AI That Works Like a Trusted Partner — Not a Black Box
Overview
Relationship Managers don’t struggle because they lack data. They struggle because they have too much of it spread across screens, reports, and documents, while every decision still rests on their shoulders. Acting fast is critical, but acting wrong is costly.
This is where the Agentic AI experience comes in.
Designing AI That Works Quietly for RMs
When introducing AI into the CRM, we made a clear decision early on: Relationship Managers should focus on outcomes, not mechanics. While AI can be powerful, exposing its complexity to RMs would slow them down rather than help them. The experience needed to feel effortless and trustworthy without asking RMs to understand how the system works behind the scenes.
ROLE
Product Designer
RESPONSIBILITIES
I led the design of an agent-based AI framework where Admins own configuration and training, and RMs only consume the results. Through the Agentic Dashboard, Admin users can create, train, and manage AI agents by defining topics, instructions and data sources. This ensures consistency, governance, and compliance across the organization.
Admins are guided through best practices while creating topics, so agents behave predictably and align with business rules. Before activation, Admins can preview and test agent behavior to validate accuracy and relevance.
COLLABRATION
I collaborated with a lean, cross-functional team that included the VP, Product Manager, Data teams and 9+ Engineers. I led end-to-end design execution, working closely with engineering and product teams to align on requirements, validate solutions, and deliver the feature efficiently in a fast-paced, high-impact environment. This involved defining role-based access, agent permissions, and lifecycle states. Rapid prototyping helped validate that Admin complexity did not leak into the RM experience, while still giving Admins enough control to manage AI responsibly at scale.
IMPACT
This approach allowed AI to scale safely across the CRM. Admins gained control and governance, while RMs gained speed and confidence. Decision-making became faster, errors reduced, and AI adoption increased not because users learned AI, but because AI stayed out of their way.
TIMELINE
2 Months
A Day in the RM’s Workflow
An RM opens the BusinessNext CRM at the start of the day. Instead of scanning dashboards and jumping between modules, they land on a Worknext Panel, a place where AI agents are already prepared to assist with real work.
Each agent has a clear purpose: summarising customer health, flagging risks, preparing follow-ups, or recommending next best actions. Nothing feels vague or magical. Every agent exists for a reason the RM understands.
Training AI the Right Way
Agents are trained using data & topics, which act as structured instructions rather than open-ended prompts.
The system guides users on:
How to write effective topics
How to add clear instructions
How to choose the right data sources
What the knowledge resourse is?
Inline guidance and best practices help prevent vague inputs, ensuring agents behave consistently and predictably.
Preview Before You Trust
Before an agent goes live, admin can preview and test its behavior. They can see how the agent processes sample input, which data it uses, and what output it generates before it impacts real workflows.
This step was critical in building trust and reducing hesitation around AI usage.
Few of the many Use Cases
By the time the experience came together, the impact was clear. RMs spent less time scanning data and more time acting on it. AI outputs felt reliable because they were explainable and, most importantly, RMs stayed in control, supported by AI, not overridden by it. Few use cases listed below will show case the outcome on CRM platform:
The CRM redesign was rolled out in phased releases to reduce risk and ensure operational continuity for relationship managers.
Started with high-impact workflows (Summary Dashboard, Lead Creation, Customer 360, Task Management, Appointmnet)
Conducted internal UAT sessions with QA & dev team
AB testing for modules
Gathered structured feedback through walkthrough sessions and usage analytics
Iterated quickly before full-scale rollout
Creating a new lead for the exciting cutomer and sending out personalised email and review with AI edits.
What RM's were able to achive on Customer 360:
See the overview of the customer, service requests and intraction insights.
Craft personalised solutions for the customer based on the insights.
Review and sent out emails.
Create a new lead for the exsiting customer.
Wireframes were validated early with stakeholders to align on flow before visual design.
Outreach agent for sending multiple emails, SMS and WhatsApp.
What RM's were able to achive with this agent:
Sending multiple Emails, SMS and WhatsApp.
Through agent monitor they can monitor the list and check review.
Wireframes were validated early with stakeholders to align on flow before visual design.
Results & Impact
This project reinforced a key belief I carry forward: enterprise AI succeeds when users can see, guide, and trust it. Designing for explainability and confidence mattered far more than adding intelligence alone — and it reshaped how AI features are now approached across the platform.