DB Control: Evaluating AI for Manufacturing and Ops
Researching and identifying opportunities for AI integration across engineering, manufacturing, and operational workflows — inside a genuinely complex industrial environment.
Role
AI Analyst
Scope
Engineering · Manufacturing · Ops
CONTEXT
Everyone wanted "AI." Nobody knew where.
dB Control builds high-reliability hardware for the defense, aerospace, and medical industries. Leadership saw AI's potential but faced a messy reality: dozens of workflows, decades of tribal knowledge, and zero tolerance for error.
I joined as an analyst to answer a strategic question, not a UI one: where, across engineering, manufacturing, and operations, would AI genuinely help — and where would it just add risk?
This was a CX problem disguised as a tech one. The "users" were engineers, technicians, accountants, marketers, and salesmen whose trust is earned slowly and lost instantly.
PROBLEM
The shiny opportunities were the wrong ones.
The workflows that looked most "AI-ready" were often the ones where errors were most expensive. Real opportunity hid in the unglamorous, repetitive work.
Chasing the impressive use case is how AI projects die in industrial settings. I had to separate where AI is impressive from where AI is safe and useful.
High visibility, high risk. Tempting demos, catastrophic failure modes.
Low visibility, high value. Tedious lookups, knowledge buried in veterans' heads.
SCOPING
I went to the work, not the org chart.
I interviewed engineers, accountants, marketing, sales, and IT, traced workflows end to end, and listened for the friction people had stopped noticing because it was just "how things are."
The richest signal came from workarounds — the spreadsheets, sticky notes, and "ask Dave" rituals that quietly hold operations together. Those are where AI could remove drag without touching anything safety-critical.
THE MAP
Rank by value × safety, not by hype.
I built an opportunity map that scored each candidate on real value and on risk — giving leadership a defensible order of operations instead of a wish list.
Telling leadership where not to use AI was, paradoxically, what made them trust the recommendations to use it.
Quadrant the field. High value + low risk became the obvious first wave.
Name the no-go's. Explicitly flagging where AI shouldn't go built credibility.
Sequence it. A phased path that earns trust with early, safe wins.
The Policy Behind the Map
No workflow on that list could touch cloud AI — dB Control’s government contracts meant everything had to stay inside Microsoft’s existing, approved tenant. Before a single automation shipped, I wrote the company’s first AI usage policy: what Copilot could and couldn’t touch, and a lightweight approval process for any new AI tool.
Do
Use Copilot for drafting, summarizing, internal docs
Route anything touching contract data through approval first
Keep everything inside Microsoft’s approved tenant
Don’t
Paste contract data or PI into any consumer AI tool
Adopt a new AI tool before it’s approved
Let AI touch a safety-critical step unreviewed
With another intern, I mocked out the first automations to run inside those rules in Power Automate — pulled straight from the research’s safest, highest-value opportunities.
— operations stakeholder
OUTCOME
Not just a plan — a policy and a working prototype.
Instead of a vague mandate to “add AI,” the team left with three things: a written AI usage policy, a first wave of Power Automate automations built inside it, and a prioritized roadmap for what came next.
The policy and the automations doubled as proof, not just a promise: engineering, manufacturing, and operations could see exactly what “safe AI” looked like in practice. Strategy is just shared understanding made actionable.
REFLECTION
What I'd carry forward.
My first real lesson in zooming out: the bravest strategy is often saying "not here, not yet."
I'd love to have followed the first-wave opportunities into delivery and watched whether the trust-building sequence actually held on the floor. Research that never gets tested is just a hypothesis.
This project is why I lead with research now. The most expensive mistakes are made before anyone opens Figma.