Skip to content
All work

NRC Health

An agentic copilot for inside sales

A Microsoft Copilot Studio agent that researched health systems and drafted outreach, with a rep reviewing every message before it went out.
  • Microsoft Copilot Studio
  • LinkedIn Sales Navigator
  • Salesforce

Summary

Challenge
Health-system outreach only works when it shows you understand the system you are writing to, and that research takes longer than the outreach itself.
What I built
An agent in Microsoft Copilot Studio, grounded in a knowledge library I maintained and run through written workflows: research the account, match the solution, draft the message.
Result
About a quarter of my work week back from prospecting research, in a tool the inside sales team used too.
Role
Designed and built the agent; maintained its knowledge library
With
The inside sales team, who used it
Status
Ran in production at NRC Health. An agent with written workflows, not a trained model

At NRC Health, before Anova. Pipeline results from the same role are listed separately, further down.

The situation

Research took longer than the outreach.

NRC Health sells experience and analytics software to health systems. The buying committees are large, procurement runs long, and a first conversation with an executive only happens if the outreach shows you understand the system you are writing to.

Inside sales covered a national book of health systems. Every good conversation started with research: service lines, recent announcements, leadership changes, and which solution fit. That research takes longer than the outreach itself, which forces a choice between generic messages at volume and a handful of accounts done properly.

The decision

Automate the research. Keep a person on every message.

The research was repeatable; the relationship was not. I decided to hand the research and the first draft to an agent, ground it in what we actually sold, and keep a rep reviewing every message before it went out.

I built it for my own book first, where I could judge whether the output was good enough to send, before anyone else relied on it.

What I built

An agent with written workflows.

Grounded in our own positioning, run step by step, reviewed by a person.

  • An agent in Copilot Studio

    I built a custom agent in Microsoft Copilot Studio, grounded in a knowledge library I maintained: our products, positioning, buyer language and the problems each solution addressed.

  • Step-by-step workflows

    Instead of one open prompt, the agent ran written workflows: research the account, match it to the solution that fit, then draft the outreach. The steps made the output repeatable from one account to the next.

  • Sources a rep would use

    Account intelligence came from the open web and LinkedIn Sales Navigator: service lines, announcements, leadership changes and the people involved. Each brief came back with its sources.

  • A prioritized book

    Segmentation ranked the book by fit, so the highest-fit health systems surfaced first and every draft arrived already matched to the solution that fit the account.

The workflow

Research, match, draft, then a person decides.

Illustrative The agent does the repeatable part, research, a solution match and a first draft, and stops there. Sources come back with every brief, and a rep decides what is sent.

Guardrails

What the agent was not allowed to do.

  • Send anything. The agent drafted; a rep decided what went out.
  • Hide its sources. Briefs came back with the sources behind them, so a rep could check a claim before using it.
  • Pitch from memory. Product and positioning came from the maintained knowledge library, not from the model's general knowledge.

How it ran

Built for one book, used by the team.

I built the copilot for my own prospecting, and the inside sales team used it too. For me it gave back about a quarter of my work week that had gone to prospecting research, and that time went into conversations.

Result

About a quarter of my work week back from prospecting research.

The agent handled the repeatable research and the first draft. A rep kept every decision about what was sent, and the time it gave back went into conversations.

Also at NRC Health

Results from the same role.

These come from the same two years at NRC Health. They are listed apart from the copilot because they are not attributed to it.

  • More than $4.5M in qualified pipeline across enterprise and new-logo accounts.
  • The year's largest new-business deal. I sourced the first executive meeting through direct outreach and co-developed the partnership proposal. I did not own the close.
  • Inbound answered within 24 hours. I mapped every inbound source to an owner and a response path, set a 24-hour standard and built the routing behind it: assignment on territory and fit, first-touch templates, and reporting that made a missed response visible.
  • AI in the curriculum. I advised the Maine College of Health Professions executive board on AI in the curriculum.

Takeaways

What carries over.

  • Ground an agent in your own positioning before pointing it at prospects. A generic model writes generic outreach.
  • Write the workflow down. Repeatable steps beat a clever prompt.
  • Measure the time given back, then spend it on conversations.

The stack

What it ran on.

AI

  • Microsoft Copilot Studio
  • Maintained knowledge library
  • Workflow design

Research

  • LinkedIn Sales Navigator
  • Open-web research

CRM and outreach

  • Salesforce
  • Executive email and social outreach

Open to senior and manager-level roles

Need the demand engine built, or someone to run it?

Demand generation, marketing operations and RevOps roles in complex B2B, remote or hybrid from Omaha. I also take on a small number of scoped builds.

Start a conversation Download résumé