Client project
EnrollIQ
An AI assisted workflow inside Zoho CRM that helps enrollment advisors review learner information and recommendations in the workspace they already use.
Problem and users
The users were enrollment advisors on a sales and enrollment team. When a learner inquiry comes in, an advisor has to work out how interested the learner is, which courses are relevant, and which payment options may suit them.
Historical learner information can help with each of those questions, but only if it reaches the advisor while they are handling the inquiry.
What I did
I worked with the client’s sales and enrollment team. Before designing the solution, I spoke directly with the team to understand their workflow, their pain points, and what they needed from a tool.
That discovery set the shape of the project. Advisors already worked in Zoho CRM, so the product was built to support them there instead of asking them to adopt a separate application.
My contribution covered:
- Discovery with the sales and enrollment team, and the roadmap that followed from it
- The Python backend
- The React interface
- The OpenAI workflow that uses retrieval augmented generation over learner information
I helped build this workflow as part of the project team, and the list above is my part of it.
How it works
- Learner inquiryAn advisor works a learner inquiry in Zoho CRM, as they already did.
- Learner contextThe Python backend gathers the learner's information and retrieves relevant historical learner records.
- RecommendationsAn OpenAI workflow uses the retrieved context to prepare recommendations on interest, relevant courses, and payment options.
- Advisor reviewThe recommendations appear in the advisor's workspace, where the advisor reviews them and decides how to respond.
The recommendations support the advisor’s judgment rather than replace it. The advisor decides how to respond to the learner.
Decision and tradeoff
Work inside the CRM instead of beside it. Discovery made it clear that advisors spent their day in Zoho CRM. A separate application would have asked them to switch tools and carry context between them. Building into the CRM kept recommendations next to the learner record the advisor was already reading.
The tradeoff is that the design has to fit the structure of the CRM rather than a blank screen.
Evaluation and limitations
This page does not report performance figures, because I do not have measurement details I can document here. It describes the workflow and my role in it.
The client code is not public, so there is no repository link.