Agentic Connectivity

The Assumption We Had to Challenge

When this project started, the internal conversation was about building a better connectivity page inside the banking portal. The assumption was straightforward: clients come to JPMC to manage their money, so they'd want to manage their integrations there too.

Research quickly proved that wrong.

"As much as possible, 100% of what I'm doing is inside the ERP. I would only go to JPMC if I got a notification." — Participant 6, CalChlor

That finding didn't invalidate the dashboard — it clarified what it was actually for. Clients saw it as a validation surface: a place to confirm connections are healthy, get peace of mind, and resolve issues when something breaks. What they didn't want was to manage connectivity proactively from a bank portal. They wanted to be notified the moment a problem surfaced, then have a focused space to diagnose and fix it. The connectivity health logs added a third dimension: clients saw real value in the audit trail — knowing exactly what data was shared, with whom, and when — for their own controllers and compliance reviews.

What I Led

In June, I ran a 2-day workshop with JPMC's FDX product, design, and engineering teams. The goal was to stop treating our three connectivity modes as separate product problems and build a coherent vision across all of them: FDX Export (JPMC account data flowing out to third-party apps), FDX Import (external bank data flowing in to JPMC analytics), and DAP (ERP supplier data integrating with JPMC payments).

The workshop produced a shared framework. I then shaped that into a set of concepts and principles, presented them to leadership in July, and prepared a more formal presentation for the EAS executive team in August.

The Vision That Emerged

The workshop output crystallized into a three-pillar framework for Connected Services 2.0. First, a centralized dashboard — giving finance teams a single source of truth across all their bank connections, third-party apps, and ERP integrations. Second, a modular design system — a set of composable consent and integration components that could be deployed inside any surface: JPMorgan's own portal, an ERP, or a chat interface. Third, an AI connectivity layer — a proactive co-pilot that would evolve from a simple assistant to an advisor over time, reducing manual intervention and surfacing issues before clients need to go looking.

AI as a Strategic Driver

The agentic piece was the most forward-looking — and the most debated. We didn't frame AI as a feature to layer on top. We mapped it as a progression: from Tool (passive, client-driven) through Assistant and Teammate, to Advisor and eventually Boss. The near-term target was Assistant to Teammate — capable enough to walk a user through connecting a bank account inside a chat window, without them ever leaving their current context. Agent J, the connectivity co-pilot we designed, sits in that range. The longer horizon is Advisor: proactively surfacing connection health issues, flagging expired consents, and suggesting actions before a finance team even notices a problem.

Modular Design as a Scaling Strategy

The third pillar was what would make the other two sustainable at scale. Rather than designing bespoke experiences for every surface and persona, we built a library of reusable modules — a Dynamic Consent Module and an Agentic AI Module — that could adapt to context. The same consent flow could run inside the JPMC portal, embedded in QuickBooks, or triggered through a NetSuite plugin. The same AI module could surface in a chat interface or as an alert inside a dashboard widget. For a finance team ranging from Faith (Founder, early-stage startup) to Ronnie (AR Manager, enterprise treasury) the underlying trust and data-sharing infrastructure stays consistent — only the wrapper changes.

What the Research Confirmed — and What It Didn't

In September, we ran a research study with 10 finance professionals across 6 companies — CFOs, controllers, AP managers, and IT directors using ERPs like NetSuite, Acumatica, and FISPAN. ERP-to-bank connectivity scored 6 out of 7. That validated the core direction. But the agentic features we were most excited about scored only 4.

"I've experienced some of those AI agents, and I end up just going around in circles. So there's concern there about how that AI is built and is it useful." — Participant 1, National Notary Association

That finding shaped how we positioned the AI work: not as automation that replaces judgment, but as optional assistance that earns trust incrementally through transparency and explicit opt-in.

The Decision That Changed the Product Roadmap

The hardest call wasn't about UI — it was recommending that the primary connectivity experience live inside the ERP, not inside JPMC's own portal. That meant deprioritizing a surface we controlled in favor of one we had to partner on. The research made the argument, but it still required conviction to take to stakeholders.

What Got Built

The consent management dashboard — one of the two surfaces I designed — is currently in implementation for JPMC's digital banking platform. In parallel, a sister team built a working agentic connectivity prototype that was showcased at an ERP conference, validating the technical direction the vision had set out.

Two surfaces anchored the design: a Connected Services management dashboard (L1) giving finance teams a single view of all active connections, and a guided consent module (L4) walking users through connecting external accounts with explicit data controls and step-by-step approval.