The Short Answer
You can build your own CRM, and a lot of small businesses should — at least at first. Four routes are realistic: a spreadsheet, a no-code database, an AI coding tool, or commissioning a build. They differ less in what they can do on day one than in where they stop.
What decides the route isn't the technology. It's how many people need to touch the data, how much of your money depends on that data being right, and whether anyone at your company wants to own the maintenance. This guide walks the four routes with published prices, real ceilings, and the failure mode that ends each one. If you're pricing a subscription CRM instead, the per-seat list prices are here.
Before Any Route: Write the Process Down
The first hour is the same whichever way you go, and it costs nothing. Write down what happens between a lead arriving and an invoice being paid:
- The stages. Not the ones a CRM demo shows — yours. "Inquiry, quoted, scheduled, done, invoiced, paid" is a perfectly good pipeline.
- The fields. What you need to know about a customer to do the work. A vehicle registration, a property address, a treatment history, a job number.
- The people. Who needs to see what. Most CRM disasters are permission problems wearing a different hat.
- The moments something must happen on its own. A quote that goes unanswered for five days. A job marked complete with no invoice raised.
If you can't write that down, no tool will fix it, and building anything now just encodes the confusion in software. It's also the honest test for whether you're ready to commission a build: describing your process is the client's job, not the developer's.
Route 1: A Spreadsheet
Start here if you're starting anywhere. Google Sheets allows up to 20 million cells or 100MB per spreadsheet, so your contact list is nowhere near the size limit. That isn't what breaks.
What breaks is everything around the data:
- No validation. The same customer ends up entered three ways, and your pipeline count is wrong.
- No history. You can see that a deal sits at "quoted." You can't see who moved it there, when, or what it was before.
- No real permissions. Anyone with the link can read every row and delete any of them.
- No automation. Nothing chases the quote that's gone quiet. A person has to remember, and eventually a person doesn't.
- Concurrency. Two people editing at once is where the trust goes.
Right call when: one or two people, a few hundred open records, one pipeline, and the sheet is a shared memory aid rather than the system of record for money.
Cost: effectively zero.
Move on when: a lead goes cold because nobody owned the follow-up, or you catch yourself maintaining a second sheet to fix the first one.
Route 2: A No-Code Database
Airtable and Notion are the serious DIY answer, and the most underrated option on this list. You get typed fields, relations between tables, views per role, forms that write straight into the database, and built-in automations — without writing code.
Published prices, checked September 2026:
| Tool | Plan | Price | Record limit |
|---|---|---|---|
| Airtable | Free | $0 | 1,000 per base |
| Airtable | Team | $20 per user per month, billed annually | 50,000 per base |
| Airtable | Business | $45 per user per month, billed annually | 125,000 per base |
| Notion | Plus | $10 per member per month | — |
| Notion | Business | $20 per member per month | — |
Two things to plan for.
It's per seat too. The usual reason people go DIY is to escape per-seat CRM pricing, and this route charges per seat. Six people on Airtable Team is $120 a month, or $1,440 a year — the same order of magnitude as a mid-tier subscription CRM for a team that size. You're buying fit and control here, not a lower bill.
Automation is often a second subscription. Native automations cover the simple cases; anything crossing tools usually means Zapier or similar. Zapier's self-serve plans price on tasks rather than users: Professional lists at $19.99 a month for 750 tasks and $49 a month for 2,000, billed annually.
Where it breaks: record ceilings on the plan you're on; anything customer-facing, because a client portal means layering on yet another product; permission rules with real depth ("technicians see only their own jobs, at their own branch"); and transactions — taking a deposit and issuing a numbered invoice against a job is where no-code starts fighting you. Your process also ends up living across three vendors' subscriptions, each with its own renewal.
Route 3: Build It With AI
This is what "build your own CRM with AI" usually means now: describe the application to a tool like Replit, Lovable or Claude and let it write the code. Replit lists Core at $20 a month ($18 billed annually) and Pro at $100 a month ($90 billed annually), with usage billed on top. It genuinely works for a first version, and a first version you can click through is worth more than a specification document.
The caveat is measurable rather than philosophical. Veracode's Spring 2026 GenAI Code Security update tested more than 150 large language models across 80 coding tasks in four languages. Syntax correctness now exceeds 95% — models reliably write code that runs. Security pass rates sat at roughly 55%, which the report describes as "virtually identical to where they stood two years ago." By vulnerability class the spread is wide: cross-site scripting passed 15% of the time, log injection 13%. By language, Python scored 62%, C# 58%, JavaScript 57% and Java 29%.
Put plainly: it's close to a coin flip whether an unreviewed AI-generated feature ships a known vulnerability class — in a system holding your customer list, their addresses, and possibly their card details.
That isn't an argument against building with AI. We build with AI every day; it's why a working system takes weeks here instead of quarters. It's an argument that AI output needs a reviewer who can tell secure code from code that merely runs, and that reviewer is what the $20 subscription doesn't include. The same logic applies to AI-written content, which we cover in is AI content bad for SEO: the tool is fine, the unreviewed output isn't.
Then there's the second 80%. Sign-in and password resets. Role-based access. Backups you have actually restored from once. An audit trail. Schema migrations for when you change your mind about a field six weeks in. Error handling that tells you something broke before a customer does. Prototypes are fast; that operational layer is where the weeks go.
Right call when: you're technical enough to review the output, or the system is internal-only and low-stakes — no customer personal data, no payments — and you're treating it as a prototype that proves the process before anyone commissions anything.
Route 4: Commission the Build
The trade is different here: a fixed fee up front, and you skip the operational layer entirely because it's someone else's job. No per-seat pricing, because a login is a role rather than a billing line. You own the code and the database, so the system doesn't disappear if you stop paying for support.
It's the expensive route on day one and the cheap one by year three if your headcount grows. Whether that trade is worth it depends on numbers you can run yourself: the break-even formula is in our CRM cost guide, and the full decision framework — five tests and a worked three-year example — is in build vs. buy software.
Two things to insist on from anyone you commission, us included. First, ownership in writing: the code, the database, and the right to take both elsewhere. Second, a change process you can actually use after launch — ours works the way website automation changes do, described in text or voice and shipped without a change-order negotiation. What a scoped system includes is on the custom CRM development page, and the same shape applies to the booking, quoting and portal systems under custom business software.
Side by Side
| Spreadsheet | No-code database | AI-built | Commissioned | |
|---|---|---|---|---|
| Cost | ~$0 | $10-$45 per user per month, plus automation | ~$20+ a month, plus your time | Fixed build fee, then a usage-based plan |
| Time to useful | An afternoon | Days | Days to weeks | Weeks |
| Price grows with | Nothing | Headcount | Usage | Usage, not headcount |
| Ceiling | Concurrency and trust | Record limits, portals, deep permissions | Whatever you can maintain | The scope you paid for |
| Who maintains it | You | The vendors | You | Your provider |
| Customer-facing portal | No | With another product | If you build it | Yes |
| Best for | 1-2 people, hundreds of records | Small teams, standard workflows | Technical owners, internal prototypes | Growing teams, specific workflows |
What Every Route Still Has to Get Right
Whichever one you pick, a CRM that works does these things. Check your build against the list:
- One record per customer. Duplicates are the most common reason a DIY CRM quietly stops being trusted.
- A stage that can't be skipped by accident. Especially the one between "work done" and "invoice sent."
- History. Who changed what, and when.
- Backups you've restored from at least once. A backup you've never tested is a hope, not a backup.
- Permissions by role. Not "everyone can see everything and we trust each other."
- A place for money. Quote, invoice, paid, with numbers that don't collide.
- A way in from the website. Inquiries should land in the CRM, not in an inbox where they compete with newsletters.
Start Small, Then Decide
The sequence that wastes the least money is almost always the same. Run the process on a spreadsheet for a month, because it costs nothing and it forces you to learn what you actually track. Graduate to Airtable or Notion when the spreadsheet's edges start showing, and stay there as long as it fits — plenty of good businesses never leave. Commission a build when the seat bill is climbing, the workarounds have become policy, or you need something customer-facing the no-code stack can't reach.
If you're somewhere on that path and want a second opinion, bring your current setup and your headcount to a free consultation. We'll run the numbers with you. If the honest answer is "stay on Airtable," that's the answer you'll get.