Why I Chose a CRM I Could Talk To

Three weeks ago, I launched the updated Lituus Studio — a solo digital transformation consultancy based in Melbourne. And I got the response that any new business owner wants: an influx of conversations.

Paid engagements starting to take shape. Partnerships forming. Collaborations emerging. The opportunity to run workshops. Exploratory chats that might become something or might not. All at once, across different people, different organisations, different timelines.

And pretty quickly, that became way too much to keep track of in my head.

I've worked with CRMs a number of times before — implemented them for clients, designed data models, migrated messy spreadsheets into structured systems. So I know what happens when you don't have a central home for this stuff. Context gets scattered. Research you've done on a prospect lives in one place, the email thread lives in another, the notes from a coffee catch-up live in your head. You forget to follow up — not because you don't care, but because the volume crossed the threshold where your brain isn't enough.

I needed a CRM. But I had a specific thesis about what that should look like.

The Two Questions That Matter Now

Part of my development thesis moving forward is simple: can I engage with the technology conversationally?

That's been the revelation of working with generative AI — getting things done through natural language. Anyone who's spent time clicking through an interface and filling in fields knows the friction of that. I wanted to see how easy it would be to get a CRM up and running that could fit into a natural workflow — one where I'm talking to Claude, not clicking through forms.

That led me to two selection criteria that aren't on most people's checklists yet:

Does this tool have an AI connector? If Claude can read and write data directly, the daily workflow isn't open CRM, find record, click into deal, update field, type note. It's tell Claude what happened — which means the time you're spending is on debriefing and analysing how the meeting went and what to do next. Particularly as a solo founder, where you can't always do that with a colleague, the conversation is the data update. For me, this is what it's all about with AI: you're not segmenting your workflow into component parts. It can all just be part of a natural way of working. (You can build these connectors yourself now, so it's not a hard limit if one doesn't exist — but that's a layer of expertise and maintenance that slows things down. Where someone else is already optimising one, lean into it.)

Can I configure it programmatically? I've always loved technology. I've always hated configuring systems — it feels slow, cumbersome, and deeply tedious. But with Claude Code, that's changing. If I can create custom fields, deal stages, and attributes through an API, then I can design the system in one place and build it in another — fast, repeatable, precise. If I can't, I'm clicking through settings screens. No thanks.

I landed on Attio. Not because it's the "best" CRM — that question is meaningless without context — but because it answered yes to both. MCP connector for Claude. Clean API for programmatic setup. A free plan that's genuinely usable. And it supports embedding custom React pages directly inside the platform — so when I start wanting to build features that go beyond what the CRM offers natively, I can build applications inside it, not just configure what's already there.

Three Modes, One System

All of this came together in about four hours across two sessions. The first: defining the concept of what I wanted and doing some initial thinking about different technology solutions. The second: two hours configuring and populating the CRM. Here's what that looked like.

Step one: Design in Claude. I opened a conversation and described the relationships I'm actually juggling. I kept it messy and raw — I essentially gave a brain dump, and Claude did what it does best: structured that into a system.

It wasn't anything particularly fancy. A three-layer architecture: Deals for what I'm pursuing, Tasks for what I should do next, Notes for what happened and why. Beyond a few custom fields, there wasn't much more than I needed.

One of the things I considered was whether to use a field for "Relationship Type" on contacts. The instinct is to create a dropdown — Client, Collaborator, Partner, Connector — but the shift we're moving into is that AI models work much better with semantic context than they do with structured data. Notes kept on each person answer the relationship question with a hell of a lot more nuance than a dropdown ever could. And they don't become a data burden — they become a searchable knowledge base.

There were other things I changed too — deal stages, custom views — but that's not the point of this article.

Step two: Build in Claude Code. The output of that design session was a spec — an overview of what I wanted to build. I dropped that into a Claude Code project and had it create everything through the Attio API. Twenty minutes later, the entire configuration had been built programmatically.

This is why the "can I configure it programmatically" question matters. It's about speed and precision. And again — it's building and configuring through conversation. Any tool that's still requiring you to click around a UI to set things up is slowing you down and creating friction that, for me at least, makes me want to bang my head against the desk.

Step three: Operate through the MCP connector. Back in Claude with the Attio connector enabled, I started populating. I'd also connected Gmail and Calendar as Claude connectors, so it had additional interaction context to draw on.

It's critical here to know that I'd been running a dedicated Claude project to keep track of all my consulting work for weeks. Every thought, every brainstorming session, every bit of research, every workshopped email — it was all there. Which meant that when I started creating deals and notes in Attio, Claude already had deep context about not just the companies but every individual I'd spoken to over the last three weeks. It could write a detailed profile on a stakeholder I'd met for coffee, referencing the conversation we'd had, the introduction that led to it, and the strategic context around the opportunity — not from a template, but from real accumulated knowledge.

Fourteen deals created. Forty-plus notes across people and companies. A dozen follow-up tasks with deadlines. All created through conversation, not through forms. Zero records touched manually.

The Honest Accounting

Here's where I need to be honest about those four hours.

"Concept to configured CRM in two sessions" is true. But those sessions sit on top of a stack of things that took considerably longer:

The rich deal notes were possible because I'd been building context about these relationships for weeks. The whole thing was seamless because I'd done the work to make it seamless.

This matters because the hype cycle wants to tell you that AI makes everything instant and effortless. It doesn't. What it does is compress the gap between expertise and execution. If you know what you want and you understand the tools, the build is shockingly fast. If you don't, the tools can't save you.

The Takeaway

Two things I want you to sit with.

The speed is real. I'm not saying anyone can set up Attio in two hours — but with the right preparation and the right tools, you can get the first draft of a new system up and running extremely quickly. The distance between "I need this" and "I have this" is shorter than it's ever been.

The criteria for choosing technology are shifting. The questions that used to dominate — what does the UI look like, what do the review sites say, how many integrations are on the features page — are being joined by new ones. Can I talk to this tool? Can I build on it programmatically? A tool that answers yes to both is a categorically different experience from one you can only interact with through its interface.

So if you're evaluating any technology right now — not just CRMs — start asking those two questions. The paradigm is shifting. And if you're not thinking this way now, you're going to be kicking yourself in a few years when you see what's possible with tools that were designed to be built on.