An owner of a $25M industrial services firm thought he had no realistic path to working less. Twelve weeks later, AI was running the part of his job he hated most, and he got his Saturdays back. Here’s exactly what we built and how.
Composite client. Details changed; pattern is real.
The owner’s situation
He’s 58. Owns and runs a $25M industrial services firm in Texas. Third-generation. Wants to retire in 5 years and hand the business to his daughter, who’s currently the GM.
The problem he’d been ignoring: he was the bottleneck for 4–5 things in the business that nobody else could do. Customer escalations. Bid reviews above $50K. Vendor disputes. Insurance renewals. Sponsor-style reporting to the bank that financed his expansion two years ago.
Each one individually was small — 30 minutes to 2 hours. Collectively, they ate 15+ hours of his week and a chunk of his Saturday mornings. They were the work that kept him in the business when he wanted to be working on the business — or, increasingly, not in the business at all.
When his daughter pushed him to think about succession, he realized: she couldn’t take over because there were 5 jobs only he did. He had to either teach her each one (slow) or eliminate them.
He picked eliminate. He came to us.
What we built
Over 12 weeks, we built four workflows that absorbed roughly 12 of those 15 hours per week. The fifth (insurance renewals) we declared not worth automating — once a year, judgment-heavy, fine to do himself.
Workflow 1: Customer escalation triage (4 hours/week recovered)
The pattern: A customer would email, copying him, when they were unhappy. He’d read the thread, decide if it needed his personal touch or could be handled by an account manager, and either respond himself or hand it off. The decision took 15–30 minutes each, multiple times a day. Cumulative time: ~6 hours/week.
What we built: An AI triage layer that reads inbound emails copying him, classifies them on three dimensions (financial impact, relationship sensitivity, technical complexity), and drafts either (a) a hand-off note to the right account manager with context, or (b) a response from him for his review. He skims, approves, sends. The whole interaction takes 2 minutes vs. 20.
Time recovered: ~4 hours/week. Customer satisfaction held (measured by escalation-to-resolution time).
Workflow 2: Bid review triage (3 hours/week recovered)
The pattern: Any bid above $50K hit his desk. He’d read it, check the assumptions, sign off or redline. He was usually checking the same 6–7 things: margin assumption, scope clarity, customer risk profile, schedule realism, dependency assumptions.
What we built: An AI review pass that checks those 6–7 things automatically and produces a 1-page summary with red/yellow/green flags before the bid reaches him. He reviews the summary in 5 minutes vs. reading the whole bid in 30. If green, he signs immediately. If yellow or red, he reads the bid with focused attention on the flagged areas.
Time recovered: ~3 hours/week. Bid quality went up because the AI catches things he sometimes missed in a hurry.
Workflow 3: Vendor dispute drafts (2 hours/week recovered)
The pattern: Vendor invoices that didn’t match POs, scope creep on subcontracts, billing disputes. He’d write the dispute email himself because they had to be firm and specific. 30–45 minutes each. Maybe 4–5 per week.
What we built: An AI drafting assistant trained on his past dispute communications (firm, fact-based, relationship-preserving). He gives it the contract reference, the disputed line item, and a one-sentence summary of the issue. It drafts the email in his voice. He edits in 5 minutes vs. writing from scratch in 40.
Time recovered: ~2 hours/week.
Workflow 4: Bank/sponsor reporting (3 hours/week recovered, prorated)
The pattern: His bank wanted a quarterly update with operational metrics, financial trends, and forward-look commentary. He’d build it from scratch each quarter — usually 6–8 hours on a Saturday morning the week before it was due.
What we built: A reporting workflow that pulls the operational data automatically, surfaces the financial trends and material changes, and drafts the narrative in his voice (trained on his past 8 quarterly updates). He reviews and edits in 90 minutes instead of writing from scratch.
Time recovered: ~3 hours/week prorated across the quarter, and crucially, he got his Saturday mornings back.
What this cost
This was scoped as an AI Office Operator engagement at $5,000/month. Three months in, all four workflows were in production. Six months in, the workflows had absorbed 12 hours of his week and the senior accounting clerk’s role had visibly changed — less administrative work, more analytical support to the team.
Total cost over 12 months: $60K (12 × $5,000).
The ROI isn’t measured in dollars saved — that’s not the point. It’s measured in what he did with the recovered time. He spent the first quarter training his daughter on bid review, a job that previously lived “in his head” and is now codified in the AI workflow. He spent the second quarter starting succession conversations with his three other senior people. He took a 10-day fishing trip with his son in October. None of that happens if he’s still doing 5 jobs nobody else can do.
Why this pattern works for owner-led businesses
We see a version of this with most owner-led services businesses we work with. The owner is the bottleneck on 3–6 things that have accumulated over 20 years. Most of them are not the things an outside consultant would identify as “AI opportunities” — they’re the small, persistent, judgment-laden tasks the owner has been doing because nobody else can.
AI is unusually well-suited to these tasks because:
- The owner has decades of past examples in their email and files — that’s training data.
- The owner can review AI output quickly because they’re the expert — they can verify in seconds.
- The tasks aren’t volume-heavy enough to justify hiring — but they’re time-heavy enough to matter.
- The tasks block succession — automating them is automating the owner’s irreplaceability.
This isn’t AI that changes the whole business. It’s AI that uses what the owner already knows to take work off their plate. The kind that quietly gives an owner their evenings back.
What this is not
It’s not “AI runs the business.” It’s “AI runs the four parts of the owner’s job the owner doesn’t need to be the one running.” The owner is still the owner. The customer relationships, the strategic calls, the people decisions — those stay human.
It’s also not a fit for every owner. Owners who actually love the bid review work shouldn’t automate it. The right workflows to automate are the ones the owner hates or the ones blocking succession — not the ones they’re good at and enjoy.
Want to run this for your situation?
If you can name 3–5 things only you do that you’d love to hand off — or that someone has to do before you can retire or scale — that’s the conversation to have. Most of those are AI Office candidates.
The first conversation is free. 30 minutes. We tell you straight whether automating the workflows you’re thinking about is realistic and what the first 90 days would look like.