Your operation loses money between the record and the floor. I find it and take it out.

The symptom shows in one department. The cause usually sits in another. I measure the whole chain before I cut.

Tell me what you’re seeing See a sample audit

Supply Chain & Operations × Deployment Strategy

How the work goes

  1. Measure

    Count what the floor holds against what the record says.

  2. Cut

    Take out the loss the count exposes, at its cause.

  3. Stitch

    Stay through go-live until the new way of working holds.

For technology teams, the same work runs in five steps, from the first audit to the handback. How a deployment runs

Why a system that works in the plan fails on the floor

Rarely the algorithm. Usually one of three things that were there before the system arrived.

  1. The record was already wrong. In one retailer’s audit, 65% of about 370,000 inventory records were inaccurate. A new system, or an AI agent, reads that record as the truth. Why records drift
  2. The incentive pays for the wrong thing. On one floor I walked, mixed, short and damaged orders led back to a productivity bonus paid per operator. The fix was the bonus, not the software.
  3. Nobody owns it after go-live. The vendor leaves, the project team dissolves, and the floor goes back to what worked. Who should own adoption and the 90-day check

The number people quote, 95%, comes from a 2025 MIT NANDA report, The GenAI Divide: 95% of organisations saw no measurable P&L impact from generative-AI pilots. It is a preliminary industry report, and it covers AI pilots, not ERP or WMS projects. The cause it names is older than AI: brittle workflows and missing context.

How I work

I walk the floor first, the 02:30 shift included, and ask what happened rather than what people think. Then I put the record and the floor on one timeline and write the working down, so finance, the floor and engineering can check it. Captures of screens, labels and racks run in plain sight with notice, never faces; an operator is recorded only if they say yes. How I diagnose

Engagements

A three-week field audit, cutover leadership, a fractional operations or deployment lead, and advisory by the hour. Engagements and fees

Proof

A 50,000-SKU distribution-centre cutover in Melbourne, a USD 30M cross-border P&L, a re-runnable triage of 2,708 marketplace claims, and tools I build in Python, SQL and TypeScript, among them an inventory and replenishment engine with 10,756 passing tests that stages every decision in draft before write-back.

See the proof or try a decision card

Working terms

Open to full-time or embedded Deployment Strategist roles with technology vendors and enterprise teams, and to fixed-scope audits and cutovers invoiced in USD from Chile. On a site visit I observe and count using the blade you issue; I do not operate your machinery or fill a post. Working terms or if you are hiring

Field notes

The Damage Has No Name

Field notes for the CFO: why the fastest driver on the report can be the most expensive one in the ledger, and how to trace the damage to a name.

PDF, 8 pages

Use It or Lose It

Field notes for the CEO: why a budget that expires on a fixed date pays for a rush in its last week, who carries the bill, and the rule you can change.

PDF, 8 pages

Who the Form Is For

Field notes for the COO: why a quality record built for the auditor teaches the floor to fill it rather than use it, and four questions to ask of it.

PDF, 9 pages

Short answers

Who should own adoption after an ERP or WMS go-live?

An executive on your side, named in writing before kickoff, who is not the deployer. Without one, the change after go-live belongs to nobody. Full answer

Why do deployers apply the same tool to every problem?

Because knowing a tool shapes what you see, even for experts. The safeguard is a rule set before kickoff, not the deployer’s open mind. Full answer

Should you customize the ERP or change the process to fit it?

Neither by default. Measure the misfit first, sort what the system lacks from what it imposes, and weigh each customization against its upkeep. Full answer

Should you replace your ERP with AI all at once?

No. The ERP keeps the books, the new piece works beside it in signed drafts, and an old module retires only after its replacement wins in parallel. Full answer

How do you check a deployment 90 days after go-live?

Log twenty system recommendations a month against what the operator would do and what doing nothing costs. If neither ever wins, audit the deployer. Full answer

Why do people follow the system even when it is wrong?

Because obeying costs them nothing today and questioning has a price. What reduces the error is making people answer for the decision. Full answer

Do manual overrides of a statistical forecast improve accuracy?

Sometimes: large adjustments tend to help and small ones often hurt. Without a name and a reason, a company cannot learn which overrides paid off. Full answer

Should the reason for an override be recorded before or after the change?

Before, and for a reviewer whose view the operator does not know. A reason written after the fact is a defence; one written to please is an echo. Full answer

Why don’t inventory records match the shelf, and what does that do to a model?

In one retailer’s audit, 65% of nearly 370,000 records did not match the shelf. A model trained on them is wrong exactly where the record is. Full answer

A customer says “not received” and the system says “delivered”: which record is wrong?

Often neither. In a public marketplace dataset, 57% of “not received” reviews on delivered orders were written on or before the delivery day. Full answer

About the author · The Deployment Chasm, a manuscript

Tell me what you’re seeing

Based in Santiago, Chile. I take on work in English and in USD in the United States, Australia, Canada, the United Kingdom and Europe.