Skip to content
Back to work

AI Outreach

Designing how AI learns product context and helps users shape personalized outreach.

Jeeva AI · AI sales assistant · Product Design

Jeeva’s Personalization step with the Refine with AI panel open: the instruction “Make this email shorter and more direct.”, quick actions, Generate and an Output Preview showing the generated version.

The context behind every message

Outreach campaigns in Jeeva could be personalized in three ways: from a template with dynamic fields, by hand, or with AI-crafted personalization, where Jeeva wrote a unique message for each lead in the campaign.

To write those messages, Jeeva needed context about the business behind the campaign: what it sold, who it was for, the problems it solved and the proof points behind it. Users described all of this in Product Info.

The Product Info step of a campaign, filled in by hand: company description, product, call to action, pain points, value proposition, proof points, a successful email sample and excluded words.
Product context gave Jeeva the information it needed to generate outreach.

Product context from a website

The product introduced a second way to create that context: from the company’s website. I designed the experience, and it shipped.

Jeeva analyzed the site and filled the same structured Product Info fields wherever it could. The result stayed editable, so users could review and correct it before it shaped any messages, and manual entry remained available.

A website could become structured, editable product context.

Steering every message at once

With AI-crafted personalization, each step of a sequence produced a different email for every lead, while the editor showed one of them as a representative sample.

When the direction wasn’t right, the main control was regeneration, which meant starting over. The product needed finer control, without asking anyone to edit each lead’s message individually.

The Personalization step before refinement: the email sequence, AI-crafted method selected, and one lead’s generated email shown as a sample.
One representative email stood in for a message generated for every lead.

Three ways to guide the AI

Refine with AI shipped with three ways to guide generation, each built on a separate underlying mechanism, and only one could be applied at a time. I designed how they worked together: one tab for each, a single question per tab, and the same generate, compare and apply loop at the end of all three.

Diagram: a sample email, then a refinement applied through Instructions, AI Search or AI Enrichment, then the step’s generation updated, then unique emails for each lead.

The preview showed one representative email. Applying a refinement changed how that step generated messages for every lead.

Instructions: guidance in the user’s own words, with quick actions for common edits.
Refine with AI on AI Search: a request to research the lead’s company and reference the most relevant update, suggested research directions, and an Output Preview using what was found.
AI Search: outside information Jeeva researches and works into the message.
Refine with AI on AI Enrichment: lead, company and behavioral fields, with Job Title and Seniority selected, and an Output Preview personalized with them.
AI Enrichment: structured lead and company data the user chooses to personalize with.

Users could change how Jeeva wrote for an entire step by reviewing one example, not editing every message.