10 Lessons Being Learned as we make a 9-Figure Fire and Security Company AI-Native

A fire and security protection / monitoring company is a great challenge for GenAI in the real world. Obligations are regulated, workflows are structured, and yet the human element is both the biggest strength and the biggest challenge. The money moves through a field platform, a CRM, and an accounting system that rarely agree. We are spending time inside one, a fire safety and monitoring company with more than 1,000 active service contracts, and the lessons below are what we would tell any owner, or any new AI hire, before they start down a similar road.

Five of these are about the team and where the money hides. Five are about the data, the ups and downs of the new AI hire, and the strategy that decide whether any of it works.

1. When the CEO is the board, everything hinges on one person

Owner / operator businesses often have the CEO as the board. Every decision routes back to them, and they are constantly pulled and pressured. The COO often cannot move without a second check. Good people wait for the bottleneck and spend time considering the dynamics of the owner instead of the fundamentals of the business. AI does not fix that. If one person is the only point of validation, an AI system just gives that person a streamlined way to see and approve or deny more. The real challenge is to spread the checking out, with clear owners, numbers everyone can see, and accountability that lets the CEO be more hands off while visibility improves.

2. Leaders who grew with the company can run the work without running the back office

Department heads in these businesses usually earned their seats by being good at the job as the company grew. That is not the same as running a back office. They can inspect, quote, and lead a crew. They were never trained to run reporting, reconciliation, or clean data entry. Do not mistake that gap for a lack of talent. Give them training, the tools they need, transparency, and the expectation that when the tools are available they will learn to use them. Most of them rise to the occasion.

3. Pay people for bookings and you will not get collections

If sales is paid on what they book, that is what you get. Booked work, not collected cash. We have watched loyal, hard-working reps leave money uncollected because nothing in their pay depended on it. Pay the booking bonus when the money is in, and accounts receivable starts to fall into line. The behavior follows the incentive, every time.

4. Transparency drives accountability. Loyalty drives patience and reluctance to leave

Owners of these companies often lean on loyalty. People have been there for years and they care. Loyalty is real, but it is not a control. It does not catch a pricing error or an unbilled assessment, nor does it drive the extra effort those things demand. When the numbers are visible and tied to the people responsible for them, accountability follows on its own.

5. Without shared numbers, nobody is wrong and nobody is accountable

If there is no shared measure, every department carries its own version of the truth and none of them are wrong. You do not have to publish net profit to fix this. You do have to measure what matters to each person who carries responsibility, and you need to show it to them. People manage what can be measured by their leadership.

6. AI runs on pristine data. A sloppy ingestion is money in and garbage out

Every promise of AI depends on two things: the quality of the data underneath it, and the cleanliness and stability of the code and models that power it. Contracts typed in by hand, scope changes lost between versions, pricing that never carried through. Point a model at data like that, let alone data that was never structured for this kind of use, and your system will confidently give you wrong answers. Cleaning and structuring the data is the foundation. It is a real investment to do properly, and if you do not get it right the first time you pay for it three times: once for the shoddy work, again in opportunity cost, and a third time to redo it. This is the kind of expensive, hidden problem we take on in the Innovation Lab.

7. Understand your new AI hire is a generalist, and should not be a specialist in ML or data science

The person a mid-sized to large company hires to own AI is ideally a generalist, closer to in-house counsel or an HR lead in terms of breadth and scope. They need to understand the business, the systems, and the vision well enough to bring in the experts and oversee them in the best interests of the business. They are not a Microsoft implementation specialist, they are not a machine learning or math genius, and they are not an expert in your specific niche, although that part can help. That is fine, as long as everyone is honest about the role and sets them up for success.

They can set the vision, align the team on the do's and don'ts, align everyone on risks and governance, and then decide to rebuild the CRM or ERP using Claude or OpenAI or whatever the leadership agrees is the most urgent and important thing of the moment. What they usually cannot do alone is the implementation, the oversight, the agent guardrail management, and the data work, or choosing the approaches that hold up and are truly in the best interests of the business. That is where they need help, from IT, HR, and Ops, or from a partner who has done it before.

It takes an MSP and strong Microsoft developers to get into Azure AI Foundry and Power Apps, measure ROI cleanly, and understand how and when to spin up an Azure instance because it is more secure and cost-effective than company Claude licenses.

8. The cheap way to ingest and the right way to ingest may not be the same (this is something the influencers don't talk about)

There is more than one way to move a large amount of data into a form a model can use, and they do not cost the same. Pick the wrong one and you can burn a budget and waste a ton of time before you have proven any value. The right call depends on how much data you hold, what kind of data it is and how it is stored, the regulations, how often you will query it, and more. Sometimes a dedicated hyperscaler is the alternative to the obvious "let's just use Claude", and other times local compute is. We wrote a separate study on the compute and cost choices underneath all of this, here. Read it before you start down that road.

9. An AI person is only as good as your IT, Ops, and HR oversight

Give one person AI tools and access to company data with no oversight and you have created a risk. There are real questions here about data retention, employee monitoring, and privacy law that are still being written. The AI role needs technical support from IT, KPIs and an understanding of the systems from Ops, and a clear line to HR, both to make sure all the "listening" they do with employee data is legal, and to make sure HR and the other departments understand the risks that come with third-party AI tools and wrappers. The AI person also has to own an honest return. If the economics of the role do not make sense for everyone, it will not last, and that person can do damage on the way out. Measure what matters with a sensible, agreed ROI formula and a tool or set of tools, so everyone is aligned on the value and return for the AI resource's efforts. Save all of their work. And make sure they never use their own personal LLM for anything to do with the business.

10. AI-native is old discipline with new tools

Centralized and structured communication and information is not new management theory. Clean data, aligned incentives, accountability through transparency. Those ideas are decades old. What changed is that the tools finally make the central organization of information interesting again to leaders who are wired to look forward, not down at operations. That is the opening. AI gives you a reason to fix the fundamentals, and it pays you back when you do.

We are moving out of the SaaS era, a time when we paid several database companies to use their apps, our data sat decentralized across all of them, and we overpaid for the complexity. Today GenAI can bring all of that data into one large dataset, and AI can use tools to connect and sync with those systems, so the CRM, the ERP, and the rest stay in sync behind a natural language interface that lets people ask questions with their voice or a simple window. Keeping that data clean and consistent is the key. Do this well and the rest of the syncing and the agents are much more straightforward.

Where to start

We learned all of this on a real mandate. If you want the specific story, including how a workshop turned into a custom app and a reconciliation that surfaced more than $300,000, read the case study.

Either way, start with the fundamentals. That is where the money is.