AI Can Help Banks Preserve Institutional Knowledge — Or Scale Your Worst Workarounds
By Meena Athinathan, VP & Head of Strategic Business Unit for Banking, Capital Markets Infrastructure and Indexes at Cognizant
Simple Subscribe
Subscribe Now!
Banks may be on the verge of automating the very habits they should be eliminating.
For years, commercial banks have worked to reduce technology debt: aging systems, brittle integrations and manual processes layered onto platforms that were never designed to work together.
Now, as banks begin experimenting with AI agents, they face a different kind of liability hidden in how work actually gets done.
Need to Know:
- Wolters Kluwer’s Q1 2026 Banking Compliance AI Trend Report found that 61% of financial institutions have either deployed AI or machine learning in production or are actively piloting it.
- But only 12.2% said their AI strategy is well defined and resourced, which suggests many institutions are putting AI into motion before they have built the operating model to control what it learns, where it acts and when it should stop.
- Call it “people debt”: The undocumented judgment, exceptions and informal steps that keep banking operations moving, but rarely appear in systems or standard operating procedures.
Why “People Debt” Can Kill AI ROI
That missing context is where AI ROI can start to break down. While a workflow can tell an agent what step comes next, it may not explain why an employee pauses, checks another system, calls another team or escalates a client request. These pauses often look inefficient on paper, but they may be the only actions keeping a wrong decision from reaching a client, a regulator or even a balance sheet.
In account reconciliation, invoicing, document review, or KYC onboarding, those decision points often determine whether automation improves the process or repeats its weaknesses. In a KYC review, for example, an experienced employee may know that one missing data field is a routine documentation issue, while another should trigger enhanced due diligence.
Key insight: Every bank has people debt. But if banks assign AI agents to these workflows before separating useful judgment from inherited workarounds, they risk preserving the old process in code rather than modernizing it.
The Three Steps to Converting People Debt
AI can help banks close the people-debt gap, but only if they understand the work. Banks can now use tools to harvest this “tribal” knowledge, create knowledge drafts and generate process graphs so work that once lived in employees’ heads becomes part of the workflow.
1. Banks should study the work before automating it: Teams should start where employees step outside the system to get work done. For example, if an onboarding team depends on experienced employees to check documents, validate information and resolve exceptions, AI can help make that knowledge more consistent. But the bank still needs clear rules for when the agent can proceed, when it should ask for review and when a human must decide.
- Map where exceptions occur and which manual steps are useful versus which exist because of legacy system limits.
- Determine which decisions need to stay with a person and which can be handled through defined rules.
- Categorize workarounds as actual risk controls, symptoms of broken systems, or processes that should be eliminated altogether.
2. Governance must be built into the operating layer: For AI decisioning to work, banks need reliable data and guardrails from the start. AI agents cannot be treated like another software layer. They act across systems, so their controls and escalation paths need to be defined before they start making or recommending decisions. In a regulated environment, there is no room for hallucination. Before assigning agents to knowledge-heavy workflows:
- Know where data comes from, who owns it, who can access it and when a human needs to step in.
- Define controls and escalation paths before agents start making or recommending decisions.
- Route agents through a controlled choke point rather than letting them run free across the enterprise until the rules mature.
3. Don’t Shy Away from Cultural Change: The organizational model matters just as much as the technical one. Recent workforce analyses suggest AI-adopting companies are creating new jobs, even as roles and skills change. Banks need to invest in a cultural mindset shift and the reskilling opportunities required to solve problems with AI rather than simply bolt AI onto existing processes.
- Give operations, risk and compliance teams role-specific education on how AI will affect their decisions, escalation responsibilities and accountability.
- Include frontline employees in supervised pilots so their knowledge informs where agents can act, where human review is required and where automation should stop.
- Create formal channels for employees to report flawed agent recommendations, recurring exceptions and process gaps, then show how that feedback changes workflows and controls.
- Hold managers accountable for preparing teams for new responsibilities, including training, pilot participation and updates to operating procedures.
Bottom line: If banks give agents broken work, they should expect broken outcomes at machine speed. To avoid teaching AI the institution’s worst habits, banks need to redesign the work, the culture and the controls before agents start making decisions. That is how institutions turn people debt into institutional resilience instead of encoding informal workarounds into the next generation of systems.
