Solution

Give the newer tech the answer, not a call back to the shop

A technician six weeks into the job gets sent alone to a crawlspace for a furnace model the company hasn't installed new in years, on one bar of signal, with the one person who'd recognize it on sight out on another call. An AI knowledge base for home services puts your equipment manuals, this property's own service history, your pricing book, and your warranty terms in his pocket instead — answered from his phone, cited to the specific document and model number, not a guess dressed up as confidence.

· Reviewed by Artur Horimoto, Founder & CEO

The gap between the tech on-site and the one person who knows

Every home services company has at least one person who can look at a control board and say what it is, who remembers that this particular subdivision was all plumbed with a supplier's fittings that fail the same way, who knows this customer called in twice last winter about the same noise. That person is valuable precisely because dozens of manufacturers, spanning twenty or thirty model years, do not document themselves the same way, and a lot of what separates a fast fix from a wasted trip is knowing which manual applies to which unit.

The problem is that person is one person. They're on a job, on vacation, or asleep when the phone rings at 9pm, and the technician standing in front of the actual equipment is someone newer — capable, but not yet carrying twenty years of "oh, that one's different" in their head. The current option is a call back to the shop, which means someone else stops what they're doing to answer a question from memory, sometimes correctly and sometimes not. A knowledge system doesn't replace that person's judgment. It takes everything that's already written down — manuals, spec sheets, service records, your own pricing and procedure — and makes it findable by whoever is standing in the basement, in the moment they need it, cited back to where it came from.

What the system does on a normal day

The questions repeat even though the equipment never does.

  • Equipment manual and spec sheet retrieval. A tech gives a model number, a nameplate photo, or a rough description, and gets back the wiring diagram, fault code table, or parts breakdown for that exact unit — not the nearest match from a different model year.
  • This property's own service history. What was installed, when, by whom; what's failed before at this address; what the last technician noted on the ticket, in their own words, not a summarized version of it.
  • Pricing and plan coverage. What this repair costs under your current pricing rules, and whether it's covered under the maintenance plan this customer is actually on.
  • Warranty terms by manufacturer and install date. Whether a part is still inside its manufacturer warranty window, and what that manufacturer requires to honor a claim.
  • Internal procedure. How this company does a specific changeout, step by step, and what the callback policy actually says — not what someone remembers it saying.

Three retrievals from a phone in a crawlspace

The clearest way to see what this looks like is to walk through it.

A fault code on a furnace nobody here has touched in years

The tech is in a crawlspace with one bar of signal, looking at a blinking error code on a furnace board from a manufacturer the company stopped installing new units from a decade ago. He photographs the nameplate and types the fault code into the system on his phone. It matches the model and year against the indexed manual set, returns the fault code table for that specific board revision, and states the manual and page it came from. If the photo is blurry enough that the model number is ambiguous between two similar units, the system says so and asks him to confirm rather than picking one — because a confident answer about the wrong model is worse than no answer, and a wrong wiring diagram in a crawlspace can do real damage.

What was installed here, and what already failed

A repeat call comes in at the same address. Before ringing the doorbell, the tech pulls up the property and sees what was installed three years ago, the two service visits since, and the previous technician's note that the capacitor "read low but within spec, worth watching." That single line changes the visit — he brings a replacement capacitor along instead of diagnosing from scratch, and when it turns out to be the actual fault, the fix takes ten minutes instead of a second trip. The history came out of your service platform's job records, not out of anyone's memory of "didn't we service this place before?"

What's covered, what it costs, and how this company does the changeout

A compressor has failed on a unit installed two years ago. The tech checks the install date against the manufacturer's warranty terms and confirms the part is still covered, then checks the customer's plan to see the labor is included under a maintenance agreement rather than billed separately. Before starting, he pulls up this company's own changeout procedure for this equipment class — the sequence the business actually wants followed, not a generic one — and the current callback policy, so he can tell the homeowner exactly what happens if the same unit acts up again in the next thirty days. Every answer in that sequence points back to the specific price sheet, warranty document, or internal SOP it came from.

What this doesn't fix — and where it gets hard

Being straight about the hard parts matters more here than almost anywhere else, because a wrong answer in the field isn't a minor inconvenience — it can mean a damaged unit or a technician working from the wrong diagram.

Model identification is the sharpest edge. Home service companies deal with dozens of manufacturers across decades of equipment, and a nameplate that's rusted, painted over, or photographed at an angle makes the match genuinely uncertain. We build the system to say when it isn't sure, rather than returning the closest match as if it were exact — a flagged gap a tech can resolve with one more question is far better than a confident answer about the wrong unit.

Connectivity is the other real constraint, not a footnote. A basement or crawlspace with one bar of signal is the actual operating environment for a meaningful share of these lookups, so the interface has to stay light — text and short answers rather than image-heavy pages that stall on a weak connection — and we design against the worst signal on the job, not the best one in the office.

Scanned older manuals add a third layer. Plenty of manufacturer documentation for equipment still in the field only exists as a scanned PDF, sometimes a poor one, and OCR quality on that material varies. Where confidence in a scanned read is low, the system flags it instead of presenting a shaky extraction as clean text.

Connecting to your service platform, manufacturer manuals, and pricing book

This runs on the same approach behind Calfy's knowledge systems work generally, pointed at the specific mix of documents a home services company actually holds.

Service history. We connect to whatever platform already holds your job records — ServiceTitan, Housecall Pro, Jobber, or a similar tool are common across the trades — and read the property and equipment history that's already there. We aren't partnered with any of them; we build to whatever API or export your platform exposes.

Manuals and spec sheets. Manufacturer documentation comes from wherever you keep it — a shared drive of downloaded PDFs, an OEM portal export, scanned paper for older equipment. The document review work behind this is what turns a stack of scanned manuals into something searchable by model number in the first place.

Pricing and warranty terms. Your current price book and manufacturer warranty terms get indexed the same way, kept current as they change rather than answered from a version that was accurate last season.

Retrieval, not invention. The underlying technique — search the actual documents first, then answer only from what was found, with the source attached — is the same retrieval-augmented generation pattern behind any properly grounded system, and the knowledge base search use case covers the general shape in more depth.

Where it sits next to your phone and dispatch systems. This page is about a technician or dispatcher pulling an answer out of your own material. It doesn't answer the phone or book the job — that's what voice AI for home services does. The two are built to work together: the same indexed property history a knowledge system surfaces to a tech in the field is available to the system answering the phone before that tech is ever dispatched.

Frequently asked questions

How do you make sure the system doesn't hand a tech the wrong model's manual?

Every answer cites the specific document and model it came from, so a tech can check it against the nameplate in front of them. Where a photo or description is ambiguous between two similar units, the system says so and asks for confirmation instead of guessing — a wrong answer stated with confidence is worse than no answer at all.

Does this actually work on one bar of signal in a basement?

That's the environment it has to work in, so the interface is built light — short, text-based answers rather than heavy pages — and designed against a weak connection rather than a good one. It won't fix a spot with no signal at all, but it's built for the poor-reception norm, not the office wifi best case.

Can it pull the service history for one specific property?

Yes, from whatever platform already holds your job records — ServiceTitan, Housecall Pro, Jobber, or similar. It surfaces what was installed, prior visits, and the notes previous technicians left, in their own words, so a tech isn't starting a repeat call from zero.

How do you keep pricing and warranty terms current?

They're indexed from your actual price book and the manufacturer terms you work with, and updated on the same cycle those documents change on. A pricing answer that's a season out of date is worse than none, so freshness is part of how the system is built, not an afterthought.

How much does this cost and how long does it take?

It depends on how many manufacturers' documentation needs indexing, how much of it is scanned paper versus digital, and which service platform it connects to for history and pricing. We scope every engagement and agree a clear price before any build work starts, and most companies see the system answering real field questions within weeks.

Bring the model number your newest tech had to phone the shop about last month. Thirty minutes on a free strategy call is enough to tell you whether a knowledge system fits your crews, and roughly what it would take to build.

See what this looks like for your business

Thirty minutes, your actual workflow, and a straight answer on whether this is worth building.

Free 30 minutes. No pitch deck. You leave with a plan either way.