From AI Prototype to Production: How Wrench Group Decides What Is Worth Scaling

‍About the guest: Tony Ferreira is Senior Director of Technology, Data, and Product at Wrench Group. He leads technology, data, product, and AI initiatives supporting a national portfolio of HVAC, plumbing, and electrical brands.

AI has dramatically shortened the distance between an idea and a working prototype. A team can now build in a few weeks what might once have taken most of a year.

That speed creates a new problem. When building becomes easy, companies produce more demos, internal tools, and proposed applications than they can responsibly maintain. A prototype can look convincing without proving that employees will use it, that it works inside the company’s real systems, or that it creates enough business value to justify its cost.

Tony Ferreira, Senior Director of Technology, Data, and Product at Wrench Group, has developed a staged approach for moving from experimentation to production. Teams have room to explore, but an idea has to earn its way into the operating business.

A promising concept can enter an isolated innovation lab built to resemble Wrench Group’s production environment. If it works there, the team may put it in front of three users and learn how they actually interact with it. The next step is a pilot at one operating brand. After roughly 90 days, Wrench Group should have enough evidence to expand the product, modify it, preserve only part of the idea, or stop investing in it.

The process allows the company to learn from real usage without treating every AI-generated prototype as a permanent product. One of the clearest examples is a live operating map Tony’s team built to connect activity across roughly 30 home-service brands.

The Map That Proved the Process

Wrench Group is a $2 billion-plus, private-equity-backed portfolio of HVAC, plumbing, and electrical businesses operating in markets across the United States. Its local companies maintain their own presidents, dispatchers, call centers, and operating teams. Tony’s organization provides shared technology, data infrastructure, and an AI operating model that help those businesses work more effectively.

That structure creates a difficult data problem. Wrench Group has customers, booked jobs, service histories, technicians, inventory, and trucks moving across many independently operated brands. The information existed, but it lived in separate systems and was difficult to turn into a unified operating view.

Tony’s team began with a simple prototype: place customer locations and service histories on a map. The first version showed where customers lived by ZIP code and identified people who had not received routine maintenance in a year or more. The team put it in front of internal users, asked whether they would use it and how it should work, and adjusted the interface based on their feedback.

As the idea proved useful, the team added more operational data. The map began showing booked work, operating capacity, technician routes, and the live locations of trucks across the portfolio. A prototype for viewing customer activity became a shared operating tool for several parts of the business.

The same underlying system now serves different roles. A dispatcher can see trucks and upcoming appointments. A technician can focus on the jobs assigned for the day. A marketer can identify customers who are overdue for service or live near an active route. A business leader can assess growth and market coverage by ZIP code.

One of the clearest workflows begins with a canceled appointment. A dispatcher can see where the technician is, identify nearby customers who are due for maintenance, and contact one of them with an immediate opening. Instead of leaving a truck and technician idle, Wrench Group can offer faster service to a customer who already has a relevant need.

That matters in home services because availability is often the deciding factor. When a pipe is leaking or an air conditioner fails during a heat wave, a customer is unlikely to wait a week for the company with the strongest brand. The provider that can arrive today often wins the work.

Better coordination can also reduce expensive repeat truck rolls. If the company understands the day’s jobs, likely equipment problems, and required inventory, it has a better chance of sending the right technician with the right part the first time. A single additional trip may be manageable, but repeat visits across hundreds of technicians quickly become a significant expense.

The map’s longer-term potential is more proactive. Wrench Group is making weather information available through its customer data platform. Future connections to equipment diagnostics and connected-device data could help the company identify likely problems earlier. A local team might know that severe weather is approaching, that a customer has not checked a generator or heating system, or that connected equipment is showing an abnormal reading.

The technician still performs the work in the customer’s home. AI helps coordinate the information needed to put that technician in the right place before a small problem becomes an emergency.

Tony said the initial map was built in a few weeks while the team continued doing its normal work. A similar project might previously have taken nine months and still left important data disconnected. The team then tested the idea with business leaders, refined the experience, and expanded it based on actual operating needs.

That progression is what made the map important. It demonstrated both the speed of AI-assisted building and the discipline needed to turn a quick prototype into something the business can use. Its success also opened the floodgates to new ideas from technicians, dispatchers, marketers, and business leaders across Wrench Group.

The Rest of the Discussion

• AI access should expand with demonstrated need. Wrench Group’s board wants broad AI adoption, but Tony has seen how quickly costs rise when everyone receives the most expensive tools by default. Employees can begin with basic access, explore a real task, and explain the limitation they have reached before moving to a more capable option.
• The observability layer is still incomplete. Wrench Group uses tools from Microsoft, Anthropic, and other providers. Their administrative views can show some usage, but Tony cannot get the complete cross-platform picture he wants without buying expensive enterprise packages or assembling several systems. Cost data alone does not reveal whether the work is valuable or whether someone is moving AI-generated code into an unapproved environment.
• Cheap software creation increases the importance of business judgment. AI makes it easy to build something technically impressive. Tony’s first question is what practical value it creates. A production system needs a clear user, a real need, and a measurable impact. Being interesting or visually impressive is not enough.
• Established workflows move to less expensive models. Tony often uses a highly capable model while defining a new task and learning what a good result looks like. Once the workflow is stable, he tests progressively less expensive models and continues moving down until the output no longer meets the standard. This lets the company reserve its most costly model usage for ambiguous work.
• Home-service businesses increasingly operate like technology companies. The service itself remains physical, but technology determines how technicians are scheduled, what information they receive, how quickly a company responds, and how consistently the customer promise is delivered. That operating layer can become a competitive differentiator.
• Wrench Group is becoming an internal AI incubator. After the map demonstrated that a useful application could move from idea to working product in weeks, employees began proposing more opportunities for technicians, dispatchers, customers, and business leaders. Tony’s team now has to prioritize which ideas deserve to enter the same testing process.
• Builders should begin with the outcome, not the novelty. Tony’s closing advice was straightforward: customers rarely care that a product is cool. They care that it helps them complete a task or makes their work easier. AI has made building faster, which makes a disciplined understanding of business value even more important.