Agent-powered procurement: from tender to award in hours
How an upstream operator replaced weeks of manual bid evaluation with a single agent workflow — and what it means for procurement teams.
A mid-cap upstream operator we work with used to run procurement tenders on a 4-to-6-week clock. An engineering lead would spend 40 hours reading bids, another 30 hours normalizing the commercial sheets, and 20 more in meetings reconciling the scorecards. Multiply that by 25 tenders a year and the team was spending over a full engineer of time just on bid evaluation.
They replaced that process with a Spawn workflow last quarter. Here's what it does now.
The workflow
- An inbound email trigger watches the tenders@ inbox. When a vendor submission lands, the workflow fires.
- A parsing agent extracts every line item, commercial condition, and technical spec from the PDFs and Excel attachments. Line items normalize against the operator's master catalog so "1-inch gate valve, flanged, CL300" maps to the same SKU across vendors.
- A compliance agent checks each bid against the tender's mandatory requirements — scope coverage, certifications, delivery terms, payment milestones. Non-compliant bids get flagged with a reason.
- A commercial agent produces a side-by-side cost breakdown normalized to the same currency, incoterms, and delivery window.
- A recommendation agent writes a draft award memo with a ranked vendor list, risks, and negotiation leverage points.
Results after 90 days
- Cycle time: 4–6 weeks → 3–8 hours from last submission to draft memo.
- Engineer hours per tender: ~90 → ~6 (review + sign-off only).
- Coverage: 100% of line items normalized — up from ~60% before, because no engineer was reading every addendum by hand.
- Savings identified: 4.2% average cost reduction on awarded contracts, attributed to the agent flagging pricing inconsistencies humans were missing.
“4 to 6 weeks became 3 to 8 hours. Engineers spend their time on judgment calls, not line-item normalization.
What the engineers do now
The interesting part isn't that agents replaced 84 hours of work — it's what the engineers do with that time back. They spend it on the decisions the agent flagged: negotiating the three contentious commercial terms instead of processing the other forty, walking vendor sites instead of re-keying price sheets, running sensitivity analyses on the top-ranked bids.
The agent doesn't replace judgment. It removes everything that isn't judgment.
What it took to ship
Two weeks of pilot work. One week was writing the prompts and tuning the catalog normalization; the other was integrating the email trigger and the operator's contract management system for the final award step. After that, the workflow has been running untouched for three months with a 97% clean-run rate (3% of tenders have a non-standard structure that needs human intervention — we're closing that gap next quarter).
If you run a procurement team and this sounds like your week, we'd like to build yours next.