From the Local Government edition of November 10, 2026
Zwolle (Netherlands) trains 800 staff on generative AI, reports 90% weekly Copilot use among trained staff, and readies a procurement-coding AI agent pending council framework approval
Gemeente Zwolle (Hyperautomation team), with Microsoft partner Wortell · Municipal government · Zwolle, Overijssel, Netherlands
- Publisher
- Computable.nl
- Original publication
- 06 november 2026 - 07:15
- Source retrieved
- 2026-11-11
What happened
Computable reported on 6 November 2026, from a presentation by Zwolle's Hyperautomation product owner and a Wortell adoption consultant, that about 800 Zwolle staff (nearly half of officials) have been trained in generative AI at levels matched to their role, and that according to the municipality 90 percent of trained staff use Copilot weekly. Zwolle's first operational AI agent, 'Joop Inkoop', helps staff pick the correct CPV procurement code from over 12,000 categories, passes data to a Power App and has RPA create a purchase request in the financial systems; the municipality says the process is fully audited and the developers say it reduces errors. A second agent supports Microsoft and Copilot use. The college and council still have to adopt Zwolle's framework for AI agents; approval for internal-process agents is expected by year-end, and use in the social domain or civil affairs is still undecided.
Why it matters
Why now: published 6 November and not in the archive. It is a concrete municipal example of moving from RPA to agent-plus-RPA workflows in procurement, with council framework approval as the gate, relevant to U.S. cities weighing agentic AI in back-office processes (transferability limited by Dutch and EU rules).
Evidence and measured results
Trade-press account of a municipal and partner presentation; usage (90% weekly) and benefit claims are self-reported with no method or error-rate data.
Limitations and uncertainty
Conference presentation co-delivered by a Microsoft partner; claims are self-reported. Framework approval is pending. Error reduction is asserted, not quantified. The municipality's own documents were not opened.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-11; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Staff struggle to choose correct procurement codes among 12,000+ categories, causing errors and manual rework.
- Stakeholders
- Hyperautomation product owner
- Procurement team
- Finance
- College of mayor and aldermen
- Municipal council
- Discovery questions
- How are classification errors measured today?
- What does the pending agent framework require before go-live?
- Which processes are excluded until the framework covers them?
- How is weekly Copilot use measured?
- Value hypothesis
- Bounded: an agent that suggests CPV codes, with RPA creating requests, may reduce miscoding and rework; local measurement is required.
- Potential engagement
- Scoped agent pilot for procurement classification with an accuracy benchmark and an approval framework template.
- Unsupported claims
- Quantified error reduction
- Cost savings
- That 90% weekly use reflects productivity gains
- Readiness for social-domain use
Pre-sales engineering
Role takeaway
- Fit
- Municipalities on Microsoft 365 with existing RPA and structured procurement workflows.
- Architecture and integration
- Copilot-based agent for classification, Power App hand-off, RPA writes purchase requests into financial systems.
- Prerequisites
- Microsoft 365 and Copilot licensing
- RPA platform connected to finance systems
- CPV reference data
- Audit logging
- Constraints
- EU and Dutch public procurement rules
- EU AI Act obligations
- Framework approval pending
- Security
- Restrict agent permissions to classification; keep system writes in audited RPA with least-privilege service accounts.
- Proof of value
- Compare agent-suggested codes to expert decisions on a sample
- Verify audit trail completeness for agent-initiated requests
- Measure rework rate before and after
Delivery
Role takeaway
- Work
- Obtain framework approval, deploy the procurement agent for internal processes, and manage it through a documented lifecycle.
- Dependencies
- College and council framework adoption
- Hyperautomation team capacity
- Procurement process owner
- Ownership
- Gemeente Zwolle Hyperautomation team, with procurement as process owner (interpretation for the latter).
- Skills and adoption
- Role-based generative AI training for about 800 staff, delivered with Wortell.
- Governance checkpoints
- Framework adoption by college and council
- Documented ownership and prioritization before build
- Lifecycle review of each agent
- Acceptance criteria
- Agent code suggestions match expert choice at an agreed rate (proposal)
- Every agent-initiated request is traceable in the audit log (proposal)
- No deployment in social domain without explicit framework coverage (proposal)
- Risks
- Unmeasured error claims
- Vendor-influenced reporting
- Expansion into sensitive domains before rules exist
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Agent handles classification, hands off to a Power App and RPA writes into finance systems: a human-in-the-loop, Microsoft 365-centred design with deterministic automation doing the system write.
Governance
Who approves, reviews and stays accountable for outcomes?
Agent deployment awaits college and council adoption of a local framework; development is demand-driven with documented ownership, prioritization and lifecycle management.
Security and privacy
What data, permissions and controls need testing?
Not detailed; extension to the social domain or civil affairs, where personal data is heavier, is explicitly undecided.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Role-based training reached about 800 staff; 300 places filled within three days of kickoff.
Procurement
What should contracts, pricing and exit terms secure?
The first agent supports procurement classification under the EU CPV scheme; the presenters include a vendor partner.
Operating model
Which teams own the service once it runs?
A central Hyperautomation team combining RPA, Microsoft 365 and AI sets priorities, quality and coherence across applications.
Publication history
- 2026-11-10Local Government · Issue 653 resources
Stable resource ID: zwolle-nl-ai-agents-procurement-cpv-copilot-800-trained-20261106