Agentic AI CMT
Agentic AI CMT
Compliance | Workflow design
Compliance | Workflow design
6 weeks
6 weeks
AlphaSights
AlphaSights
problem.
problem.
The client management tool gave the team everything in one place, but the hardest part of the workflow was still manual: figuring out which clients needed outreach, why, and what to say. We wanted to use agentic AI to speed this up without replacing the critical thinking that makes outreach personal.
The client management tool gave the team everything in one place, but the hardest part of the workflow was still manual: figuring out which clients needed outreach, why, and what to say. We wanted to use agentic AI to speed this up without replacing the critical thinking that makes outreach personal.
constraints.
constraints.
AI came into the picture after the MVP had already shipped, so whatever we built had to sit inside an existing product without breaking workflows people had already adopted. We also had to be deliberate about where AI decided and where it suggested, the client services team needed to stay in control of the relationship.
AI came into the picture after the MVP had already shipped, so whatever we built had to sit inside an existing product without breaking workflows people had already adopted. We also had to be deliberate about where AI decided and where it suggested, the client services team needed to stay in control of the relationship.
in-chat widgets.
in-chat widgets.
Part of the challenge was designing a chat experience that could respond to the task at hand, rather than defaulting to plain text for every interaction. That meant designing the UI for a set of widgets suited to different use cases.
widget 01. client relevance
widget 01. client relevance
Comparison called for a widget over plain text - it's easier for the eye to compare clients or outreach options side by side than to parse it from a paragraph.
Comparison called for a widget over plain text - it's easier for the eye to compare clients or outreach options side by side than to parse it from a paragraph.
widget 02. research comparison
widget 02. research comparison
Surfacing research needed a similar comparison view, but also had to distinguish how that research would actually appear in the outreach. The primary action is copying it in, with a secondary option to open the full piece in the CMT.
Surfacing research needed a similar comparison view, but also had to distinguish how that research would actually appear in the outreach. The primary action is copying it in, with a secondary option to open the full piece in the CMT.
widget 03. email draft
widget 03. email draft
Drafted emails were the clearest case for a widget. Users needed to review what the AI had written before sending, so the primary action is sending directly from the chat rather than switching context to do it elsewhere.
Drafted emails were the clearest case for a widget. Users needed to review what the AI had written before sending, so the primary action is sending directly from the chat rather than switching context to do it elsewhere.
reflection.
reflection.
Retrofitting AI into an existing flow is a different design challenge to building with it from the start. Some parts of the tool were designed before AI was a factor, and balancing what we'd already shipped with what AI could now do was harder than I expected. Next time I'd push to define where AI sits in the experience before the first release, even if the capability isn't there yet, so the architecture leaves room for it.
Retrofitting AI into an existing flow is a different design challenge to building with it from the start. Some parts of the tool were designed before AI was a factor, and balancing what we'd already shipped with what AI could now do was harder than I expected. Next time I'd push to define where AI sits in the experience before the first release, even if the capability isn't there yet, so the architecture leaves room for it.