A Contractor's Performance Is Becoming Public. Who Decides What 'Good' Means?

A contractor can be very good at showing up on time and less impressive on the quality of the work. Another can have the opposite strengths. Give both companies one overall score, and you can make that difference nearly disappear.

This isn't a thought experiment. I was looking through Homh, the home-services marketplace powered by ServiceTitan, and found something I haven't seen presented quite this way in local business discovery: individual contractors with public scores for on-time arrival, being hired for the job, and job quality and satisfaction.

We've mostly compared service businesses using star ratings, reviews, and marketing claims. Now a company that runs contractors' operating software is publishing measures of how the work gets done. The question is how those measures should inform recommendations, particularly recommendations made by AI.

Two contractors, almost the same overall score

On October 20, I reviewed two public Homh profiles serving the Los Angeles area. Monkey Wrench showed a 7.9/10 job performance rating, including 8.6/10 for on-time arrival and 7.4/10 for job quality and satisfaction. The Right Choice showed an overall 7.8/10, with 7.4/10 for on-time arrival and 8.4/10 for job quality and satisfaction.

These are marketplace-published scores, not measurements I independently reconstructed. The profiles don't explain the formula or the component scores' reference periods. The companies differ in size and service mix, so this is not a controlled comparison of quality. I haven't established whether any consumer AI agent uses these scores.

Still, look at the problem they reveal. The overall scores differ by a tenth of a point, while the component ratings pull in different directions. If I need someone to repair an air conditioner today, arrival reliability could be central. If I'm choosing a company for a complicated installation, I'm likely to care about demonstrated experience with that exact type of project and how the finished work holds up. Neither the overall score nor either component can decide that for me.

ServiceTitan's October 6 announcement describes Homh as a selected contractor network with operating-performance signals, availability, and confirmed booking. Its contractor page names punctuality, hiring frequency, and customer satisfaction among its criteria. It's a more ambitious approach than aggregating reviews. It also creates a harder problem: who defines and combines those measures?

Every performance score contains a judgment

Take "hired for the job." A higher score might indicate that a contractor explains the work clearly, offers competitive prices, or is good at converting estimates. It might also reflect the kinds of jobs the business accepts or the customers it serves. Without the metric's definition, denominator, and context, we don't know precisely what behavior it measures, much less whether a higher number benefits a particular buyer.

Homh also displays jobs completed during the prior 30 days, useful evidence of scale. But a company handling hundreds of small repairs and a specialist doing fewer complex installations may not compete for the same work. Job count measures activity; it doesn't decide which company fits a particular problem.

The distinction gets sharper when a system has to choose. If two contractors have different strengths, which quality gets the most weight? The marketplace's? The homeowner's? Or an AI assistant's guess about what the homeowner probably wants? An overall score is convenient because it gives a complicated decision a single number. Convenience is also how important differences get lost.

The right measure changes with the job

Consider the kinds of decisions people make across the service economy. A homeowner choosing emergency plumbing help may care most about response time, availability, and whether the company can handle the specific failure. That same homeowner planning an expensive renovation may place more weight on comparable projects, permits, warranty terms, and which crew will do the work.

For a dental patient, a practice's appointment volume may establish experience and capacity. It doesn't establish experience with a particular procedure, whether the treating clinician has relevant training, or whether the patient's insurance is accepted. A law firm's number of closed matters may indicate scale without telling a prospective client whether the attorney assigned to the case has handled similar disputes. An auto repair shop's quick turnaround is valuable, but a driver facing a recurring electrical fault might care more about diagnostic capability and whether the repair lasts.

Some industries have already spent years confronting the problem of turning performance into a consumer score. The federal Centers for Medicare & Medicaid Services' nursing-home rating system separates health inspections, staffing, and quality measures instead of relying exclusively on one undifferentiated number. Even there, the published ratings are decision aids. A family still needs to establish whether a particular facility can meet someone's clinical, personal, and geographic needs.

These examples are illustrative, not observed consumer prompts. Appropriate weighting depends on the assignment. There is no universal definition of a good service business.

The software company is becoming more than a software company

Here's the part I find especially interesting. Operational platforms historically collected information to help businesses schedule jobs, bill customers, manage staff, and measure performance internally. Homh shows a possible next use: publishing some of those measures as part of a consumer-facing selection experience.

That changes the competitive landscape. A contractor inside one software ecosystem may have detailed performance dimensions made public while an equally capable company using another system has only reviews and a website available for comparison. Missing measurements should not be mistaken for poor performance. Nor should the availability of a proprietary score automatically make it the correct one for a buyer's needs.

There's a governance question when a company collecting operational data also operates a marketplace for selected customers: how transparent are the selection and scoring criteria? What if a contractor questions its score or changes software? Does it cover the whole company, a branch, or only recorded jobs? These questions don't imply Homh manipulates its scores. Nor is there evidence here that AI assistants favor businesses with them.

An October 20 Yelp announcement adds another dimension: scheduling connections to ServiceTitan and Housecall Pro, alongside lead-management connections to Jobber and Workiz. It doesn't mean Yelp is adopting Homh's scoring methodology. It does show that the boundary between marketplaces and the software running service businesses is becoming more porous. Information about a business and the ability to act on it are increasingly moving through connected systems.

What I would want an AI assistant to do

Suppose a customer asks an agent to find an HVAC company for a complicated heat-pump conversion in an older house. The answer should establish whether the company performs that work locally, and distinguish general job volume from comparable experience. If the agent uses a performance score, it should explain the metric and its source. If relevant experience isn't documented, an overall rating can't substitute for it.

For a routine repair, the emphasis might change. Same-day availability, service geography, dispatch fees, and reliable arrival can legitimately matter more. The important feature isn't that one metric beats all the others. It's that the recommendation fits the decision being made, and the reasons remain inspectable.

This is a useful challenge for us at TrueSignal, too. We work on turning a business's operating history and declared capabilities into structured facts AI can find and use. But a sourced metric is only a starting point. Definitions, time periods, job categories, coverage, and limits determine whether another system can interpret it responsibly. And even a perfectly computed number can't decide how much a particular customer ought to care about it.

I haven't tested whether Homh scores alter ChatGPT or Gemini recommendations. Two profiles don't establish any rating's accuracy or predictive value. That requires repeated tests in consumer-facing products, clear scoring definitions, and independent outcome checks. It's an investigation worth doing, not a finding to announce.

What we can see today is consequential enough: operational data is beginning to travel from business software into public choices about whom to hire. That could make service-business evaluation substantially better. It could also replace one set of simplistic reputation signals with a more technical-looking set if nobody insists on context.

The next question for AI recommendations isn't just whether a business has evidence. It's who decides which evidence counts for this customer, for this job, and why.