AI Product Leader · Retail AI · US-remote
Chad Holdridge
Twenty-five years inside retail operations — now building the AI that runs them.
Enterprise retail-AI across six enterprise grocery and convenience retailers: a multi-tenant analytics platform, an LLM insight engine that never computes its own dollar figures, an evidence-audited test engine, and a shipped B2B SaaS — each built to be checked, including by me.
I have been the person a shelf system sends forty tasks to before the store opens, and I have been the person deploying that system into someone else's stores. I know which of the forty actually get done, and I know that the other thirty-eight are what teaches a store team to stop trusting the product.
That is why the work below is shaped the way it is. When a customer asked me to exclude weekends from a metric, I measured the distortion before changing anything: weekends were 25.8% of the denominator carrying 0.35% of the numerator, and exactly one metric was affected. The fix touched only that one.
The builds
All projects →Enterprise work shown at the level client agreements allow.
Multi-Tenant Retail Analytics Platform
One config-driven codebase I wrote and maintained, serving six enterprise grocery and convenience retailers across four countries in three currencies
AI Insight Engine: Deterministic Where It Matters
A production LLM pipeline for retail execution data where the model never computes a dollar figure
Evidence-Audited Test Engine for an AI Retail App
The AI authors and triages; deterministic code delivers every verdict
B2B CPQ Sales SaaS
Multi-tenant configure-price-quote SaaS for industrial distribution — my own venture, built end-to-end
How I work
The builds above were made inside a multi-agent system I run and govern: typed agents for discovery and review, retrieval that answers from my own projects, and hooks that enforce the rules at the tool layer. A hook refused the commit that shipped this page until its verification had actually run.
Below, the delivery gate refusing a spawn until the agent carries a delivery contract, then running it once the contract is attached.
$ agent spawn researcher-dashboards --type researcher
✗ AGENT DELIVERY CONTRACT MISSING — refusing to spawn.
$ agent spawn researcher-dashboards --contract file-first
✓ spawned · findings write to disk before the report
✓ delivered 19.6 KB · resolved: collected · 0 work lost
Depth on the practice: the orchestration ecosystem and the governance console.
Shelf-intelligence computer vision: at Focal Systems I ran the labeling operation behind the models and shipped Planogram Scan-in, the self-serve onboarding feature that fixed the cold-start problem every shelf-CV deployment hits.
The domain depth under all of it: grocery retail operations, the industry this work instruments. Full P&L ownership across 20 departments; led a store organization of 150+, around 600 including dotted-line.
What I bring to product
Conducted time & motion field studies across pilot stores, then built the automation that eliminated manual reporting entirely
Led a lost-sales and out-of-stock analytics methodology grounded in the published academic research of Gruen and Corsten rather than vendor benchmarks, so the numbers held up under executive scrutiny
A single metric dictionary as the source of truth: every KPI defined once, with a lockstep test so copies of a definition cannot drift
A 1.0-scoring fixture ships alongside a reversed-layout control that must score 0.125, and a CV system's run-to-run noise gets measured on identical inputs before the first reading counts
Turn what is actually breaking on a shelf into a specification an engineering team can build from, then stay with it through deployment into the stores it was written for
Root-caused a production outage to five distinct mechanisms and cut a query of more than 170 seconds to 0.64s