When the CRM Is a Spreadsheet, an Inbox, and Someone's Memory

Most sales teams have one system of record on paper and another in practice. The real state of a deal lives in a rep's mailbox, the quote lives in a folder, the history lives in a spreadsheet somebody has kept for years, and the CRM gets updated after the fact — if at all. A generic CRM doesn't fix that. It was built for the average sales process, not yours.

I build the alternative: a custom CRM and sales platform shaped around how your team actually sells. It's for teams that have outgrown a generic CRM, or that run sales out of spreadsheets and inboxes. The pipeline uses your stages and your rules. Mail and meetings land on the deal they belong to. AI does the reading and the drafting; people make the calls.

The proof is Breadcrumb — a platform I built for a client's sales team that started as market intelligence and lead management and became the team's system of record for deals. What's described on this page comes from that build. See the Breadcrumb case study.

Pipeline board showing the Leads lane with Identified, Working, and Engaged columns, each headed by its move-on rule, and cards showing rank, days in stage, off-track and open-commitment flags, and the next task with its due date. Company names, people, task text, and dollar values blurred.
// The pipeline board in Breadcrumb — every card has a next task, and every column states its move-on rule. (Company names, people, task text, and dollar values blurred.)

What Gets Built

Six subsystems around one deal record — each one built to your process, not a vendor's.

Pipeline

Owner, Stage, Next Task

Every deal has an owner, a stage, and one open next task. Stages have written exit rules, a board view shows the whole pipeline, and deals that aren't ready can be parked for nurture.

Mail

Threads Filed onto Deals

Your team's Microsoft 365 mailboxes are read through Microsoft Graph, with organization-level access granted by your company's admin. Threads are filed onto the right deal, and reps send from their own mailbox from inside the CRM.

Meetings

Transcripts That Do Something

After a Microsoft Teams meeting, the transcript is filed on the deal, facts are extracted, and a recap is drafted for the rep to confirm. In Breadcrumb, a follow-up task proposed from the call is still running in shadow mode.

Deal Brief

An AI Brief with Sources

Each deal carries an AI brief built from a sourced fact store — mail, documents, and transcripts. Reps can correct or pin a fact when they know better.

Quotes

Template-Driven Quotes

Quote documents are generated from the deal's own data through your templates, with revisions kept. The template writes the quote, not a language model.

Leadership

The Weekly Overview

A weekly pipeline overview and reports for sales leadership, drawn from the same deal records the reps work in.

A single deal page showing its stage on a stepper running from lead stages through deal stages to post-sale, the open follow-up task, and email threads filed onto the deal with a reply-waiting flag. Company, people, subjects, message text, and deal value blurred.
// One deal — its stage on the path from lead to active customer, the open next task, and the mail threads filed onto it, with a flag on the reply that has been waiting. (Company, people, subjects, message text, and deal value blurred.)

Around the Pipeline

  • Tasks and a Today queue. Each rep gets a daily queue. Tasks are also proposed from promises made in mail, and a person accepts them; in Breadcrumb, doing the same for promises made on calls is still running in shadow mode.
  • Per-person permissions. Enforced fail-closed in one place. Outside reps see only their own territory.
  • History kept. In Breadcrumb, history from the team's previous CRM and a long-running spreadsheet register was imported and kept as searchable history.
  • Guided walkthroughs. In-app walkthroughs show people how a page works, inside the page itself.
  • A change-request loop. Requests are raised inside the app, and each shipped change gets a plain-English note.

Works from Claude

The CRM comes with a Claude connector (MCP), so reps can work it from Claude on web, desktop, or phone. In Breadcrumb the connector exposes 35+ tools, read and write, covering the routines a rep runs every day: a morning brief, call prep, call wrap-up, mail sorting, and account review.

  • Signed in as themselves. Each person connects with their own sign-in (OAuth), so Claude acts as that person and nobody else.
  • The CRM's own checks. Every tool call first passes the CRM's own permission check for the page it stands in for, and every write runs through the same code as the web UI.
  • Attributed and logged. Changes are attributed to the person who made them, and logged.
  • Email is draft-only. Claude can draft a message; the draft waits for the rep to approve and send inside the CRM.
  • Approval before writes. Requiring approval before each write is a setting your organization turns on in Claude.

AI Proposes. A Person Decides.

A CRM holds the record your team is accountable for, so the AI in it is engineered to stay in its lane.

  • Proposals, not actions. Tasks are proposed for a person to accept, recaps are drafted for the rep to confirm, and facts on a brief can be corrected or pinned.
  • Shadow mode first. Most new AI behaviors first run in shadow mode — proposing, not acting — before they are switched on. In Breadcrumb, several are still in shadow.
  • Customer mail is never auto-sent. A person approves every customer-facing send, whether it was drafted in the CRM or from Claude.
  • One set of permissions. Every Claude connector tool call passes the CRM's own permission check.
Deal chat and AI brief side by side: the chat answers from the record and states what is not on record, and the brief lists what is still needed from the customer and from the team to finish the stage. The question asked, one line of the answer, and three checklist items blurred.
// The deal chat and AI brief in Breadcrumb — the chat answers from the record and says so when nothing is on record; the brief lists what the stage still needs, and from whom. (Deal-specific lines and three checklist items blurred.)

The Other Half: Finding the Deals

Breadcrumb is one platform with two halves. The first finds and ranks the market — every company, plant, and decision-maker, tiered against an ideal customer profile. That is Market Intelligence & Custom Lead Management. This page is the second half: what happens once a lead becomes a deal. The build is documented in the Breadcrumb case study.

Tools & Technologies

Chosen per project, never per habit. Common building blocks include:

  • Microsoft 365
  • Microsoft Graph
  • Microsoft Teams
  • Claude connector (MCP)
  • OAuth sign-in
  • Python

How Projects Work

  1. Survey — map how deals actually move today: the stages people really use, where mail and quotes live, what leadership asks for every week.
  2. Draft — a drawing of the system: the pipeline and its exit rules, who sees what, what AI proposes, what stays with people.
  3. Build — the pipeline, the mail and meeting filing, quotes, permissions, and the connector, with an automated test suite alongside.
  4. Prove — bring the history across and run new AI behaviors in shadow mode before they are switched on.
  5. Operate — ship it, run it, and keep changing it through the change-request loop as the team's process moves.

FAQs

When does a custom CRM make sense instead of an off-the-shelf one?

When the team has outgrown a generic CRM, or never really moved into one — deals live in spreadsheets, inboxes, and memory, and the CRM is something people update after the fact. A custom CRM is built around the stages, rules, quotes, and reports your team already uses, so working a deal and keeping the record are the same act. If an off-the-shelf CRM fits how you sell, I'll say so.

Does it connect to our email and meetings?

For Microsoft 365 and Teams, yes — that is what the Breadcrumb build runs on. The team's Microsoft 365 mailboxes are read through Microsoft Graph with organization-level access granted by the company's admin. Threads are filed onto the right deal, and reps send from their own mailbox from inside the CRM. For Microsoft Teams meetings, the transcript is filed on the deal, facts are extracted, and a recap is drafted for the rep to confirm; a follow-up task proposed from the call is still in shadow mode, not yet switched on.

Will the AI send email or change deals on its own?

Customer-facing mail is never auto-sent — a person approves every send. The rule across the system is that AI proposes and a person decides: tasks are proposed for a person to accept, recaps are drafted for the rep to confirm, and facts on a deal brief can be corrected or pinned. Most new AI behaviors first run in shadow mode — proposing, not acting — before they are switched on.

What can my team do from Claude?

In Breadcrumb, the Claude connector exposes 35+ tools, read and write, so a rep can run a morning brief, prep for a call, wrap one up, sort mail, or review an account from Claude on web, desktop, or phone. Each person signs in as themselves, every tool call passes the CRM's own permission check for the page it stands in for, every write runs through the same code as the web UI, and changes are attributed to that person and logged. Email from Claude is draft-only — drafts wait for the rep to approve and send inside the CRM. Approval before each write is a setting your organization turns on in Claude.

What happens to the history in our current CRM and spreadsheets?

It can come with you. In the Breadcrumb build, history from the team's previous CRM and a long-running spreadsheet register was imported and kept as searchable history, so the record of past work sits alongside the live pipeline.

Who can see and change what?

Permissions are set per person and enforced fail-closed in one place, so anything not explicitly allowed is refused. Outside reps see only their own territory, and sales leadership gets a weekly pipeline overview and reports.

Where Do Your Deals Actually Live?

Tell me how your team sells today — the CRM, the spreadsheet, the inboxes — and where deals get lost. I'll map a system built around that process, and mark what AI should and should not touch.