ADSGENCY AI X DATA MANAGEMENT

Enabling teams to reuse the assets they create instead of rebuilding them by turning one flat shared folder into a rule based library that shows each role only what belongs to them

80% +

80% +

Returning customer retention

Returning customer retention

2X

2X

Design to launch speed

Design to launch speed

100%

100%

Client and stakeholder success

Client and stakeholder success

Role

Founding product designer

Founding product designer

Timeline

2025- Present

2025- Present

Team

Design, PM, Engineering

Design, PM, Engineering

Deliverables

Flows, Claude code mockups, Component Specs, Handoff

Flows, Claude code mockups, Component Specs, Handoff

PROBLEM

PROBLEM

Digital assets had nowhere to live, and no rules for who could reach them.

Digital assets had nowhere to live, and no rules for who could reach them.

AdsGency generated ad creative in minutes and then the flow ended. There was no library to return to, so teams rebuilt work they had already made. Building that library surfaced a second problem underneath it

Four kinds of user groups reach into the same library, all different user flows

Mapping how each user group interacts with an asset, the four collapsed into two. The difference that matters is not job title, it is whether you use the library or govern it.

SOLUTION + OPPORTUNITY

SOLUTION + OPPORTUNITY

Design for one library, branch it at permissions and governance

Auditing the two profiles component by component, the parity was near total. Same company profile, same integrations grid, same team list, same asset library, same four tab detail rail. What changes is not what exists but what you can act on.

Three card components parity

One IA and one component set serve both profiles, built and maintained once

Three card components parity

One IA and one component set serve both profiles, built and maintained once

Permission as state, not subtraction

Controls degrade to read-only badges instead of vanishing and leaving holes in the layout

Permission as state, not subtraction

Controls degrade to read-only badges instead of vanishing and leaving holes in the layout

Pull line, own row

Standard is read plus light metadata. Admin is standard plus everything that governs it

Pull line, own row

Standard is read plus light metadata. Admin is standard plus everything that governs it

CORE FLOWS

CORE FLOWS

Same library, two permission states.

Same library, two permission states.

Both profiles land on the same page, run the same search and filters, and open the same asset drawer. Walking standard first and admin second makes the layering visible, because everything in the second pass is an addition rather than a replacement.

Standard view, the asset manager

Standard view, the asset manager

View only, with one honest wall. Integrations and team are read-only, and billing, roles, and security collapse into a single locked card naming the owner.

Standard view, the profile

Standard view, the profile

View only, with one honest wall. Integrations and team are read-only, and billing, roles, and security collapse into a single locked card naming the owner.

Admin view, the profile

Admin view, the profile

View only, with one honest wall. Integrations and team are read-only, and billing, roles, and security collapse into a single locked card naming the owner.

Admin view: same library, extra components

Admin view: same library, extra components

Upload assets

Admins get an upload button in the library toolbar that opens a modal, dropping files into the same auto-tagging pipeline generated assets already flow through

Governance panel

An entire section above the library, admin-only, holding three tabs: users and permissions, rights management, and version oversight.

DESIGN PROCESS

DESIGN PROCESS

Mapping which feature interactions belong to which profile, and which controls belong on which asset, then drawing low fi screens for user profiles.

Mapping which feature interactions belong to which profile, and which controls belong on which asset, then drawing low fi screens for user profiles.

Segmentation lives in two layers: what a profile can do and what controls an asset carries. A screen is just where the two meet. Claude held the matrix steady and caught combinations I'd have missed.

Mapping out user entry points, mental models via research and competitive analysis

Mapping out user entry points, mental models via research and competitive analysis

We traced how assets arrive, by hand or auto-generated, and mapped each entry point to how people expect to find an asset later, not how it technically got in.

INFORMATION ARCHITECTURE + ITERATION AND TESTING

INFORMATION ARCHITECTURE + ITERATION AND TESTING

Upon iteration, we found governance was scattered across the product, so the controls were consolidated into one panel at the top of the admin view.

Upon iteration, we found governance was scattered across the product, so the controls were consolidated into one panel at the top of the admin view.

PostHog data analytics, google analytics and 1:1 stakeholder conversations from the MVP build showed admins bouncing between the grid, drawer, and profile for one governance task, so the controls were consolidated into a single panel at the top of the admin view.

Figma MCP to Claude Code

Figma MCP to Claude Code

I mapped the IA in our existing asset-management design system, fed Claude that context and custom skills through Figma MCP, let it build the foundation while I iterated, then added new views in parallel

I mapped the IA in our existing asset-management design system, fed Claude that context and custom skills through Figma MCP, let it build the foundation while I iterated, then added new views in parallel

Claude Code ran the build loop, turning each round of testing into working screens: the asset detail popup, list-view inline edits, user-profile colors matched to the design system, and countless micro-interactions, all iterated live

IMPACT

IMPACT

The asset library lifted returning-customer retention from 35% to 80%, making the ad-generation product stickier

The asset library lifted returning-customer retention from 35% to 80%, making the ad-generation product stickier

RETROSPECTIVE

RETROSPECTIVE

My biggest learning here was Adapting to new tools, building features fast with Claude in an ambiguous, fast-paced environment

My biggest learning here was Adapting to new tools, building features fast with Claude in an ambiguous, fast-paced environment

Working solo as designer, builder, and reviewer, I moved fast through ambiguity with Claude. The risk was bias: mistaking my own intent for clarity. Internal and client testing kept my assumptions honest.

Make Claude reason before it builds

Asking it to explain each decision surfaced flawed logic early, before it was built into screens

Make Claude reason before it builds

Asking it to explain each decision surfaced flawed logic early, before it was built into screens

Wearing multiple hats through handoff

I learned handoff gets smoother when engineers receive working components, not specs to interpret.

Wearing multiple hats through handoff

I learned handoff gets smoother when engineers receive working components, not specs to interpret.

Stakeholder management

Balancing client conversations with design work sharpened how I communicate and align stakeholders


Stakeholder management

Balancing client conversations with design work sharpened how I communicate and align stakeholders


Designed and built with purpose | © 2026 Divya Mavinkurve. All rights reserved.

Designing and built with purpose

© 2025 by Divya Mavinkurve