Dev48
Language
  • About
  • Services
  • Industries
  • Technologies
  • Articles
  • Contacts
Book a call
    Home/Articles/Fix the process first lessons from reuters momentum ai finance
Dev48
info@dev48.org+375 29 320-40-96WhatsApp@isky_by

© 2026 · All rights reserved.

Fix the process first: lessons from Reuters Momentum AI Finance

Source: Camunda•November 19, 2026 at 09:17 PM

Fix the process first: lessons from Reuters Momentum AI Finance

Bank leaders kept reaching the same conclusion: re-engineer the work before agents run it.

When TIAA printed every screen a service representative uses to answer a retirement participant’s question, the stack stood 81 inches tall.

That's a LeBron James-sized stack of paper.

Sastry Durvasula, TIAA’s chief operating officer, told the room in New York what happens when AI meets a workflow that is clunky: “AI will actually amplify that complexity.”

His point surfaced in nearly every session I attended, including the panel I spoke on about trust and control with agentic AI. McKinsey’s State of AI in 2026 survey points the same way. Only about 6% of organizations qualify as AI high performers, and nearly three-quarters of them have fundamentally redesigned workflows because of AI, compared with one-quarter of everyone else.

The leaders on stage kept tracing that gap to the same cause: AI added to work that needed a redesign first.

Agents expose years of process debt

Durvasula named process debt alongside technical and data debt, and it’s the hardest of the three to see. It lives in workarounds, re-keyed data and handoffs nobody owns from end to end. Aadesh Gandhre, chief audit officer at DTCC, said most processes “cannot be done in a lift-and-shift manner,” which makes any modernization the moment to ask whether a process should exist at all. And if it should, what needs to change now that AI agents can handle complex work.

Re-engineering also means deciding where an agent doesn’t belong. Nastassja Hagan, head of applied AI at BNY, said it plainly on our panel: “The solution, once you figure out what that process is, is not always to use AI.” Where judgment is needed, the strongest designs pair an agent’s reasoning with deterministic rules that run the same way every time.

Schwab took that path with its Charley assistant, starting with easily verified actions such as updating an address before widening what it can do. Firms that redesign first spend their AI budget on the steps that need judgment.

Redesigned value streams change who owns the outcome

Rachel Levi, head of AI, analytics and automation at Swift, drew the line that shapes the new operating model: a task makes someone own an output, while a process makes them own the outcome. Mike Carr, head of technology strategy and innovation at Vanguard, said the biggest payoff comes from “high-complexity, multi-division workflows with handoffs and wait time.” That’s where the money leaks.

One panelist told of a high-net-worth client who moved $2 billion out of a bank after its new app asked him to upload the same KYC documents for each account. The app was new, but the journey behind it hadn’t been redesigned. High-net-worth or not, that’s an experience nobody deserves.

In a redesigned value stream, agents and systems move work at machine speed while people set the rules, handle the judgment calls and stay accountable, a stance panelists called “human in the lead.” Trust sets the pace: costs often rise first while agents and people run in parallel, and savings follow once the organization trusts the system.

Central control and local delivery let reuse compound

Hagan summed up the delivery model: “It is ultimately about centralizing that control but decentralizing the delivery.” A central team runs one internal platform as a service and sets the baseline every agent must meet, from access controls to a kill switch with clear exit criteria. Engineering teams across the lines of business build inside those boundaries using approved patterns and resources to speed up time to market. BNY’s payment-repairs agent shows how autonomy grows within them: it started with a single, recipe-based correction and earned more permitted actions as evidence built up.

Reuse and enterprise scale came up in almost every session I attended, yet few programs had built it into how they fund and deliver work, so teams keep rebuilding the same things over and over again.

Lúcia Soares, CIO at Carlyle, described the habit that breaks the cycle: “If you’re prioritizing your use cases, you can actually fix the data problem in a smaller area to create that blueprint that then allows you to scale across the firm.” Processes built from approvedbuilding blocks compound the same way. Each one launches faster and clears review sooner, lowering the cost of the next.

Durable value lives in how your firm works

Jeff Grimes of Perplexity described frontier models improving every few days, so an advantage built on one model’s strengths erodes on that schedule. The Ship or Shelve panel offered a sturdier test: a durable AI product compounds on proprietary data or workflow advantages competitors can’t copy, and it gets better as models improve.

Zhen Zhao, chief AI officer at New York Life, made that test practical with three questions: who owns it, whether it’s embedded in the actual process, and whether anyone would notice if it were switched off. HIs last point is by far the most important.

A capability that passes Zhao’s test has a named owner, real volume and a result people depend on. That’s where your policies, escalation paths and expert judgment get built into how the work runs, and they stay yours when the model underneath changes. The model becomes a component you can replace, and the process becomes the asset you keep.

Start with one value stream and build it to last

Pick one value stream with real money at stake, map how the work actually moves today, and redesign it before adding agents. Decide which steps need judgment and which should follow fixed rules, name who is accountable for the outcome, and build each piece for the next team to reuse.

The Benchmarking Reality panel closed with questions for leaders to take back to their teams on Monday. Three belong on the table for your next agentic use case:

  • Can every number in an AI-generated output be traced and explained?
  • Has the process been re-engineered to remove non-AI friction before AI is added?
  • Is AI creating a competitive advantage, or only automating work you already do?

See how banks are re-engineering core processes for AI agents.

← All articles

What this article says

More in Software Development

All →
How to read an OCR benchmark before you trust it
Datalab

How to read an OCR benchmark before you trust it

Ignite 2026: Building the frontier ecosystem
Microsoft

Ignite 2026: Building the frontier ecosystem

AI Will Influence $75.8 Billion in Cyber Week 2026 Sales: What Online Sellers Should Know
Ordoro

AI Will Influence $75.8 Billion in Cyber Week 2026 Sales: What Online Sellers Should Know

Amazon Black Friday 2026 Runs 12 Days: What Sellers Need to Know
Ordoro

Amazon Black Friday 2026 Runs 12 Days: What Sellers Need to Know

Elastic and HoneyHive join forces to advance AI observability and evaluation
Elastic

Elastic and HoneyHive join forces to advance AI observability and evaluation

Intuit Reports First-Quarter Results and Enters Agreement to Sell Mailchimp to Bending Spoons
Intuit

Intuit Reports First-Quarter Results and Enters Agreement to Sell Mailchimp to Bending Spoons

More from Camunda

Every process in healthcare is legacy
Camunda

Every process in healthcare is legacy

Where to Start with ProcessOS: A Practical Guide to Re-Engineering Your First Process
Camunda

Where to Start with ProcessOS: A Practical Guide to Re-Engineering Your First Process

Migrating From Pega to Camunda: Where AI Helps and Where it Doesn’t
Camunda

Migrating From Pega to Camunda: Where AI Helps and Where it Doesn’t

From Backup to Recovery: Making Restore Routine in Camunda 8
Camunda

From Backup to Recovery: Making Restore Routine in Camunda 8

Recapping Camunda's Toronto Roadshow: The Case for Process Re-Engineering
Camunda

Recapping Camunda's Toronto Roadshow: The Case for Process Re-Engineering

Camunda 8.10 Webinar Recap
Camunda

Camunda 8.10 Webinar Recap