Jon-Torrey

co-founder SearchLight, raw data to insights for HVAC - Best data insights for contractors:

[Jon-Torrey @jon_torrey]

Quick thread in response to @garyvee biggest challenge getting ppl to execute his advice. I hope this helps many of you. 2 yrs ago I started to get bad anxiety. Stuck in my job. Wanted to do more. Randomly put on GV + he spoke about garage sales as a start-now way to build skills

[Gary Vaynerchuk @garyvee]

@princeofpvnk its my biggest challenge ...

[Carolina Hurricanes @Canes]

WHO WANTS A BEER SKATE?!

RT for your chance to win the hottest drinking device in sports.

[Jon-Torrey @jon_torrey]

The SearchLight Data Lab is my favorite AI-enabled project I've done this year.

The SearchLight Data Lab is my favorite AI-enabled project I took on this year. It's not done at the snap of a finger and still requires a lot of context, understanding of data, trends, taxonomy, etc. but without Claude, I wouldn't be able to do this.

See how your Google Ads CPL stacks up in your region: https://t.co/OoSldOpo3Z

[Jon-Torrey @jon_torrey]

🚨 New Case Study Alert - RevSync: A 25% Lift in ROAS When Google Ads Gets Real Revenue Data

Google explicitly recommends feeding offline conversion data back into their algorithm: 'Combine your online website measurements with your offline first-party data (from order management or data platforms) in Google Ads to improve campaign performance.'

RevSync is an activation product within the SearchLight data platform that automates this feedback loop with accurate, normalized data pulled from multiple call tracking providers, conversion tools (including forms and chats), and CRMs.

This case study tracks one contractor's Google Ads performance across 10 months: 4 months before RevSync (June-September 2025) and 6 months after (October 2025-March 2026).

It then compares the results against a control account with comparable Google Ads spend and no RevSync.

Full case study: https://t.co/rV95ep33IR

[Jon-Torrey @jon_torrey]

A residential electrical contractor paid Google LSA for this exact call last month, but our lead grading product caught it.

For years, we've said that lead volume and cost per lead told only part of the story.

But now, that's evolved even further because marketing has gotten more complicated and messy.

More channels, more conversion tools, more vendors, and fewer eyes on the quality of those contacts.

The most-asked question we get from our clients when they see ROAS is whether it could be a lead-quality issue.

So, we responded and built a proprietary domain-trained model, using hand-labeling, to grade contacts across all conversion tools with deep marketing integration to catch waste.

This transcript is the tip of the iceberg, and we recently published a case study about how, in just two weeks, we used SearchLight Lead Grading to improve a contractor's Google Ads ROAS from 0.1x to 3.8x.

Full case study: https://t.co/7FCChK6eJH

[Jon-Torrey @jon_torrey]

@HVACSEO Built this and all interactive views seen here in a weekend - https://t.co/7tJIwJDx2l - some overhead with updating and working through some issues with Claude but largely easy with some focused time and the output is phenomenal

[Jon-Torrey @jon_torrey]

Pumped to introduce SearchLight's data lab with live, interactive benchmark reports for The Trades - https://t.co/nSBPUVa61J

Google Ads CPL by Region for Q1 and ChatGPT / LLM lead and revenue gen for Q1 is live.

#hvacmarketing #trades #homeservices

[Jon-Torrey @jon_torrey]

@MSUKyleBrown Hell yeah - love seeing this!

[NFL @NFL]

Eagles agree to 1-year deal with CB Riq Woolen. (via @CameronWolfe)

[Jon-Torrey @jon_torrey]

GPT vs Gemini vs Claude vs ClaraT: Which AI Model Performs Best for Lead Grading in Home Services?
We ran 6,000 real home service transcripts through every major AI model:

GPT-4o-mini
GPT-5.2
Gemini 3 Flash
Claude Opus 4.6
Claude Sonnet 4.6
ClaraT™ (SearchLight’s domain-trained model with an ensemble architecture)

and measured how accurately each one classified leads into three categories:

Whether the lead was bookable
Why it was unbookable (if applicable)
Why a bookable lead did not convert

Why?

Because our clients make real decisions based on this data. Staffing decisions. Budget decisions. Decisions about which marketing channels to keep and which to cut.

If the classification underneath those decisions is wrong even 10% of the time, the downstream cost is significant.

General-purpose models clustered between 73–90% accuracy on bookable vs. unbookable.

That's impressive for tools built to do everything, but for a GM trying to figure out why Tuesday's phones were slow, "pretty good" isn't good enough.

ClaraT™, SearchLight's domain-trained model built on years of structured home services data and human experience, achieved 98.05%, more than 8% higher accuracy than the next-highest general model.

That gap led us to explore a concept we're calling Decision-Grade Intelligence (DGI).

Decision-Grade Intelligence is the point where data becomes accurate enough that you'd actually change how you run your business based on it.

We've worked in data for over half a decade, and a core reason why we've been able to drive success for our clients is the willingness to get into the weeds and the details to maintain a standard of accuracy.

Artificial intelligence, specifically LLMs, unlocks a new wave of productivity, and we're here to figure out how to close the accuracy gap between general models and domain-trained outputs.

You can read the full analysis here: https://t.co/vnQgHw2TfV

[Jon-Torrey @jon_torrey]

ICYMI, we broke out Cost per Lead for Google Ads (averages from 816 HVAC/Plumbing businesses) in January 2026 across Brande campaigns, non-branded, PMax, and a blended average.

Read more here:

https://t.co/GLnLgkTzaf

[Jon-Torrey @jon_torrey]

Compared 3 AI CSR providers across on-call book rate, human transfer request rate, planned follow-up rate, and more in my latest newsletter.

I also provide some commentary on 'book rate' - there's a lot more nuance that meets the eye and digging into that nuance can help stop revenue leaks.

TheDataDrivenTrades dot Substack dot com

[Jon-Torrey @jon_torrey]

@Seth_Parker_ 🤝honored for that call out. Our team has poured so much of our lives into searchlight. To stay bootstrapped and solve messy data problems isn’t easy but we do genuinely care. Kevin’s dad is a plumber and that’s the seed for everything we do.

[Jon-Torrey @jon_torrey]

@MSUKyleBrown @SMBDistribution @STLChrisH Ah I see - no we don’t. Haven’t gotten there yet - we are still bootstrapped and growing like crazy so just have to prioritize but should that change I’ll let you know! Always happy to try and help where I can

[Jon-Torrey @jon_torrey]

@MSUKyleBrown @SMBDistribution @STLChrisH Our largest sample is hvac/plumbing but we have roofers lawn care garage door pest control, but it’s a battle of time to get all that out, can you clarify what you mean by distribution?

[Jon-Torrey @jon_torrey]

Humbled that @STLChrisH shared this content - Substack being down right now sucks but all my channel analysis for contractors lives here - https://t.co/rV7lyMPFy0

[Chris Hoffmann @STLChrisH]

In January, HVAC and plumbing contractors paid an average of $104 just to show up at the front door of a new customer.

Before paying for wages, vehicle, insurance, overhead…

The “Google tax” is real, and painful for customers and contractors alike.

[Jon-Torrey @jon_torrey]

@SMBDistribution @STLChrisH The 5x is blended - I broke ROAS down by unbranded pmax etc - https://t.co/rV7lyMPFy0

[Jon-Torrey @jon_torrey]

@HockJohannes @STLChrisH Service type details are in the article but Substack is down right now :( - https://t.co/rV7lyMPFy0

[Jon-Torrey @jon_torrey]

We tracked LLM-referred customers across 1,000 home service companies through December.

The results are a bit of Jekyll and Hyde.

Top performers are seeing real pipeline. The average contractor is getting almost nothing.

Read more here: https://t.co/FSN07v6xMe

#hvacmarketing #trades #ai #marketing

[Jon-Torrey @jon_torrey]

Unbranded vs. Branded Google Ads performance for December is now LIVE in our latest newsletter!
Just going 'one level deeper' with data can yield much more insight, but it always comes back to the fact that your business, your goals, your market, etc. are unique and have different needs.

Read more here: https://t.co/WozQSO9JEg

#hvacmarketing #homeservices

[Jon-Torrey @jon_torrey]

Does Facebook drive revenue for HVAC contractors?

We analyzed $2.3M in Facebook ad spend across 262 HVAC advertisers to answer it, using closed revenue only, not modeled or view-through conversions.

The results surprised us (especially the gap between average and top performers). It CAN work, but there's a gap and it needs to be monitored w/ proper expectations (demand gen vs. demand capture)

Full breakdown, methodology, and 2025 benchmarks:

https://t.co/Ln0rCseeYH