Part II of an Applied AI Systems Series
I didn't start KOAH Chiropractic as a technology experiment.
I became a co-owner of a real healthcare practice — payroll, rent, insurance billing, marketing costs, and the quiet pressure that comes from knowing overhead doesn't wait.
There is something clarifying about a business where revenue and expense timing actually matter. Architecture debates disappear. Cash flow becomes the system constraint.
From the beginning, I approached it the way I approach enterprise platforms: not as a collection of activities, but as an operating system.
And this time, AI wasn't the product.
It was the multiplier.
The First Phase: Using AI to Reduce Unknowns
When you start a business, the number of questions is overwhelming.
Licensing. Entity structure. Tax treatment. Lease negotiation. Marketing positioning. Competitive density. Insurance payer mixes. Local demographic trends. Pricing strategy. Cash flow modeling. Vendor contracts.
Some of these are black and white. Yes or no. Required or not required.
Others are gray. Strategic. Contextual.
In the past, this meant relying heavily on lawyers, accountants, consultants, marketing agencies, and a network of people who had “been there before.” That network still matters. But AI fundamentally changed the dependency model.
I used large language models to perform early market analysis — evaluating local population density, competitor presence, keyword demand, payer mix realities, and positioning opportunities. Instead of hiring a marketing firm to generate a broad overview, I generated dozens of hypotheses myself and refined them interactively.
The cost was time and API tokens.
The return was speed and clarity.
AI didn't replace professional advice. It allowed me to ask better questions before engaging professionals.
That distinction matters.
Hundreds of Questions, Compressed
Starting a business generates hundreds of micro-decisions.
Should we structure distributions this way? Is this clause typical? What are the risks of this payer contract? What does AR aging above 90 days usually signal? What benchmarks exist for patient visit average? What marketing channels typically convert in healthcare?
Individually, these are small questions. Collectively, they create friction.
AI became a compression engine.
For black-and-white questions, it accelerated confirmation. For nuanced questions, it generated structured perspectives — trade-offs, risks, alternative strategies. It forced clearer thinking.
Instead of waiting days for responses or scheduling consults for every decision, I could model scenarios in minutes.
The volume of learning condensed into months.
Treating a Chiropractic Practice Like a Platform
Once operational, the surface problem appeared to be growth.
“How do we get more patients?”
But that's a marketing question. I reframed it as a systems question:
“How does this system produce predictable revenue?”
That shift changed the operating model.
We introduced structured P&L discipline. Patient acquisition cost was tied to marketing channels. Patient Visit Average (PVA) became a tracked metric. Growth thresholds were defined before owner distributions. AR aging reports were analyzed with intent, not just reviewed.
Marketing campaigns were segmented by intent — auto injury, sports, pediatric, lifestyle. Landing pages were aligned to keyword clusters. Conversion paths were instrumented.
And then AI was layered in — not to automate blindly, but to accelerate analysis.
Large language models were used to interpret Google Analytics trends, identify anomalies in campaign performance, suggest hypothesis-driven refinements, and evaluate messaging coherence across ads.
AI did not replace marketing judgment.
It compressed feedback loops.
Instead of hiring a marketing agency to interpret performance and suggest refinements, we built lightweight intelligence workflows internally. Structured data exports fed into models. Targeted prompts generated comparative analysis week-over-week. Summaries were formatted for decision-making.
The practice gained a feedback engine without enterprise overhead.
AI as Internal Intelligence, Not External Feature
There's a pattern here.
AI is often discussed as something you sell.
In KOAH, AI became something we used.
It helped evaluate financial forecasts under reimbursement lag. It supported cash flow modeling across growth scenarios. It surfaced risk when marketing spend outpaced receivables. It helped simulate break-even thresholds.
Payroll depends on these decisions.
That changes how seriously you treat instrumentation.
Healthcare businesses operate under reimbursement delay. Insurance claims age. Accounts receivable accumulate. Cash timing becomes as important as revenue totals.
Technology was applied to forecast volatility, model reserve thresholds, and align marketing spend to revenue timing.
This was not theoretical optimization.
It was operational survival.
Governance and Boundaries
Healthcare adds a constraint layer many startups never encounter.
HIPAA compliance. Controlled patient data access. Secure communication channels. Vendor risk evaluation.
AI integration could not bypass these boundaries.
Even when using models for analytics or internal insight generation, data exposure had to be considered deliberately. The architecture respected security constraints first.
AI does not eliminate governance.
It increases the importance of it.
The Failed Experiment: Autonomous Agents
At one point, I attempted something more ambitious.
I experimented with running Playwright automation driven by an isolated local LLM agent. The goal was semi-autonomous execution — letting an agent interact with web dashboards, gather data, and synthesize insights without manual export steps.
It didn't hold.
Local models were too slow. Tool orchestration was brittle. Error recovery wasn't robust enough for operational reliability.
The experiment was shelved.
But not abandoned.
Models are getting faster. Tool calling is improving. Cost efficiency is increasing. That path will be revisited — but only when reliability matches operational risk tolerance.
That's another pattern: applied AI requires timing discipline. Not every technically possible experiment is operationally appropriate.
What Changed
With AI as a constant analytical partner, the dependency model shifted.
We did not eliminate lawyers or accountants. We reduced unnecessary dependency. We did not eliminate professional judgment. We entered those conversations more informed.
The volume of small decisions no longer required outsourcing by default.
The chiropractic practice became more than a clinic.
It became a structured operating system:
- Financial discipline tied to patient economics
- Marketing treated as infrastructure, not sporadic campaigns
- AI applied to internal intelligence, not hype
- Governance embedded from inception
- Experiments attempted under controlled risk
The same architectural patterns used in enterprise platforms translated cleanly to a small healthcare business.
The scale changed.
The principles did not.
The Broader Pattern
Across Lalo Meals and KOAH Chiropractic, a consistent lesson emerged:
Technology only creates leverage when it integrates into a coherent system.
AI accelerates feedback.
Automation reduces friction.
Instrumentation clarifies decisions.
But none of these matter without discipline.
The model is not the advantage.
The operating system is.
