Procurement agents that draft savings memos run on a two-system architecture. Coupa and Workday Strategic Sourcing hold the purchase orders, supplier master, and contract terms. Dock holds what the agent concluded after reading them: the consolidation candidate, the off-contract spend pattern, the price variance the agent flagged. A procurement lead reviews the memo in Dock and approves before any supplier conversation begins.
The architecture
Coupa and Workday Strategic Sourcing stay the system of record for the raw spend data: POs, invoices, supplier records, contract clauses, sourcing event history. Dock is the system of record for what the agent interprets from that data. Each Dock row carries a pointer back to the source record (coupa_po_id, wday_supplier_id), the agent identity that drafted the memo, the procurement reviewer, the decision, and the timestamp. The agent re-fetches platform data via fresh API reads when it needs current state, because POs close and contracts renew between the time a memo is drafted and the time it is approved. See Dock for procurement for the full pillar.
The Dock surface
A savings_opportunities table, one row per drafted memo.
| supplier | category | coupa_po_ids | annual_spend | opportunity | agent_draft | reviewer | status |
|---|---|---|---|---|---|---|---|
| Iron Mountain | records storage | PO-44120, PO-44910 | $312k | Consolidate 2 active contracts; one expires Q3 | agent:procure-01 | lead:r.okafor | approved |
| Veeva CRM seats | software | PO-51002 | $878k | 19% off-contract usage above tier; renegotiate at renewal | agent:procure-01 | lead:r.okafor | pending |
| LinkedIn Recruiter | talent tools | PO-48771, PO-48772 | $204k | Duplicate seat purchase across two BUs | agent:procure-01 | lead:r.okafor | needs-rework |
The workflow
The procurement agent pulls twelve months of PO line items from Coupa and matches them against active contracts in Workday Strategic Sourcing. It clusters spend by supplier and category, identifies the three patterns it has been instructed to surface (consolidation, off-contract drift, duplicate purchasing), and writes one row per opportunity to savings_opportunities. The memo body sits in the row as a doc field, with the underlying PO IDs linked back to Coupa.
The procurement lead opens Dock, sorts by annual_spend, and works the queue. Approval moves status to approved and triggers a downstream notification to category management. Rework sends it back with reviewer notes attached to the row. Nothing reaches a supplier without that human gate, which is the same pattern we use in Dock for legal and Dock for accounting.
Why this matters
Deloitte's 2025 Global CPO Survey found that 72% of CPOs name margin improvement via cost reduction as their top response to macroeconomic pressure, and that top performers allocate up to 24% of procurement budget to technology to deliver it. The bottleneck is rarely opportunity identification. It is the reviewer hours needed to validate each candidate before action.
The Hackett Group's 2023 Key Issues Study put a number on the same problem: procurement workload is rising 10.6% with little headcount growth, producing a 7.4% productivity gap. Agents close that gap only when the reviewer can audit the work without re-reading the source data. Dock makes the agent's reasoning, source pointers, and decision history a queryable artifact rather than a chat transcript. See agent audit and compliance for how that audit trail is structured, and agent identity for why every row carries an agent principal.
Finance teams running the same pattern for vendor consolidation work the queue from the same table shape. See Dock for finance.
Start with one supplier category
Point a procurement agent at one category in Coupa, give it the savings_opportunities schema, and review the first ten memos it drafts. The shape of the review queue tells you whether the agent's pattern detection is calibrated.
FAQ
Does the agent take action in Coupa or Workday Strategic Sourcing? No. It reads. The savings memo lives in Dock until a procurement lead approves it. Approval triggers downstream work, which a human or a separately scoped agent executes.
What happens if a PO closes between draft and approval? The agent re-fetches state from Coupa on approval. If the underlying PO has been closed or the contract renewed, the row is flagged for the reviewer rather than auto-approved.
Why not just run the analysis inside Coupa? Coupa stores the spend data. It does not store the agent's interpretation, the reviewer's decision, or the audit trail that links them. Putting interpretation in the platform mixes raw state with derived conclusions.
Can multiple agents write to the same opportunities table? Yes. Each row records the agent identity that drafted it. Reviewers can filter by agent, category, or confidence to triage.
Confirmed: file written to /tmp/phase-2b-ops/content/blog/procurement-spend-analysis.md
