The Real Bottleneck Isn’t the AI
Welcome back to Founder Mode!
I keep hearing the same thing from teams rolling out AI.
“The model isn’t performing the way we expected.”
Most of the time, that’s not true.
The bottleneck is not intelligence. It’s alignment.
At Pretty Good AI, we have seen this pattern repeat across healthcare groups, service companies, and even tech-forward operators. Teams think implementation means turning on AI. In reality, implementation means cleaning up everything underneath it.
And that is where most pilots stall.
Turning On AI Is the Easy Part
When we deploy AI at PGA, the technical activation is usually the fastest step. Connecting APIs. Training on basic knowledge. Configuring routing logic.
What slows everything down is alignment.
Mapping provider schedules to physical locations.
Agreeing on what counts as a new patient.
Defining routing rules.
Cleaning CRM fields.
Standardizing appointment types.
No one budgets time for this work. Everyone assumes it already exists.
It almost never does.
The New Patient Definition Disaster
One team told us they wanted to prioritize new patients. Simple enough.
We asked what defines a new patient.
Silence.
One manager said it was someone who had never been seen before. Another said it was someone who had not visited in three years. Someone else said it depended on insurance billing rules.
The AI was configured using one definition. The billing team used another. The front desk had their own interpretation.
The result was confusion. The AI was technically correct based on its rules. The organization was misaligned.
Once we aligned on one clear definition and updated the CRM accordingly, everything stabilized.
The issue was not intelligence. It was an inconsistency.
The Three-Year Rule Routing Error
In another case, routing logic depended on the three-year rule for returning patients.
The problem was that half the historical data was incomplete. Some visits were logged in one system. Others were missing entirely.
The AI routed based on what it could see. Staff assumed it had perfect context.
It did not.
We had to go back and reconcile years of fragmented records before the routing started to feel right.
Again, not a model problem. A data alignment problem.
Location Mapping Chaos
Multi-location groups are common in healthcare. Providers rotate between clinics. Schedules change weekly. Tribal knowledge fills the gaps.
One group had doctors working across three clinics, but the scheduling system did not reflect that clearly. Humans knew where to place patients. The AI only knew what the database told it.
Patients were booked correctly by rule, but incorrectly by reality.
We paused the rollout and rebuilt location mapping from scratch. Only then did performance improve.
The AI was doing its job. The spreadsheet was not.
Alignment Beats Innovation
Founders want to ship features. Operators need to clean spreadsheets.
That tension is real.
As a builder, I love pushing the product forward. New capabilities. Better voice quality. Smarter routing. But I have learned that none of it matters if the foundation is messy.
The most important work in AI deployment is boring.
It is documentation.
It is naming conventions.
It is workflow mapping.
It is CRM cleanup.
Configuration is strategy.
If your rules are unclear, your automation will amplify confusion.
Why Most AI Pilots Die
When I look at failed pilots, they rarely collapse because the AI could not answer a question.
They die during setup.
Weeks spent arguing over definitions.
Conflicting process owners.
Fields that mean different things to different teams.
Hidden rules that no one ever wrote down.
Teams want magic. What they need is alignment.
The organizations that succeed with AI are not the most innovative. They are the most disciplined.
Start With the Spreadsheet
If you want AI to work, start with the spreadsheet.
Audit your CRM.
Define your appointment types.
Clarify routing logic.
Map providers to locations.
Document escalation paths.
Once those pieces are clean, AI feels powerful.
Without them, it feels broken.
At Pretty Good AI, we now assume that eighty percent of implementation is alignment. The model is often the smallest piece.
5 Key Takeaways
- The bottleneck is alignment, not intelligence.
- Most AI failures are workflow failures.
- Clear definitions matter more than advanced models.
- Configuration is strategy, not busywork.
- If AI feels broken, audit your spreadsheet first.
Final Thoughts
Building Pretty Good AI has humbled us.
I used to think the breakthrough would come from smarter prompts or better models. Now I know it comes from clarity.
AI reflects the system it is plugged into. If that system is messy, AI will surface the mess.
If that system is clean, AI will look brilliant.
So the next time something breaks, resist the urge to blame the model.
Ask a harder question.
Where are we misaligned?
Fix that first.
See you next week!
-kevin
Recent Social Posts
Recent Podcasts
2810 N Church St #87205, Wilmington, DE 19802
Unsubscribe · Preferences
