How a Multi-Location Business Built an AI Operating System

Twelve sites, twelve slightly different operating models. Weekly reporting arrived in twelve different formats, customer response quality varied by manager, and institutional knowledge lived in individual heads. Leadership had visibility into revenue but no reliable read on operating consistency — and every new site launch reset the learning curve.

We reframed the problem from 'add AI' to 'build an operating system.' The engagement began with a two-week diagnostic mapping the ten workflows that materially drove customer experience and manager time. From that baseline, we designed a shared operating layer with three components: a governed knowledge base, AI-assisted site reports on a common template, and standardized customer-response drafting reviewed by local managers.

Rollout ran in three phases over 90 days. Phase 1 stood up the knowledge base and locked the reporting template with two lead sites. Phase 2 extended the customer-response layer and trained managers on review — not authoring — as their primary job. Phase 3 scaled to all 12 sites with a weekly executive rollup, a decision log for exceptions, and a lightweight governance cadence owned by the COO.

By week 12, leadership received a one-page executive rollup every Monday, customer response times were consistent across sites within a defined band, and manager reporting time dropped by roughly 25%. More importantly, the operating system survived turnover: two managers changed roles during the engagement and their sites continued running without a visible dip.

• Standardization is the prerequisite for AI leverage. Automating twelve different processes produces twelve different problems.
• Managers accept AI drafting far faster than AI decisioning. Keeping humans in the review seat preserved trust and shortened rollout.
• The single largest source of value was not time saved — it was consistency of customer experience across sites, which unlocked the ability to open a 13th location on the same playbook.

• Treat multi-location AI as an operating-system problem, not a tooling problem.
• Invest in the reporting template and knowledge base before layering AI on top.
• Design the review workflow first, then work backward into what the AI produces.
• Sequence rollout so early sites de-risk the model for later sites.