QBall CRM

Customer relationship management software in capital markets has a structural problem that predates AI. The system that is supposed to capture client intelligence is also the system nobody wants to feed, so coverage notes go stale, engagement history lives in inboxes, and the record of what a client actually cares about ends up in the head of whoever last spoke to them. Then generative AI arrived, and the industry's answer was to put a chat window next to all of it.

That answer is comfortable because it is easy to demo and easy to buy. It also asks the user to do the hardest part of the job, which is knowing what to ask. A sell-side analyst about to publish does not want to interrogate a chatbot about which accounts to send it to. They want the system to have already noticed. As the repository puts it:

The role of AI in an AI-native CRM is not that of a chatbot. The chat interface is essential, but secondary. Instead, AI is deeply embedded into user workflows, often enhancing and augmenting them with critical insights.

The second problem is that "capital markets CRM" is three different products wearing one badge. A research analyst, a sales trader, and a mergers and acquisitions banker share a client universe and share almost nothing else about how they work. Generic tooling averages those three jobs into a shape that fits none of them.

Three sell-side personas inside an investment bank, each with a distinct primary workflow, all covering an overlapping client base.

The prototype makes a specific architectural argument: an embedded agent needs two interaction surfaces, and they need to coexist rather than compete.

The first surface is a command bar, persistent at the bottom of every view and focusable with a keystroke. It expands into a conversational panel for intent-driven requests, and its outputs are built to be acted on rather than read. A list comes back as interactive tiles. A drafted note comes back as an editable block with send and edit-in-full-view actions attached. This is the user-initiated surface, and it is deliberately not the centrepiece.

The second surface is inline, and it is where the argument actually lives. Qubie appears inside the structured interface as annotations on contact rows, nudge badges on account tiles, stage-gate offers when a deal card is dragged between columns, and morning-brief summary cards at the top of boards. All of it is pre-computed insight paired with a concrete offer to act, placed where the user was already looking. This is the system-initiated surface, and it does the work the chat window would otherwise offload onto the user.

Both surfaces resolve into the same structured product interface, which is the part that keeps the concept honest. The agent augments a real application rather than replacing it with a text box.

On top of that sit the three persona views. Equity research covers contact engagement scoring, Institutional Investor vote campaigns, and report distribution. Sales and trading covers account coverage tiles, consumption analytics, and revenue attribution. Investment banking covers a deal pipeline kanban with stage-gate assistance and pitch deck generation. A floating segmented control switches between them, and it sits deliberately outside the product interface because it exists for navigating the demo rather than as a feature of the thing being demoed.

The whole prototype is a single self-contained HTML file generated through Claude Design, with no build step, no backend, and no dependencies, built on the QBall design system for tokens, components, and type. Both themes ship: Vibrant for light, Midnight for dark. The interaction patterns that matter are real and clickable, including drag-to-stage, audience assembly, and pitch generation.

The showcase is live and walkable at ashaykubal.github.io/QBall-CRM, with v1 from May 2026 preserved at its own path so the design evolution stays inspectable. It does the one job a prototype showcase exists to do: it makes a contested position concrete enough to argue with in a stakeholder review, across three personas rather than one flattering happy path.

Being precise about the scope matters as much as the argument. This is a showcase and nothing beyond one: there is no backend, no real data, and no authentication, and the repository says so plainly. Nothing here has been built to run. The artifact is the argument, and the argument is that an AI-native CRM earns its keep through embedded intelligence, with the chat interface present, essential, and secondary.