There's a phrase circulating in venture capital right now: services-as-software. Eric Siu, in a widely-viewed X post/video, laid out the logic behind it, and it's worth understanding even if you have zero interest in pitching Sand Hill Road. Because underneath the VC framing is a genuinely useful question for anyone who runs a business built on labor and expertise — which, if you own an HVAC, plumbing, roofing, electrical, or garage door company, is exactly what you run.
The thesis, in Siu's terms
Siu's argument starts with a simple ratio: for every $1 spent on software, roughly $6 is spent on services. Software has been the easy part to digitize for the last twenty years — SaaS ate a huge swath of the software dollar. The services dollar — the $6 — has mostly resisted that, because services have always required a human doing the work: a consultant building a deck, an agency running your ads, a technician diagnosing a furnace.
Siu's claim is that AI changes that. For the first time, you can build AI-native systems that actually deliver a service outcome — not just software that assists a human doing the service — and price it accordingly. That's the "services-as-software" model: agents plus human judgment where it's actually needed, replacing the traditional billable-hours, headcount-scales-with-revenue structure that services businesses have run on forever.
He argues this is exactly why a16z, Sequoia, and YC are, in his words, "chasing services, not SaaS" right now. Their thesis is that the firms who figure out how to restructure service delivery around AI-native, outcome-based models — rather than bolting AI onto an existing agency to cut headcount and pad margin, which he calls "playing the small game" — are the ones capturing outsized value. Siu ties a specific number to this: he claims AI-native service firms are commanding roughly 30x multiples right now, in the segment he's describing, which skews toward software-adjacent and agency-style service businesses, not home services specifically. Worth being precise about that scope, because it's easy to round it up into a promise it isn't.
Why should a trades owner care about a VC talking point?
Fair question. You're not raising a round. You don't need a16z to like your multiple. But strip out the capital-markets framing and there's a real operating insight underneath, and it applies to you directly, arguably more directly than it applies to a marketing agency.
Here's the reframe: your business already is a services business. You sell labor and expertise, not software. The thesis isn't telling you to become something new — it's asking a question about how you're organized to deliver what you already sell. Are you structured around billable hours and headcount, where the only way to grow revenue is to hire more people and bill more hours? Or are parts of your delivery restructurable around outcomes, with AI handling the pieces that don't require a human hand or judgment, freeing your people to focus on the parts that genuinely do?
That's the whole question. Not "should I add a chatbot," but "which parts of how I deliver work are actually billable-hours logic dressed up as necessity, and which parts require a licensed, experienced human being standing in someone's garage?"
What this could plausibly look like in a trades business
To be clear upfront: these are illustrative possibilities, not case studies, and not a promise of what AI will do for your shop. Nobody has proven a specific outcome here — this is a way to think about where the lever might apply, not a result to expect.
- Intake, scheduling, and follow-up as an end-to-end system, not a queue for your team. A lot of trades businesses lose money not because techs are slow, but because intake is inconsistent — calls missed after hours, follow-ups that fall through, scheduling that depends on whoever picks up the phone that day. An AI system that handles intake, books the job, and manages follow-up end-to-end, escalating only the genuine exceptions (an angry customer, an ambiguous job scope, a scheduling conflict that needs judgment) to a human, is a plausible version of "agents handle what doesn't need a human hand, your team handles what does." It's not hypothetical technology — the building blocks exist — but whether it changes your numbers depends entirely on your execution, not on the concept.
- Diagnostic and estimate support that gives technicians back billable time. Technicians often spend real chunks of a shift on documentation, photo logging, and writing up estimates — necessary, but not the part of the job that required their training. AI-assisted diagnostic notes and estimate drafting, reviewed and finalized by the tech, is a plausible way to shift more of a shift back to billable, hands-on work. Again: illustrative, not demonstrated.
- Post-job follow-through — reviews, warranty registration, maintenance-plan upsell — run as a system instead of a task nobody owns. This is the kind of work that's genuinely low-judgment but chronically under-resourced because it competes with dispatching the next job. It's a reasonable candidate for the "agent handles it end-to-end" model precisely because it rarely needs a human judgment call.
None of these are proof that "AI adds X to your business." They're a way to see your own operation the way the services-as-software thesis asks you to: not "where can I bolt on a tool," but "which parts of my delivery are structurally billable-hours, and which parts could be restructured around the outcome instead."
The honest caveat
Siu's 30x figure is about a different market — AI-native service startups, largely software and agency businesses being underwritten by venture capital. It is not a claim about home services multiples, and nobody should read it as one. If you're thinking about your business's eventual value, AI adoption is one lever among several that buyers screen for — it doesn't offset customer concentration, an aging fleet, or how dependent the business is on you personally. Those still move the number more than any AI system will.
What the thesis is actually useful for is sharper than a multiple prediction: it's a lens for looking at your own delivery model and asking where you're still organized around hours instead of outcomes.
If you want help figuring out where that lever actually applies in your operation, and where it doesn't, that's the conversation we have. Apply at advisy.com/#apply.