How one AI startup cut its CRM bill from $40,000 to $1,200 — and ended up with a better product in the process.
There’s a particular kind of frustration that comes from paying a lot of money for software that makes you feel stupid. Not because it’s too advanced — but because you only need 20% of it, and you can’t figure out how to turn the other 80% off.
That was the situation at Atonom, a startup building AI agents that handle sales development, customer support, and operations. They were paying $40,000 a year for Salesforce. For 25 to 30 people. CRO Gabe Larsen put it bluntly: “This is crazy for the functionality we need.”
What happened next is a case study in what’s becoming possible when software teams — or even non-engineers — stop asking “which tool do we buy?” and start asking “what do we actually need?”
The Usual Options Didn’t Help
Atonom’s team looked at the usual suspects. HubSpot. Pipedrive. Others. Every one of them offered a different price point and a different interface, but the same underlying deal: pay for a platform built around someone else’s assumptions about how you should work.
The team considered the nuclear option — just moving everything to Excel or Google Sheets. “The CRMs were too much work or too expensive,” Gabe explained. It wasn’t a joke. For a lean team optimizing for speed, a spreadsheet you actually use beats a CRM you don’t.
“It was 40 grand a year for 25 or 30 people. We were just like, this is crazy for the functionality we need.” — Gabe Larsen, CRO, Atonom
The Pivot Came From Finance, Not Sales
Here’s where it gets interesting. The solution didn’t come from the sales team, a product manager, or an outside consultant. It came from Jason, Atonom’s Head of Finance and Legal.
Jason had been experimenting with Lovable, a tool that lets non-engineers build functional web apps through natural language prompts. He told Gabe: “I’m going to build you a CRM.” Gabe laughed it off.
Jason took it as a challenge. Three hours later, he had a working prototype. They showed it to the sales team. Started collecting feedback. Iterated.
Within weeks, nobody was logging into Salesforce anymore.
What $1,200 Buys You (When You Build It Yourself)
The Lovable-built CRM now runs Atonom’s entire sales operation. Annual cost: roughly $1,200, including hosting. That’s a $38,800 reduction — and the savings compound beyond the subscription.
They eliminated a dedicated CRM admin. No implementation cycles. No dev team tied up in configuration. No ongoing maintenance overhead. Jason runs the whole thing in his spare time alongside his actual job.
The system covers everything they genuinely need: lead capture and tracking, lead source categorization, automatic account and opportunity creation, ARR/MRR tracking, probability weighting, weekly pipeline focus flags, sales dashboards, and basic reporting.
“This is where we spend 80% of our time. What deals, which ones are hot, what are we going to close this week.”
Does it have every bell and whistle of HubSpot or Salesforce? No. Does it need to? Also no. It’s purpose-built for how they actually operate — which is the point.
Then They Connected Their Own AI to It
Because Atonom owns the codebase, they could do something no off-the-shelf CRM would easily allow: they wired it directly into their own product.
Atonom builds AI sales agents. One of them is named Zoey. Zoey handles inbound leads — phone calls, email follow-ups, SMS, LinkedIn outreach, meeting scheduling. The whole sequence.
Now when a new lead submits a form, the Lovable CRM creates the record, generates an opportunity, and automatically triggers Zoey to start working the lead. Slack alerts keep the team in the loop in real time.
The CRM doesn’t just store pipeline. It activates it. That’s the difference between software you buy and software you own.
The Pattern That Emerged
Once the CRM was stable, Jason kept building. Finance now runs contract-based revenue modeling, budget tracking, forecasting, and executive dashboards — all inside Lovable. End-of-month reports that used to require manual assembly now update automatically.
Marketing is next. They’ve been running project management in spreadsheets; this quarter they’re building a purpose-fit tool instead of buying another SaaS seat.
The pattern is the same every time: identify a spreadsheet-level problem, build a working version in hours, collect feedback, iterate. Each function at Atonom owns one internal build.
Why This Matters for Marketers and Builders
There are two lenses worth applying here.
For marketers: this is a story about product-market fit applied internally. Atonom didn’t succeed by buying better software — they succeeded by getting ruthlessly specific about what they actually needed and building to that spec. No more, no less. That discipline is rare, and it’s worth imitating.
For builders: the tool category here is significant. Lovable is part of a wave of “vibe coding” and natural language app-building platforms that are fundamentally changing the build-vs-buy calculus. When a Head of Finance can ship a production CRM in an afternoon, the old argument — “we don’t have the dev bandwidth to build internally” — loses most of its weight.
“It’s a little bit off if we’re selling AI agents and telling people to be more agentic — and yet we’re using Salesforce and hiring 500 people.”
There’s also a deeper point about alignment. Atonom sells the idea that AI can make teams leaner and faster. Running a bloated SaaS stack internally would be a contradiction in terms. Building your own tooling isn’t just a cost-saving measure here — it’s a proof of concept.
The New Math
Off-the-shelf software promises convenience. What it actually delivers is someone else’s opinion about how you should work, packaged inside a subscription you’ll keep paying because switching feels like too much trouble.
That bargain made sense when building was hard. It makes less sense every month.
The tools to build are finally good enough that the old calculus no longer applies. The question isn’t whether your team can afford to build. It’s whether you can afford to keep not building.
Atonom went from $40,000 to $1,200. They got a better product. Their head of finance runs it in his spare time. And their AI agent is plugged directly into it.
That’s the story. The math is pretty hard to argue with.