We Built a CRM Most Employees Rarely Need to Open

Most people at BitSafe rarely open our CRM database. They still use it every day.

An account executive can post a short call update in Slack. An agent resolves the company, creates or updates the right records, and routes the next step. Meeting notes connect to the same customer context automatically.

This works because Notion already holds the company's shared knowledge and project state. The CRM is one layer of that system, used mainly by sales, marketing, and product. Engineering works from the same workspace without needing the customer records. Slack gives the business teams a faster route into the CRM, while agents and deterministic automations keep the data structured.

We started with a project-management problem. Teams had local conventions and useful documents, but cross-company questions still required someone to reconcile the answer.

The redesign expanded into the internal knowledge base, meetings, CRM, and shared directories. Companies, apps, opportunities, contacts, customer or partner contracts, documents, projects, and tasks became related records instead of separate lists.

That foundation let us move from Salesforce to Notion in about eight weeks. We kept the fields and reports that supported decisions instead of reproducing every Salesforce object. Customer context could then sit beside the meetings and work that changed it. The migration is documented in We Replaced Salesforce With Notion in Eight Weeks.

A CRM stores customer state and gives employees a way to update it. We separated those jobs.

Notion handles structure. An opportunity has an owner and stage, then relations connect it to the company, app, meetings, documents, and follow-up tasks.

Slack handles much of the interaction. An account executive can write:

The agent resolves the records, asks for any required detail it cannot infer, and writes a well-formed update into Notion. Rules assign the owner. Autofill adds derivable metadata. The follow-up stays connected to the opportunity.

I prefer this setup to Salesforce because customer records live beside the knowledge and work that shape each relationship. The team can work from Slack without using a dedicated CRM screen as the only doorway into customer data. Salesforce is moving toward a similar conversational model in Slack.Salesforce's conversational CRM overview

People report what happened and make the judgment calls. Agents populate fields that can be derived from approved context, while formulas handle fixed transformations.

This supports a deeper schema than we would ask an account executive to maintain manually. Every property still needs a downstream job: a decision, routing rule, risk check, or retrieval task. Fields without one become stale complexity.

Meetings feed the same system. Notes connect attendees to companies and opportunities, preserving decisions and follow-ups as operational context. We kept Fathom because it joins calls automatically; dependable capture matters more than forcing the entire workflow into one tool.

Ramp followed the same broad sequence, consolidating work before scaling its agent layer.Ramp runs on Notion

Our registry lists 75 active agents, but the useful distinction is what each one owns.

We start interpretation-heavy workflows with stronger models. Once the steps become predictable, mechanical work moves to a cheaper model or deterministic automation. Notion holds the state; NanoClaw covers external data and repeatable execution outside it.

Trust starts with scope. Every agent or automation has an owner, purpose, trigger, and permission set. Instruction changes move through a recorded propose, approve, apply process. Hygiene agents flag malformed records, and Slack carries actionable exceptions to the people who can fix them.

For one-to-many customer communication, a human approval gate stays in place. Notion drafts the message and assembles the context. NanoClaw distributes it only after approval.Slack's agent-design guidance

Agents failed when the underlying records were inconsistent. Dashboards built before the data stabilized looked polished and produced numbers people could not trust. Broad alerts became noise.

The order that worked was straightforward:

Following that order gave us a CRM employees can use without constantly visiting it. The team spends less time translating between systems, while Notion preserves a durable company record. AI can work across BitSafe because the company is structured for it.

Subscribe to the BitSafe newsletter for more from the Canton ecosystem and occasional behind-the-build notes on how BitSafe operates.