Never Trust the Sales Call: How IRBIS Stress-Tests Every AI Tool Before It Touches San Jose

This is the sixth installment of our Home Services Leaders in AI series. We sit down with operators who are actually using AI and automation to run their companies — no theory, no hype. (Catch up on the first interview with Sarah at Golden Rule Cleaning, the second with Dave at Glass Therapy Clarksville, the third with Adam at Blanton & Sons, the fourth with Bekka at Advanced Air Conditioning + Heating, and the fifth with Kara at Air Repair Pros.)

IRBIS is a home-services company in San Jose, California that does HVAC, plumbing, and electrical. Tim Alagushov founded it in 2019 and runs it today with about 100 people, 50 of them technicians. He's 32. He personally designs the company's sales presentations, works alongside his team every day, and has assembled one of the deepest AI stacks of anyone we've talked to in this series. He is also, by policy, the hardest AI buyer to sell. "In our company, we have a policy: never trust any software in the beginning."

Don't Decide on the Demo

Tim's starting assumption about any new tool is that a polished demo and daily use are two different things. "On a sales call, everything is beautiful," he says. "But when you actually start working with the software, you're going to see a lot of issues there." So IRBIS never decides on the call. Before any tool gets adopted, the company writes a technical assignment for it, a document spelling out exactly what they need and what they expect the software to do. They build visual boards. Then they assign an integration specialist and the relevant department manager to stand the system up together, test it, and live in it for roughly a month.

Only after that month does IRBIS decide whether to keep the tool. And Tim negotiates the trial terms so the month doesn't cost him if the answer is no: a one-month free trial, or a discounted first month, plus one to two months of opt-out if it turns out not to be a fit. The result is a buying process that filters out what most operators get burned by, which is software that demos beautifully and behaves differently in production.

This isn't AI-specific. Tim points to a non-AI rewards dashboard IRBIS once tried and dropped after the test month as proof the policy earns its keep across every category. The lesson he draws is simple: make the software prove itself in your shop, on your processes, for a month before you rely on it.

Why a 50-Tech Shop Says Yes to So Many Tools

The flip side of Tim's skepticism is an unusual openness to spend. A lot of owners we talk to refuse to put AI anywhere near their company. Tim's math runs the other way, and it starts with ticket size. "In our industry, a ticket is huge. One client can bring you $60,000. So that's why we are not greedy about additional tools that can bring more revenue. Eventually, when you use them properly, it's going to pay off."

The operative phrase is "when you use them properly." Tim thinks the reason most shops refuse AI isn't cost. It's that they have no one who can implement and manage it. "Most businesses don't have operational staff who can implement them, manage them, in order to get return. Maybe that's why they refuse to add all these things." IRBIS does have that staff: testers, an integration specialist, department managers. Tim is 32, his co-founder is a technician, and he frames being a young company as a structural advantage. "We all grew up around technology. When you're younger, you can be faster. Other companies have way older people in management who are not that tech savvy." High ticket plus an integration team is what makes a long tool list rational instead of reckless.

The Stack: Hatch, Rilla, Lace, Talent on Tap, Ply, Claude

IRBIS's system of record is ServiceTitan, and around it sits a layer of specialized AI, each tool aimed at one job.

• Hatch: the front-of-house workhorse. It answers the phone, handles AI text and voice on both inbound and outbound, and runs as IRBIS's lead-management platform with automation, notifications, and follow-ups. It consolidated a patchwork of separate tools into one system.
• Rilla: listens to how technicians talk with customers in the field and coaches their performance, the virtual-ride-along pattern that has become standard for conversation intelligence in the trades.
• Lace AI: the CSR-side equivalent. It listens to how customer service reps talk, scores them, and runs the reports IRBIS uses to improve the front desk.
• Talent on Tap: hiring. It conducts first-round interviews with candidates, reviews resumes, and hands IRBIS highlights so a human only spends time on the candidates actually worth watching. Tim uses it for CSRs and field roles alike.
• Ply: inventory management. It wasn't originally an AI product and is now adding AI features. Tim works closely with their team on improvements.
• Claude: the office brain. Marketing, IT, managers, and operations, anyone working with large data, has a Claude account for processing and mechanical work. Technicians separately use their own ChatGPT for diagnostics and everything else.

Different shape than the five named UpSmith agents Bekka runs at Advanced Air Conditioning + Heating, where one vendor's roster covers the whole front desk. IRBIS instead buys the best point solution for each function and makes them talk to ServiceTitan, which is exactly why integration discipline matters so much to Tim.

What He Won't Buy, and Why

What Tim says no to is as informative as what he buys. He passes on point tools that duplicate something his core platforms already do well, and he's wary of products stitched together from other people's parts rather than built for the job. The clearest example is a pitch for an AI tool that did technician follow-ups and nothing else. He turned it down flat. "Getting AI for every little thing in your company? I have Hatch that has everything." If a platform he already trusts covers the task, a single-purpose add-on has to clear a very high bar.

Out of those decisions Tim has built a four-part screen he runs before a tool ever reaches the month-long test. First, does it actually solve a problem we have, or are we about to buy AI to paper over an internal issue we should fix ourselves? ("You don't really need AI follow-ups with good technicians. If a technician is not doing his follow-ups, he's not a good technician.") Second, is the platform built from the ground up to solve this problem, and how cleanly will it connect to the rest of the ecosystem? Tim strongly prefers a real API integration over brittle workarounds. Third, who is the person behind the product? "There are a lot of young students trying to solve a problem they never actually had personally." He wants someone who has been in the industry and knows the little issues around the problem. Fourth, trial and opt-out terms, negotiated before onboarding.

Never trust any software in the beginning. Whatever they say, whatever they call, never trust. Always go through tests and check to ensure it works.

— Tim Alagushov, IRBIS

What He'd Want From an AI-First CRM

Tim's bigger-picture point isn't really about any one platform's features. It's about architecture. Layering modern AI onto a system that was designed years ago is genuinely hard, he argues, because the older the underlying data model, the harder it is for new features to get clean access to the data they need. For a 50-tech shop doing tens of millions in revenue, he'd love to see a CRM built AI-first for the trades, and he doesn't think a fully satisfying one exists yet.

So Tim describes the system he'd build, and it starts from the physical world rather than the database. "There is a house. There are equipment. Whenever you get any information, that's where the system should start building the whole environment. Then there's a technician who'll go there, whatever car they use, whatever tools they have. Everything should start optimizing between them." The core requirement is a flexible data model where all the columns and data can talk to each other. He says he knows of at least one startup already building an AI-first CRM for the trades. They invited him onto the board and he declined because IRBIS is its own full-time project. "Maybe in a couple of years we're going to see some pretty strong players."

Tim's Framework

1. Don't decide on the demo. Everything looks good on a sales call; the real issues only show up once you're working in the software every day.
2. Make every tool survive a month in your shop. Write a technical assignment defining what you need, build visual boards, and assign an integration specialist plus a department manager to test it before you commit.
3. Negotiate the exit before you onboard. A free or discounted first month plus one to two months of opt-out means a failed evaluation costs time, not money.
4. Check whether the problem is actually internal first. Sometimes you don't need AI, you need to fix a process or a bad hire. And don't buy a separate tool for every little task when one platform already covers it.
5. Favor platforms built for the job with clean, native integrations. A tool purpose-built for your problem that connects properly will hold up far better than one stitched together from other people's parts.
6. Vet the person behind the product. Favor people who've lived the problem in the industry over those solving a problem they've never personally had.

The Bottom Line

IRBIS isn't an AI success story because Tim bought a lot of tools. It's a success story because almost none of them get in without earning it. He runs Hatch, Rilla, Lace, Talent on Tap, Ply, and Claude not out of enthusiasm for AI in the abstract, but because each one cleared a written technical assignment and a month of real-world testing by people whose job is to find the flaws. The high average ticket pays for the experiments, and the integration team makes the experiments safe.

Tim's bet is that the next real wins in home services won't come from a flashier chatbot. They'll come from rebuilding the unglamorous workflows on a data model that can actually talk to itself. Until then, the safest thing you can do with any AI tool, in Tim's world, is hold off on trusting it until it has survived a month in the field.

Ready to Automate Your Home Services Business?

If IRBIS's disciplined approach to AI sounds like how you'd want to run your own shop, that's where Robby comes in. We help HVAC, plumbing, and electrical companies wire up AI-powered estimate generation, automated documentation, and integrations across ServiceTitan, HouseCall Pro, and the rest of the stack, with scripts to organize the price book so the options sync cleanly. We've ridden along with operators across the country and seen what works at 50 techs and what works at three.

Book a free demo and we'll walk through your specific workflow and show you where the biggest wins are.

One more thing worth a look if you're an operator: Tim recently launched his own home-services marketing course, Snow Leopard Marketing, built on the same playbook he used to grow IRBIS in San Jose. If the operator's-eye view on growth and tooling that runs through this interview resonates, that's where he teaches it in depth.