Building Hyper-Personalized Email Agents for Scalable Sales Campaigns at Grafana Labs
Building effective email writing agents for GTM When you care about quality, you can’t accept what comes from one-shot prompts and basic LLM workflows. When deciding how to scale sales email campaigns at Grafana Labs, there were a few important constraints:
- 1.
Our customers are engineers — can’t sound salesy.
- 2.
The products are very technical — can’t get the details wrong.
- 3.
Personalization matters — truly creative hyper-personalization.
The end-state for email personalization Personalization in the context of email has a few levels:
- 1.
Static template with variables
- 2.
Static template with a sentence and paragraph bank
- 3.
Spintax, i.e. templates rewritten with slight variation
- 4.
Lucid, agentic hyper-personalization (end state)
Let’s walk through how we built the fourth level. The System customer.io is the email automation platform. It contains a database of our marketable contacts (most of this comes from the data warehouse), enables building audience segments, it’s a workflow builder for email campaigns, and it…sends emails! n8n receives webhooks from customer.io, processes LLM-enabled workflows, and sends the generated emails back to customer.io. Workflows include email generation, formatting, writing guardrail evaluation, observability/logging, and tool-calling, such as getting personalized content recommendations to be added to the emails. Octave receives all the context and writes the emails. It understands our business and the playbook behind the campaign, and crafts hyper-personalized messaging. In a few hours, I can have an agent that weaves prospect and account research with campaign value propositions, personalized on persona and arbitrary segments (e.g. OpenTelemetry-first organizations). Slack and their dashboards are where we notify reps that a new email is about to be sent. For high-priority emails, the generated emails are given to the sales team to review, modify, and approve for sending (e.g. a director+ title in a tier 1 account). These are sent to existing customers and opt ins, so we don’t have a need for inbox warming/rotation and other components like that. Right now, this system does the work of +10 reps. And this is just for driving consumption and pipeline in our self-serve book of business. What comes next for our agentic email writing program:
Feed winning/losing emails back into the writing agents for fine-tuning.
Automatically mine email data and build new writing agent use cases.
Create email writing skills for a central agent.
Scale campaigns to post-sales teams.
We’ve proven that AI can write better than our sales and success teams. I’m always open to trading notes about this kind of work! sharing for Jordan G.!
