It's a Tuesday afternoon at a hypothetical twelve-electrician contractor, and the office manager has three browser tabs open. The first is a free trial of an AI estimating app. The second is a developer's proposal for a custom assistant that would read panel photos and draft quotes. The third, opened almost by accident, is the settings page of the field-service software the company has used for four years, where a section called “automations” has never been touched.
All three could plausibly help with the same headache: estimate requests piling up while the estimator is out on site. It feels like a technology choice. It's mostly a question about the shape of the work, and you can answer it without knowing anything about how AI models are built.
There are five options, not two
“Build or buy” leaves out the cheapest answers. Before comparing a custom system with a product, put all five on the table:
| Option | What it means | Good fit when |
|---|---|---|
| Skip | Leave it manual for now, or fix the process instead | It happens rarely, or nobody agrees on the steps |
| Switch on | Use a feature in software you already pay for | The tool you work in every day already offers it |
| Connect | Glue between tools you already have | Each tool works; the hand-offs between them don't |
| Buy | A focused product built for this one job | Your version of the job looks like most businesses' version |
| Build | A custom system shaped around your process | The process is unusual, spans several systems or is sensitive |
The connect option is easy to miss. Plenty of “we need AI” turns out to mean “information is stuck in one place and needed in another.” A carpenter's site photos sit in a text thread while the quote lives in a spreadsheet; a new lead arrives by email while the schedule lives in the field-service app. Connecting those can matter more than adding anything clever.
Flowchart
Build, buy or skip: a decision flow
- StartOne chore that eats time, e.g. “turn EV charger inquiries into quotes”
- DecisionDoes it happen at least weekly, following mostly the same steps?NoA personSkip for now: write the steps down and keep a person on itYes
- DecisionDoes software you already pay for offer it?YesStepSwitch it on and test it with ten real examplesNo
- DecisionIs the real gap moving information between tools you already use?YesStepConnect them; no custom AI requiredNo
- DecisionDoes a focused product handle your version without workarounds?YesStepBuy it, on a monthly plan you can cancelNo
- A personOwner confirms the process is stable and names who will maintain a custom system
- DoneBuild custom, scoped to that one chore
Connecting is less glamorous than either buying or building. In practice it looks like a web form that creates a job record instead of an email, a missed-call text whose answers land in your CRM, or a supplier's emailed price list that updates the cost column in your estimating sheet. Some of that uses built-in integrations; some needs a little AI to read messy inputs, like pulling the main breaker size off a photo of a panel label for the electrician to confirm. If every tool already does its own job well and the only complaint is retyping, start here.
If the flow leaves you weighing a product against a custom build, this is how the two usually differ:
| Off the shelf | Custom | |
|---|---|---|
| What you pay for | A subscription, often plus usage; setup is mostly your own time | Design and build up front, then hosting, usage and upkeep |
| Time to first use | Usually shorter: account setup, settings and testing | Longer: scoping, building, then testing on your own past work |
| Who maintains it | The vendor updates the product; you keep your settings current | You, or whoever you hire, each time prices, forms or systems change |
| Fit to your process | You adapt to how the product works | It's shaped around how you work |
| Control over data | Set by the vendor's terms, so read them | You choose where it's stored and who processes it |
| If you leave | You export whatever the vendor allows | Depends on who owns the code and configuration |
When something off the shelf is enough
If your version of a task looks like most businesses' version, someone has probably built a product for it. Appointment reminders, review requests, basic website chat, call answering and receipt capture are all well covered. The tests are practical, and you can run them during a free trial:
- Use ten real examples, not the demo data. Pull last month's actual inquiries, including the vague ones and the ones with blurry photos.
- Count the workarounds. If your team has to retype, copy or fix something on every item, the product doesn't fit your work.
- Check where the output lands. A great answer stuck in the vendor's dashboard, when it needs to be in your field-service or estimating software, has only moved the copy-and-paste.
- Read the data terms. Find out whether the vendor may train its models on your data, and how you'd export everything if you left.
Don't skip the switch-on option. Field-service, accounting and practice-management platforms add automation features regularly, often within plans you already pay for. Fifteen minutes in the settings menu, or one email to your account rep, can save you a purchase.
When custom makes sense
Custom work earns its keep in a narrower set of situations. Consider a three-person shop that builds custom built-ins. Its quotes come from the owner's own assemblies (cabinet boxes, face frames, finish options) priced against whatever the last lumber delivery cost. Generic quoting apps assume a price book of fixed items; his work is closer to a recipe.
A custom drafting tool that reads his measurements and site notes and builds a quote from his assemblies, for him to check line by line, is a reasonable case.
Signals that point toward custom:
- Your process genuinely differs from your competitors', and the difference is part of why customers choose you.
- The job touches three or more systems, and no single product connects to all of them.
- The data is sensitive enough that you want control over where it's stored and who processes it.
- You've already tried an off-the-shelf tool and can point to exactly where it broke on real work.
One signal that doesn't count: “we want something that's ours.” Ownership comes with upkeep. Someone has to update a custom system every time your prices, forms or other software change, and that someone should have a name before the build starts.
Before and after
Buying a tool for a hand-off problem
Bought a new AI quoting app
- 01Inquiry photos arrive by text on the owner's phone
- 02Measurements go into a notebook on site
- 03Owner retypes both into the new app at night
- 04App drafts a quote; owner retypes it into the invoicing tool
ResultTwo more places to copy information, same late nights
Connected what was already there
- 01Inquiry form collects photos, sizes and budget up front
- 02Site notes go by voice memo into the same job record
- 03Draft quote is built from the job record and price list
- 04Approved quote flows straight into the invoicing tool
ResultOne record per job, typed once
Three questions only a custom build raises
If the flow points you to custom, get written answers to these before signing anything:
- 01When our prices, forms or other software change, who updates the system, how quickly, and on whose bill?
- 02What will it cost to run each month, separate from what it costs to build?
- 03Who owns the code, the prompts and the configuration if we part ways?
The rest of your due diligence (where data goes, how it's tested, how you leave) applies to any outside help, custom or not. Our 12 questions for hiring an AI consultant cover those.
If you'd like someone to run the flow with you, it's a standard part of our consulting work, and any fair adviser will tell you when the answer is “buy” or “skip.” For the money side, see what AI automation really costs, including the lines that rarely appear on a quote.
What stays with a person, whichever you choose
Build, buy or connect, the boundaries don't move. For an electrical contractor, load calculations and code questions stay with a licensed electrician; software can collect photos and answers, not decide whether a panel can take a charger. When a caller mentions a hot panel, sparks or something burning, the system gives only the instructions you've approved (keep away from it; leave and call 911 if there's smoke or the smell gets stronger) and pages the on-call electrician at once.
For a carpenter, structural choices and anything that has to pass inspection stay with the builder. In both trades, no price reaches a customer until a person has approved it. Those boundaries also make the decision easier: a product that can't be set up to hand those cases to a person is ruled out on the spot, however good its demo looked.