Framework · How to think about it
When AI actually helps and when it’s theater
We’ve worked through the cloud cycle, the mobile cycle, and the low-code cycle. Each promised to change everything; each changed some things. This framework is how we tell AI value from AI theater before the money is spent.
The core idea
Every technology cycle produces the same two failure modes: companies that ignore the real shift, and companies that burn budget performing it. The tools change; the job doesn’t. The job is still finding where friction actually lives in your operation.
AI is genuinely good at one broad thing: repetitive cognition reading, classifying, extracting, drafting at a volume and speed people can’t match. Where your bottleneck looks like that, AI helps enormously. Where it doesn’t, AI is decoration with an invoice.
The framework
Three honest verdicts
Verdict 1
AI helps here
A real bottleneck made of repetitive cognition, with a human review step available.
This is you when
- ✓Skilled people spend hours reading, sorting, or re-typing.
- ✓Volume is high and the pattern mostly repeats.
- ✓A human can review output where errors matter.
Verdict 2
AI helps later
The use case is real, but the foundations aren’t ready. Fix those first; they pay off either way.
This is you when
- ✓The data AI would need is scattered or untrustworthy.
- ✓The process itself is still unstable or undocumented.
- ✓Nobody can define what “working” would measure.
Verdict 3
This is theater
AI applied for the announcement, not the operation. The most expensive kind of pilot.
This is you when
- ✓The problem is coordination, not cognition systems that don’t talk.
- ✓Errors are intolerable and nobody can review the output.
- ✓The honest goal is to say “we’re doing AI.”
FIG. 01 Every AI pitch lands in one of three places: real friction removed, foundations first, or theater.
How to decide
Follow the friction, not the trend
List the three slowest, most repetitive steps in your operation. If one of them is essentially reading-and-deciding at volume, that’s an AI candidate worth a pilot with a measurable definition of success. If the slow steps are handoffs, re-entry, and systems that don’t talk the answer is integration and automation, and AI can wait. The discipline isn’t being pro-AI or anti-AI. It’s being pro-friction-removal, whatever removes it.
Common questions
AI in operations FAQ
How do we start without a big bet?
+
Pick one workflow step, define what “working” means in numbers, and pilot against it. A good first AI project measures in weeks and thousands, not years and millions.
What data do we actually need?
+
Less than the hype suggests, but it must be reachable and trustworthy for the specific step you’re automating. An honest data read is part of any first engagement.
Everyone else is announcing AI features. Aren’t we behind?
+
Most announcements are theater and theater burns budget that operators can spend on real friction. Moving second with a working use case beats moving first with a press release.
Have a place AI might fit?
Bring it to a strategy call. We’ll tell you honestly which verdict it gets help, later, or theater.
