Case Study
CPG Manufacturing: Retiring a six-figure AI platform that hadn’t shipped anything.
A mid-market cup and lid manufacturer with nine plants and an expensive “AI platform” pilot that wasn’t doing much.
- Sector
- CPG manufacturing · $120M revenue
- Engagement
- Fractional CAIO · 8 months
- Status
- Completed, ongoing advisory
Situation
The COO had signed an 18-month contract with an “industrial AI platform” vendor the year before. $340k committed, $180k drawn. The vendor’s quarterly reports showed “dashboards stood up” and “models trained,” but no plant manager on any of the nine sites used them.
Renewal was four months out. The CEO asked whether to renew, expand, or walk. Internal opinion was split three ways.
What we did
- Spent two weeks on-site. Ate lunch with shift leads. Watched the production lines. Asked every plant manager what they’d actually pay $5k/month for.
- Ran a structured review of the vendor’s deliverables against the original SOW. Seven of the twelve stated outcomes were either unmeasured or not instrumented.
- Drew up a two-page “what we’d build instead” memo: a shift-handover summarizer used by 42 shift leads, and a scrap-cause classifier run against QC photos.
- Built the handover tool in ten days with one engineer from the internal IT team. Scoped the classifier to a 6-week build with a specialist partner.
- Negotiated the vendor contract down to a 90-day wind-down instead of the renewal.
Outcome
- Saved $220k in year-one software spend versus the renewal case.
- Handover tool in daily use across all 9 plants within six weeks of shipping. Shift-to-shift miscommunications down a measured 31% in the first quarter.
- Scrap-cause classifier cut unexplained QC rejects on the two largest SKUs by ~14%, holding through the next quarter.
- COO and CEO promoted an internal engineer into a permanent ML lead role instead of hiring an external VP. Retainer wound down as planned.
If this looks like the kind of problem you’re sitting on, an intro call is thirty minutes and costs nothing. We’ll tell you if we’re the right firm. If we’re not, we’ll often know who is.