The five-hour CRM: what changes when custom software becomes affordable

The agency already had the raw parts of a sales system. Leads arrived through Google Forms and spreadsheets. Conversations happened in Gmail. Meetings lived in Google Calendar. Files landed in Drive. The failure happened in the handoffs. Information had to be copied between tools, follow-ups disappeared under new email, and the pipeline depended on someone remembering the next step.

Past experience with traditional CRMs made the tradeoff clear. Large platforms could handle the data, but disconnected interfaces and rigid workflows created hours of weekly data entry. A specialist vendor quoted a monthly retainer that did not make sense for the agency’s lead volume. Buying more software would have added cost without fixing the shape of the work.

AI makes software easier to create, but easier does not mean automatically worthwhile. The session emphasized a useful rule: build only when a custom workflow creates a meaningful operational advantage over a mature product.

A standard product is usually the better choice when the process is common, the integration is reliable, and the subscription costs less than maintaining your own system. Custom development becomes compelling when the process is specific, the missed handoffs are expensive, and the people closest to the work can describe the desired behavior precisely.

The CRM connects Gmail, Google Calendar, Google Sheets, and Google Drive into one working sales loop. It can import leads from a CSV, prepare email campaigns, track verified opens, and move a contact into a visual pipeline when a form arrives or a prospect responds.

Once an opportunity is active, the system follows it from first conversation through close. Revenue and commission estimates update with the deal stage. Before a discovery call, the CRM reviews available lead data and prepares a meeting flow with likely objections. It also sorts tasks by urgency so the team can see what needs attention now, between meetings, or later.

The first version ran through Claude, but scheduled polling introduced a delay of five to ten minutes. That is acceptable for some background work and frustrating for moments that should feel immediate, such as a new form submission or an incoming reply.

The team began testing a hybrid approach with Gemini handling Google Workspace events as they happen, then handing more complex reasoning to Claude. This separates fast event detection from deeper decision-making. It also reflects a broader pattern for AI systems: use the smallest dependable tool for each part of the job instead of forcing one model to do everything.

The demonstration was candid about what the prototype was not. It was built as an internal tool, had been tested by the team, and was about 90 percent complete. It was not presented as a certified system for regulated medical or financial data.

That distinction matters. Natural language can accelerate implementation, but it does not erase responsibility for access control, data retention, backups, testing, or compliance. A five-hour prototype can prove a workflow. Production use still requires review proportional to the sensitivity and consequence of the data involved.

The strongest takeaway was not that every company should build a CRM. It was that every company should inspect its recurring software costs and manual handoffs with fresh eyes.

Start with one workflow that is repetitive, measurable, and annoying. Write down where information enters, who touches it, where it gets copied, and what happens when somebody forgets. Then compare three options: use the current product better, connect the current tools, or build a narrow internal tool. The right answer is the least complicated option that reliably removes the bottleneck.