Four initiatives.
None measured
anything.
A national logistics and distribution operator had four active AI initiatives, three software vendors, and no single metric that told them whether any of it was working. AI investment was being driven by vendor relationships, not operational problems.
Reduction in manual exceptions
Across national distribution network
Agentic workflows in production
Scoped, built, and handed over in 16 weeks
Duplicate AI initiatives stopped
Three vendor-originated, one internal
Full engagement duration
Audit through live handover
Investment without measurement is spend without direction.
The operator's AI portfolio had grown initiative by initiative, each sponsored by a different function. Procurement owned one. Operations owned two. The innovation team had sponsored a fourth directly with a vendor it had met at a conference.
None of the four had a defined baseline. None had agreed on what success looked like in operational terms. Two were solving the same problem with different vendors. Neither vendor knew the other existed.
Vendor-shaped priorities. No operational grounding.
When we asked what problem each initiative was solving, every internal champion gave a different answer. When we asked what metric would tell them it was working, none could answer without consulting the vendor.
AI strategy had been outsourced to the people selling the tools. The operator was paying for vendor roadmaps, not operational outcomes.
Audit. Measure. Build. Hand over.
Initiative Audit
Catalogued all four AI initiatives across the national network. Found significant overlap: two vendor-sourced initiatives were solving the same routing optimisation problem with different tools, neither aware of the other. Internal measurement was absent. Success was defined by vendor reports.
Measurement Framework Design
Built a measurement framework from scratch. Established baseline figures for manual exception rate, processing time per exception, and escalation volume. Defined what 'working' would look like in operational terms. Not model accuracy, but exceptions handled without human intervention.
Agentic Workflow Design and Build
Scoped two agentic workflows against the measurement framework: an exception triage agent and a carrier communication agent. Built both inside the existing tech stack. No new platform, no new vendor. The exception triage agent went live in week 10, producing data from day one.
Handover and Portfolio Closure
Transferred ownership of both workflows to the operations team. Ran four weeks of side-by-side operation, then exited. Closed the three duplicate initiatives. The fourth was folded into the new framework with revised objectives and a named internal owner.
"We thought we needed a better AI platform. We actually needed to know what problem we were solving. GBW gave us that before we spent another dollar."
Two workflows live. 44% fewer manual exceptions.
The exception triage agent went live in week 10. By week 16, manual exceptions had dropped 44% across the distribution network. The carrier communication agent was live at handover, already handling routine carrier interactions without human intervention.
Three duplicate initiatives were closed. The fourth was reformed with new objectives and internal ownership. Total vendor spend on AI dropped by 60%. And the two remaining workflows were producing more measurable output than the four had combined.
The operations team runs both workflows independently. No external support contract. Measurement is embedded. The team knows exactly what working looks like.
Paying for vendor roadmaps instead of operational results?
We audit what you have, measure what matters, and build what works.