Skip to main content

8. CRM Diagnostics

Note: The video covers material not in the guide below — please watch in full.

Action Step

Complete this before moving on.

Follow along with the video — this is a workshop. Copy the BrightLoop Salesforce folder from Academy Practice Assets into your Academy Exercises folder. Feed the metadata files to Claude Code and start asking questions about what the CRM data tells you. Then have Claude create draw.io diagrams based on the metadata — pipeline architecture, data quality heat maps, object relationship maps. Explore beyond the video: try combining the metadata with your kickoff prep doc or a playbook to build a before-and-after comparison.

Comment in Slack

Post your answer in your onboarding channel.

What was your biggest takeaway(s) from this training?

Training Guide

You've been building deliverables from transcripts and playbooks — qualitative stuff. Now you're working with raw system data. CRM metadata is the layout of how everything is structured inside a CRM — objects, fields, pipeline stages, data types — without any individual contact or deal records. It's the blueprint.

What you can do with that blueprint, using Claude Code, is surprisingly powerful.

(You don't need to be a Salesforce admin to do this)

Setting Up the Exercise​

In VS Code, go to Academy Practice Assets and find the BrightLoop Salesforce folder. Right-click it, copy it, then paste it into your Academy Exercises folder — create a subfolder called something like "Salesforce Analysis."

Inside you'll find five files of mock Salesforce metadata:

  • Opportunity field usage summary — shows how often each field is actually being filled in
  • Opportunity pipeline configurations — stages, probabilities, exit criteria
  • Report exports — pipeline summaries, forecast by rep, lead source breakdowns
  • Object and field list — every object and field in their Salesforce with data types
  • Sample opportunities — real deals showing how reps actually use the system

Open a few. Don't try to absorb everything — just notice the shape of the data.

(This is where Claude earns its keep)

Feed It to Claude Code​

Copy the path to the entire Salesforce metadata folder and paste it into Claude Code. Give it context about who you are and what you're doing — something like: you're a go-to-market architect working with BrightLoop as a customer, you have their Salesforce metadata, and you want to understand what you can do with it.

Claude comes back with a structured breakdown. It identifies the field usage, pipeline stage issues, data quality problems — and it maps all of this directly to the kickoff prep work you did earlier. The qualitative pain points from the kickoff now have quantitative evidence behind them.

You can go further and ask it questions: What does the lead source report tell you? Which deals look stuck? Which fields are abandoned? What's wrong with their pipeline stages for a Series A B2B SaaS company?

(Push it when answers are too general — ask for specifics, ask for numbers)

The Pre-Kickoff Hypothesis​

Here's where this connects to the bigger picture. At LeanScale, there's a four-phase approach to implementing projects. Inside the first phase — strategy — there's a pre-kickoff sub-phase. The idea: come to the kickoff call with a 70% prefilled hypothesis of what the customer needs, based on their actual CRM data.

Before AI, this wasn't realistic. Now, within a few minutes of feeding metadata to Claude Code, you have a data-backed assessment of what's broken, what's unused, and what to fix first. Combine that with the kickoff prep doc you already built, and you're walking into the meeting with evidence instead of a blank whiteboard.

Creating Diagrams from Metadata​

You already learned how to create diagrams programmatically. Now apply that to the CRM data. Tell Claude Code to create draw.io diagrams based on the Salesforce metadata — let it decide what diagrams would be most useful for stakeholders.

It produces things like: pipeline stage architecture, data quality heat maps, rep performance comparisons, object relationship maps, lead source attribution funnels, and field cleanup roadmaps showing what to retire, fix, or keep. Each tab in the draw.io file targets a different stakeholder — the VP of Sales sees something different from the demand gen lead.

(Fork the conversation to run the diagram creation in parallel while you keep asking questions in the main session)

Managing Multiple Agents​

This training naturally involves multiple agents running at once — research agents investigating Salesforce CLI capabilities, the main agent writing files, a forked conversation building diagrams. A few things to keep in mind:

When a subagent finishes, it reports its findings back to the main agent. That report eats your main context. If you know a research agent will produce a large report, tell the main agent upfront: have the subagent write the report to a separate file and only report back that it's done. This preserves your main context window.

Once a subagent is launched, you can't update its instructions. Everything has to be scoped in the initial prompt. If you find yourself needing this pattern regularly, you can set it up in a universal prompt file that gets loaded every session.

What You Just Did​

You took raw CRM metadata — field lists, pipeline configurations, usage summaries, sample deals — and turned it into analysis, diagrams, and a diagnostic hypothesis without ever logging into Salesforce. You asked Claude Code questions about the data, combined it with context from previous trainings, and had multiple agents working in parallel to produce deliverables.

The pattern isn't limited to Salesforce. The same workflow applies to any system where you can export metadata — HubSpot, Gong, Clay, any tool with an API. Feed the data, ask questions, cross-reference against best practices, and build something a client can act on.