification
- Policy findings (pass, warn, block) with reasons
- Approval chain preview with roles and limits
- Submitted PR with status and audit trail
- Purchase order issued when final approval lands
Where teams use Intake Agent
Questions
Does it need my ERP?
The live demo runs on a shared procurement record (requests, POs, receipts, invoices, contracts, and spend). In production you connect your ERP or AP system via API or scheduled file exchange so matching uses your real PO and goods-receipt data. Until then, every rule runs the same way on the demo dataset.
Does it depend on an AI model to work?
No. Every decision runs through Stellitron's deterministic rules first: tolerance math, playbook comparisons, weighted bid scoring and classification maps, so results are repeatable and auditable. Stellitron Intelligence adds document reading, drafting and plain-language explanations on top, and the agents keep working if it is unavailable.
Who approves?
Agents recommend and trace; people approve. Intake routes to managers, category owners, Finance, and the CPO by amount and category. Invoices on exception stay on hold until AP approves, requests a credit note, or rejects a duplicate. Contracts, sourcing awards, and supplier blocks follow your owners in the app — the agent surfaces the evidence, not the signature.
What if nothing in the catalog matches?
The agent still drafts a line with an estimated price and category guess, and adds a policy warning that sourcing will need to quote. You can edit lines before submit.
Can approvers reject or comment?
Yes. Each step supports approve or reject with an optional comment. Rejection stops the chain; approval advances to the next approver until the PO is issued.