How to create a weekly field-insights digest agent in Demodesk AI Crew

What and why

This guide shows you how to build a custom AI agent in Demodesk AI Crew that listens to your team's external meetings, extracts the topics you care about, and sends you a structured digest on a schedule you set — for example, every Friday at 3pm.

It replaces the manual “ask AI chat after every call” workflow with one agent that runs in the background.

The agent is built using natural language. Describe what you want, connect it to your meetings data, and AI Crew sets it up. No code, no templates.

Who this is for

Sales managers, onboarding leads, and revenue leaders who run multiple teams and want a structured weekly read on what's being discussed in customer conversations — without watching every recording or writing prompts from scratch every Friday.

Prerequisites

• A Demodesk Coaching & AI seat with AI Crew access
• Recording enabled for the teams you want to summarize (sales, onboarding, customer success)
• A clear list of the topics you want surfaced (objections, competitor mentions, product feedback, deal risks)
• An email address or Slack channel where the digest should land

Steps

1. Open AI Crew from the Home sidebar — In the Demodesk app, go to Home. In the left sidebar, find the AI Crew section. This is where your custom and installed agents live.
2. Click the plus icon to add a new agent — Next to the AI Crew section header, click the + icon. This opens the Add to your Crew screen, which is the AI Agent Builder.
3. Describe the agent in natural language— In the prompt field, write what you want the agent to do. Be specific about three things: which meetings to look at, what to extract, and when to send the result.

   Example prompt for a sales digest:

   “Every Friday at 3pm, summarize all external sales meetings from the past 7 days. Focus on: objections raised, competitor mentions, feature requests, and deal risks. Group findings by topic, not by meeting. Send the summary to my email.”

   Example prompt for an onboarding digest:

   “Every Friday at 3pm, summarize all onboarding calls from the past 7 days. Focus on: common customer questions, setup blockers, feature confusion, and positive feedback. Group findings by topic. Send to my email.”

   Two separate agents — one for sales, one for onboarding — keeps the reports clean and role-specific.

4. Confirm the data source — AI Crew will ask which meetings the agent should use. Select external meetings for the relevant team (sales, onboarding, or both). If you only want recorded calls with customers, exclude internal meetings here.
5. Set the schedule and delivery channel— Confirm the run schedule (Friday at 3pm in your timezone) and pick a delivery channel: email, Slack, or Teams. You can add multiple delivery steps if you want both.
6. Review what AI Crew built— Before saving, AI Crew shows you the agent's configuration: trigger (schedule), inputs (meetings), processing (the prompt), and output (delivery). Read it once. Adjust anything that doesn't match what you described.
7. Save and run a test— Save the agent. Then run it manually against the past 7 days of meetings to see what the digest looks like. If the output is too long, too vague, or misses a topic, go back into the prompt and add specifics. Most agents need one or two refinements to hit the right level of detail.
8. Let it run on schedule— Once the test output looks right, the agent runs automatically. Your digest arrives every Friday without any manual work.

Tips

• Group by topic, not by meeting.A list of 12 meeting summaries is hard to use. A digest organized by “Objections,” “Competitor mentions,” and “Product feedback” is what you want to read.
• Be specific about the topics.“Summarize meetings” produces generic output. “Surface objections, competitor mentions, and deal risks” produces a usable report. The more specific your prompt, the better the digest.
• Run separate agents per team.One agent for sales, one for onboarding. Mixed reports lose signal. AI Crew includes 1,000 runs/month free, so multiple agents won't push you over the limit.
• Add a “what changed this week” line.Ask the agent to flag topics that came up more often than the previous week. That's where the real field intelligence is.
• Forward the digest to your team. The same report that helps you spot trends helps your reps see what their peers are running into.

Related skills and agents

• AI Crew on Demodesk — the in-app section where custom agents live
• Demodesk Marketplace — browse pre-built agents and skills, including templates for field reporting
• AI Analyst product page — the pre-built agent that surfaces deal risks, competitor mentions, and product feedback across your pipeline

FAQ

How long does it take to build an agent like this?
Five to ten minutes for the first version. Most users refine the prompt two or three times before the output is right.

Can I build separate agents for sales and onboarding?
Yes. Two role-specific agents produce cleaner, more relevant digests than one combined report.

Does the agent only summarize recorded calls?
Yes. The agent works from transcripts of meetings the Demodesk Notetaker captured. If a meeting wasn't recorded, it won't appear in the digest — one more reason to push recording adoption above the 70–80% benchmark.

Can I send the digest to Slack instead of email?
Yes. AI Crew supports multiple delivery channels including Slack and Teams.

What if my team only records audio, not video?
Audio-only recording is fully supported and produces the same transcript quality. The agent works identically whether your team records video and audio or audio only.

Is my data used to train Demodesk's AI?
No. Customer conversations and CRM data are used only to serve you. Your data never trains our AI.

How many agents can I run?
The Starter tier includes 1,000 agent runs per month free with any Coaching & AI plan. A weekly digest agent uses 4–5 runs per month, so you can run many agents in parallel.

Where can I see what an agent did?
Each agent has an execution history in AI Crew showing every run, the data it used, and the output it produced. Useful for debugging and for spotting when a run didn't complete as expected.