Migrating from AI Agent Personality to Post-Chat Survey.
A comprehensive UX redesign that replaced personality configuration with a dedicated Post-Chat Survey system, enabling teams to collect structured feedback from customer interactions while improving the AI Agent's behavior through data-driven insights.
I led the migration strategy from the legacy Postchat Survey plugin to a scalable GenAI-based solution — defining the approach, analyzing customer data, and coordinating across engineering and TAM to deliver a full, single-cutover migration with no customer effort required.
Context
Part of a company-wide re-architecture to meet Enterprise requirements, this meant deprecating Zowie's legacy bot-script layer. The Post-Chat Survey plugin was one of several products still running on it. An engineer brought the problem to me — not as a design brief, but as a constraint: migrate all customers without it costing them time or effort. The scope of the solution was mine to define.
Before any design: understanding what existed
I started in BigQuery, analyzing how customers had configured the legacy plugin across all environments. Most were using dedicated CSAT and NPS question types correctly. The notable exception was a subset who had manually replicated CSAT and NPS scales using single-choice questions — those mapped cleanly to the new data types anyway. The data confirmed a single automated cutover was viable. The only real edge case: one or two customers had built fully custom responses using emoji and bespoke labels — something the new GenAI solution, which has no button UI, couldn't replicate.
The decision that kept the system clean
There was a genuine question on the table: build button support into the new system to handle this? It would have solved the edge case but deviated from the GenAI solution's design principles. With only one or two customers affected, I proposed a different path — TAM would reach out directly and guide them toward standardized formats. Both customers agreed. The system stayed architecturally clean.
Handoff
For the backend engineer: a BigQuery analysis in Markdown broken down by environment and usage, a mapping document translating legacy bot-script responses to Zowie Engine equivalents, and UI designs for the customer-facing migration prompt.
For TAM: a direct call with the responsible account manager — enough context to handle the client conversation confidently without over-engineering the process.
