Guides on CRM setup, automation, and data management.
I give away a 48-hour Attio audit. You add me as an expert, no seat, and I send back a ranked teardown. After doing a stack of them, the findings rhyme. The workspace is almost never broken. The data it captures almost never gets used. Here is the pattern, and why it changes what I build.
I built a tool that creates Attio objects, attributes, and records from a CSV or another CRM. The Attio REST API is good. The parts that cost me time were not in the happy-path docs: required fields that look optional, attribute types you can make by hand but not by API, and relationships that force a two-pass design. Here is the real list.
My best lead source is a directory I can't climb. The tier above me is gated on a number my business structurally does not produce. So I stopped trying to rent a better spot on someone's list and started building the kind of visibility the web hands you on its own.
For two years I sold the same judgment over and over, by hand, while I was in the room. The moment an engagement ended, it stopped. Now the judgment ships as a Claude Code skill the client runs themselves, and it keeps working after I leave. Here is what productizing a consulting playbook actually looks like, and what it changes about the business.
A few weeks ago I rebuilt my whole strategy around getting cited by AI instead of climbing a directory I cannot climb. Here is what I actually did first, and why it was a free tool and not another article.
I asked ChatGPT to recommend an Attio expert. It named four of my competitors and not me. So I traced exactly where that answer comes from, found the lever I needed was structurally closed to me, and changed my whole content strategy because of it.
I run an Attio implementation business. I had a working theory for why it sells, but it was fuzzy. Then a 20VC panel spent two minutes on the AI expertise gap and said the thing I had been circling for a year. Here is what they got right, and where my lived experience says they are slightly off.
AttiBooks is live on the Attio Marketplace. For a few days in April it was not. The invoice sync kept dying, and the reason was not QuickBooks or a slow API. It was an assumption I made about the runtime. Here is the real bug, the fix that did not work, and the one that shipped.
Not content. Not cold email. Not ads. The single best source of customers for my Attio implementation business is a free listing in a directory I do not own. Here is why it works, what most people get wrong about directories, and how I would copy it from scratch.
Every AI agent demo on the timeline is either a consumer chatbot or an enterprise pilot. Almost nothing in the market is shaped for the segment where agents pay off fastest: the small services business. Here is what the missing agent looks like, why nobody is building it, and what I see clearly from running one.
Every CRM vendor's 2026 roadmap is a longer feature list. The team's actual problem is not a missing field. It is the work that sits between the fields: the pipeline hygiene, the meeting prep, the follow-up that requires reading a transcript and writing four updates. None of that is a feature. All of it is an agent.
Most cold outreach calls itself personalized and means firmographic. The first line is the company name in a sentence. Real personalization happens per person, takes longer than founders are willing to spend, and is the only kind that earns a reply. Here is the rule I run our outreach by, and the format I use to keep it cheap.
The standard advice is to pick one: a services business or a product. I run both. Craftt does Attio implementations. AttiBooks is a product I shipped. Here is why that is not the contradiction it looks like, and the one rule I use to keep it from becoming one.
Most of an implementation business is repeatable knowledge work: discovery, a data model, a migration, automations, agents, a monthly cleanup. I stopped treating Claude Code as a coding tool and started running the whole business on it. Here is what that looks like, and what it changes about the math of a small services firm.
The normal way to ship an AI agent is a server, an API key, and a token bill. That makes you infrastructure. I ship agents a different way: a markdown skill that runs on the customer's own Claude Code subscription, scheduled in the cloud, with no server I own and no key I hold. Here is the architecture and what it changes.
Three humans, three named agents, one fixed price. The shape of a 7-day Attio implementation, the design choice that compresses the timeline, and the three things that still happen at human speed.