AX assessment and roadmap

Background

The client is a 3PL logistics company with 450 employees and four warehouses. Executives agreed they “had to do something with AI” but had been unable to decide where to start for six months. Vendor proposals each pointed somewhere different, and managers each spoke only of their own team’s pain. A painful warehouse management system replacement three years earlier had also left the company wary of large projects.

The brief was clear: “Show us what to do, in what order, and why, using our own data. Then run something, however small, to prove the judgment is right.”

What we did

The 12 weeks were split into four weeks of assessment, two of roadmap, and six of pilots. The point of the project was not to hand over a roadmap and leave, but to run two of its processes ourselves and verify the roadmap’s first numbers.

1. Site interviews and process inventory (weeks 1–3)

We interviewed 42 floor staff, managers, and executives across the four warehouses and head office, and shadowed work for half a day at a time. The 30 processes gathered went into one table: staff hours per month, error frequency and cost, related systems, and where the needed data lives and in what form.

2. Scoring and prioritization (weeks 3–4)

Each process was scored on expected impact (time, error cost, customer effect) and difficulty (data readiness, system integration, scale of change on the floor) and placed on a matrix. Nine high-impact, low-difficulty processes became the first wave; eight high-impact processes that need data work first became the second; and 13 were deferred with reasons (“human judgment is the core,” “only 3 hours a month,” “not until the system is replaced”). The deferred list is a decision too.

3. Roadmap (weeks 5–6)

A two-year roadmap by quarter. Each process has a business owner and IT lead, success metrics, prerequisite data work, and estimated cost. The sequence was adjusted twice in executive workshops before being confirmed.

4. Pilot 1, inbound inspection (weeks 7–12)

Inbound inspection, where workers compared packing lists with goods and recorded by hand, was replaced with automatic packing list reading matched to scanned quantities. Only discrepancies are shown to the worker. It ran for four weeks on two inbound lines at one warehouse.

5. Pilot 2, carrier invoice reconciliation (weeks 7–12)

Monthly invoices from 12 carriers are matched automatically to shipment records and contracted rates, and only mismatches go to the settlement staff, who check the cause and request adjustments from the carrier. Confirming the invoiced amount stays with a person. Verified on two months of invoices.

How it was carried out

Weeks	Work
1–3	42 site interviews, job shadowing, inventory of 30 processes
3–4	Impact and difficulty scoring, priority matrix, data readiness table. Manager review
5–6	Two-year roadmap. Two executive workshops to confirm the sequence, owners named per process
7–12	Two pilots built and run. Remaining roadmap benefits re-estimated from measured results
12	Final report. Roadmap, pilot results, three items to start next quarter, and data preparation tasks

Measured pilot results came in below the estimate for inbound inspection (estimated -75%, measured -68%) and above it for invoice reconciliation (estimated -80%, measured -93%). The benefits of the remaining seven processes were re-estimated with these differences, and the first-wave total was confirmed at 18,000 hours a year.

How executive decisions changed

The question “what first,” which had circled for six months, was settled in four weeks by a single table. Instead of comparing vendor proposals, executives decide by seeing where the company’s own 30 processes sit. The 13 deferred processes carry a written “why not now,” so the same debate does not repeat.

Above all, the roadmap’s first numbers rest on measured pilot results. The three items starting next quarter each have a named business owner and are reviewed against metrics every quarter. The company now runs on a system, not a project.

The table below compares each task before and after.