Deploying AI across a company, tied to a board inventory target.

The situation

Individual use of AI, with no company setup behind it.

Several employees were using Claude on personal subscriptions, mostly as an advanced search engine. There was no policy on what data could go into it, no connection to the company's systems and no shared work between colleagues. A board deck prepared with the tool had shown what it could do when used with intent, and management wanted that result across the company rather than in a few individual accounts.

At the same time, the board had set a goal to reduce inventory by 10%, with a significant share of the stock moving slowly. The planning team of three spent a large part of each week producing recurring reports by hand.

The company did not need another executive. It needed a defined project with a fixed scope and a capped fee, led by someone who has been accountable for the numbers, the systems and the teams, and who would leave the organization able to continue on its own.

What I did · foundations

Governance and systems first, training second.

The project started with a working session with IT, because training people on a tool that is not yet configured wastes their time and teaches habits that later have to be corrected.

• Mapped the systems and data the teams would connect to: ERP, ecommerce platform, point of sale, EDI and Microsoft 365, with data owners, available exports and connectors, and integration risks flagged.
• Drafted the AI policies for approval: acceptable use, data classification (what may and may not go into the tool), connector approval and access governance, and a review standard for work produced with AI.
• Moved the company from individual subscriptions to a company plan. The license cost roughly tripled against the initial assumption. The recommendation rested on control rather than price: single sign on through the company's Microsoft accounts, shared projects so that a report or a process is built once for a whole team, and policies that IT can enforce rather than leave to each employee.
• Measured proficiency before training with a 36 question survey across four categories, placed each person in a beginner or intermediate cohort, and replaced generic training with a curated video library for the fundamentals and 90 minute sessions built on the survey results.

What I did · inventory

A financial target as the first use case.

The first team to work with the new setup was inventory planning, on the board's target. This gave the project a measurable result in dollars and gave the rest of the company a working example built on its own data.

• Translated the 10% goal into dollars by brand, category, location and channel, with agreed measures of weeks of supply, turns and aging.
• Built a holding cost rate (cost of capital, storage and logistics, shrink, expiry and obsolescence, insurance), so that ordering decisions weigh the cost of holding stock against lost sales on real numbers.
• Segmented the slow moving stock by age, brand and product, with a disposition lever for each segment: returns to vendor, vendor funded promotions, bundles, loyalty offers, markdowns, off price channels or donation.
• Reset the ordering policy: reorder points and safety stock by ABC class, order frequency against minimum order quantities, and a review of open purchase orders.
• Rebuilt the recurring reports with written specifications and validation checks, run on a schedule, together with a vendor scorecard, a single action tracker and a weekly summary for the CEO and the board.

I led the first week, we co led the second, and the team led the last two with my review. The plan belongs to the people who will run it after I leave.

What I did · departments

Each department builds its own setups, with coaching.

Once the foundations were in place, each of six departments received a fixed allocation of time: a discovery session to list pain points and select the two or three use cases with the best return, two build sessions in which the team built them in a shared project, and follow up coaching to correct setup mistakes early.

Typical use cases included contract review against a playbook and lease abstracts in legal, monthly variance analysis with follow up questions drafted to the owner of each variance in accounting, driver based sales forecasting in FP&A, store labor scheduling around peak sales in retail operations, scheduled market and competitor monitoring in marketing, and first line support triage in IT.

Every department recorded the hours saved and the outside fees avoided in a savings register. The consolidated register and a prioritized backlog of use cases, with owners, formed the basis of the leadership readout at the end of the project.

How the project was run

A capped fee, measured against a savings register.

The fee was capped and billed only for hours worked. The case for it did not depend on optimistic assumptions: team productivity alone returns three times the fee if each employee frees about 19 minutes a day, before any inventory benefit or reduction in outside fees.

I used Claude myself, under my direction and review, to produce the project documents, models and training material. This kept the billed hours on the business, the numbers and the people, and it showed the teams the working method they were learning.

What changed

A company setup that continues after the project.

[To complete at week 8 with measured results: adoption rate across the 30 users; hours saved per week from the savings register; outside fees avoided; inventory reduction achieved or committed in dollars against the 10% target; number of use cases live by department.]

The company now works from approved policies, connected systems and shared projects rather than individual accounts, and each team knows how to build and maintain its own setups.