The Return of the Boring Business: Why Vertical Software for Plumbers Beats AI Wrappers

Here's a company that doesn't get invited to AI conferences. ServiceTitan sells scheduling, dispatch, and invoicing software to plumbers and HVAC technicians. Its customers are small business owners who drive trucks, not people who attend Y Combinator demo days.

ServiceTitan's fiscal year 2026 revenue: $951 million.

Not ARR. Revenue. From a company that most of the tech industry couldn't name.

Meanwhile, the AI wrapper landscape looks like a mass grave. The GitHub repository "awesome-ai-wrappers" peaked at 2,300 entries in mid-2025. By January 2026, roughly 40% of those links were dead.

Something is wrong with our pattern recognition.

The Structural Advantage of Boring

The trades — HVAC, plumbing, electrical, roofing, landscaping — represent a $2.1 trillion market in the United States alone. That number comes from SignalFire's December 2025 analysis of construction and home services spend.

This market has three properties that make it structurally superior to most AI startup addressable markets:

1. The customers can't build it themselves. A plumbing company owner with a fleet of 12 trucks is not going to spin up a custom scheduling system. They're not going to evaluate LLMs. They need software that works when they open it at 6 AM, and they'll pay $500–$2,000/month for it without flinching.

2. The switching costs are enormous. Once a field service company loads 3 years of customer records, job history, invoicing templates, and technician schedules into a platform, they're not moving. ServiceTitan's gross retention is north of 95%. Not because the product is irreplaceable, but because the data is.

3. The competition is local, not global. A new AI coding assistant competes with GitHub Copilot, Cursor, Claude Code, and every other global player on day one. A new HVAC scheduling tool competes with whatever the local distributor recommends. The go-to-market is trade shows, distributor partnerships, and referrals — channels that don't scale virally but also don't face global competition overnight.

The AI Wrapper Graveyard

Contrast this with the typical AI wrapper startup. The pitch: "We built a beautiful UI on top of [OpenAI/Anthropic/Google] for [specific use case]."

The problem: the commoditization clock starts ticking the moment you ship.

The three-body problem of wrappers

Body 1: The foundation model provider. OpenAI, Anthropic, and Google are all moving upstack. ChatGPT added custom GPTs, canvas mode, deep research, and operator. Claude added Projects, MCP, and artifacts. Every feature the wrapper offers is one product update away from being absorbed by the platform.

Body 2: Other wrappers. If your entire value proposition is "GPT-4 with a nicer interface for lawyers," the barrier to entry is a weekend hackathon. There are currently 47 AI-powered legal research tools on Product Hunt. Forty-seven.

Body 3: The customer's own team. As AI literacy increases in enterprises, internal teams build their own solutions. A Fortune 500 legal department doesn't need a startup's wrapper when their IT team can build the same thing with the API in two sprints.

The result: AI wrapper startups face margin compression from above (platform features), competition from the side (other wrappers), and disintermediation from below (customer self-build).

Revenue Per Employee: The Real Scorecard

The metric that exposes the difference between boring businesses and AI wrappers isn't revenue growth. It's revenue per employee.

ServiceTitan: $951M revenue, ~2,024 employees. Revenue per employee: ~$470K.

The median AI wrapper startup with $5M ARR employs 30–40 people. Revenue per employee: $125K–$167K.

Jobber, which sells field service management to small contractors: estimated $200M+ ARR with ~900 employees. Revenue per employee: ~$220K.

Housecall Pro, acquired by ServiceTitan in 2024 for reportedly $500M+: was running approximately $100M ARR with ~500 employees at the time.

The pattern: vertical software companies serving trades generate 2–3x the revenue per employee of horizontal AI startups, because their products solve operational problems that customers can't solve any other way.

Why "Boring" Means "Defensible"

The word "boring" in this context is a synonym for "defensible." Here's why:

Boring means domain expertise

Building scheduling software for HVAC companies requires understanding seasonal demand patterns, technician certification requirements, parts inventory management, warranty tracking, and local building code compliance. This domain knowledge takes years to accumulate and can't be replicated by a foundation model.

Boring means regulatory moats

Field service companies need software that handles contractor licensing verification, permit tracking, EPA compliance for refrigerant handling, OSHA reporting, and state-specific lien waiver requirements. Every regulation is a barrier to entry for competitors.

Boring means integration depth

ServiceTitan integrates with equipment manufacturers for warranty processing, parts distributors for inventory management, financing companies for customer payment plans, and insurance providers for claims processing. Each integration is a negotiated partnership that takes 6–12 months to establish. An AI wrapper has no equivalent integration depth.

Boring means data gravity

A field service company's 5-year history of job records, customer interactions, equipment service histories, and technician performance data creates genuine data gravity. This data makes the software more valuable over time — predictive maintenance recommendations, optimal technician routing, demand forecasting. The longer a customer uses the product, the harder it is to leave.

The AI Layer for Boring Businesses

The real opportunity isn't building AI wrappers that compete with boring businesses. It's adding an AI layer to boring businesses.

ServiceTitan is already doing this. Their AI features include:

• Smart dispatch: Matching technicians to jobs based on skills, location, and predicted job duration
• Revenue prediction: Forecasting which service calls will convert to equipment replacement sales
• Call analysis: Transcribing and analyzing customer calls to identify coaching opportunities for dispatch teams

These AI features are valuable precisely because they're embedded in a product with deep workflow integration and years of operational data. The AI isn't the product. The product is the product. The AI makes the product better.

This is the pattern that will define the next five years of SaaS: boring operational software, enhanced by AI, sold to industries that can't build it themselves.

The Valuation Disconnect

As of March 2026, ServiceTitan trades at roughly $73/share with a market cap of approximately $4.5 billion. That's about 4.7x forward revenue for a company growing 20%+ annually with 95%+ gross retention in a $2.1 trillion addressable market.

Compare this to a hypothetical AI wrapper startup at $20M ARR growing 100% annually with 80% gross retention in an addressable market that shrinks every time a foundation model ships a new feature. VCs valued this company at $200M last year (10x ARR) and are now struggling to find a lead for the next round at $150M.

The market is slowly recognizing what the trades have always known: the most valuable software solves problems that don't go away when the next model drops.

What This Means for Founders

If you're starting a company in 2026, here's the uncomfortable advice: consider the trades.

Not because they're exciting. Because they're a $2.1 trillion market served by incumbent software that mostly hasn't been updated since 2015. Because the customers pay reliably, churn rarely, and don't read Hacker News. Because the competitive dynamics favor deep domain expertise over raw technical speed. Because AI makes your product better without making it commoditizable.

The next ServiceTitan isn't going to be built by a team that spent three years at Google Brain. It's going to be built by someone who spent three years riding along in HVAC trucks and noticed that every company was doing dispatch on a whiteboard.

That founder probably isn't reading this article. They're too busy talking to customers.