Skip to main content

3. RapidRatings - Build + Enablement

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

Resources

Reference material from this engagement.

Action Step

Complete this before moving on.

This is a real client build and closeout, so use it as a reference model. Watch the overview section and make sure you can explain the one repeatable pattern every agent follows: a Salesforce trigger → a Power Automate flow → the Copilot agent → a write-back to the CRM. Then trace the two-step QA (access test in Copilot Studio, then the flow test in Power Automate) end to end, and skim the closeout guide in the Resources box to see how the assumptions, open questions, and per-person requests are documented. If anything's unclear, drop it in your onboarding channel.

Comment in Slack

Post your answer in your onboarding channel.

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

Training Guide

This is the closeout for the RapidRatings engagement — part mini-course, part enablement guide. It walks the client team through every agent we built, how to edit and extend them, how to QA before promoting to production, and the open items left to finish. If you're learning, treat it as a reference model for how a real Copilot build gets handed off.

The One Repeatable Pattern​

Every agent follows the same shape, and once you see it once you see it everywhere: a Salesforce field changes, an automation fires, that automation calls a Copilot agent, and the agent reads from the CRM (and the web) and writes its answer back to Salesforce.

That breaks into three parts:

  1. Copilot Studio — the brain. Each agent has a description and instructions (the main prompt, which is where most of the QA happens), plus the tool access that lets it reach the data.
  2. Power Automate — the plumbing. A flow watches Salesforce on a trigger, calls the agent, and writes the result back.
  3. Salesforce — the fields. The output lands here, along with confirmation fields so an agent doesn't re-run on a record it already processed (which saves tokens).

The Four Agents​

Four agents were built. The Discovery agent is the proven one — it was the proof of concept, and most of that time went into nailing the service-level account and security permissions with Nate. Once that foundation was solid, the other three were built on top of it the same week, fully wired in Sandbox and awaiting the team's QA and promotion to prod.

  • BDR Messaging agent — turns a new lead into research-driven, persona-tuned outreach (cold email, follow-ups, a voicemail opener) that lives in Salesforce for the BDR to review and copy.
  • Discovery agent — turns a new lead or opportunity into a cited pre-call brief so the rep walks in informed. (Live / final QA in progress.)
  • Qualify & Develop agent — turns a stage change into a coaching brief on what evidence is missing and what to do next.
  • Prove Stage agent — turns a Prove-stage deal into a validation brief: is it sponsored, budgeted, vendor of choice, and paper-ready?

All four currently write into the Salesforce agent confirmations tab (on both the lead and opportunity objects). Now that the foundation exists, the team can re-route those outputs to Microsoft Teams or email whenever they want.

How to Make Edits​

This is the part that empowers the team to keep building after the engagement ends.

Copilot Studio. The agent's behavior is driven by the instructions prompt. To change it, paste the current prompt into Copilot, Claude, or any AI — "here's the prompt I'm using, here's how I want to tweak it, what do you recommend?" — and iterate. The description doesn't affect behavior, but keep it accurate. Critical: after any change, click Publish → Force the newest version, or the Power Automate flow won't pick it up.

Power Automate. The flow is made of nodes, just like n8n, Make, or Zapier. The first node is the trigger (object + filter query). The second is the execute-and-wait Copilot node, where you select the published agent and add a short precursor message plus dynamic tokens from the previous step. The last node updates the record — match the record ID from the trigger and map the fields to write. Connection type (service-level accounts, prod vs. sandbox) is Nate's area of expertise.

Salesforce. Outputs live under agent confirmations on the lead and opportunity objects — and again, you can redirect them elsewhere later.

The Two-Step QA​

There are two places to QA, and the order matters.

  1. Access test (Copilot Studio). On the right side, ask the agent "do you have access to Salesforce?" and probe whether it can see specific leads and opportunities. If it fails access here, it'll fail at the Power Automate level too — so this gate comes first.
  2. Flow test (Power Automate). Put the flow in Test → Manually, then go to Salesforce and change a record to match the trigger criteria (move the stage, flip a checkbox, or reassign the owner). Watch it run, open the successful runs to inspect the output, and hand the raw output to Claude if you need help debugging.

After both pass, the real work is output QA: is it structured right, in the right tone and format, and actually useful for whoever receives it? If not, fix the instructions prompt and re-run.

Gong Integration (Coming Soon)​

Gong released its MCP this month and is rolling it out across client instances, so RapidRatings should get access soon. Once Gong is connected in Copilot Studio, every agent gets meaningfully smarter — it can read Gong transcripts alongside Salesforce. The closeout guide already contains Gong-enabled prompt variations for all four agents; when Nate adds the connection, the team swaps the Salesforce-only instructions for the Gong versions.

Open Items & Next Steps​

The Discovery agent's design is confirmed with Crystal and Hooman. The other three were wired in Sandbox based on educated assumptions about the existing Salesforce setup — so each has a "for the team, please confirm" note in the guide.

  • For Nate — a Sandbox connection issue is hiding some newly created fields from the flow. The BDR Messaging agent is missing both its write field and its checkbox; Qualify & Develop and Prove Stage are missing checkboxes. These need the connection/field access fixed.
  • For Crystal — approve the new Salesforce fields (or propose a different architecture; the alternatives we evaluated are documented).
  • For the team — run the two-step QA on all agents, then promote to production. Connect Gong later as a future enhancement and swap in the Gong prompts.

Every architectural assumption, open question, and alternative is written into the closeout guide so the team can own and extend this after the engagement ends.