I’ve noticed a consistent pattern across financial institutions that have tested general-purpose AI tools but struggle to identify and measure impact. Leadership makes the decision to invest, then implements across teams and waits for the impact. After a few months, the confusion hits. Results aren’t clear, processes aren’t improving, adoption is lagging, and the numbers remain unchanged. The ROI is not evident, and they point to the tool as the problem, but the reality is that when you implement a general tool for general productivity gains, there isn’t a meaningful way to determine ROI.
When I talk with banks and credit unions about AI, I try to drill down and identify the problem they’re trying to solve for with it. The answer often reflects a tools-forward approach rather than a problem-forward approach. They’re given technology and told to be productive, but no one is taking the time to evaluate which workflows are compromised, where the bottlenecks live, and what success metrics look like for the institution. This identification and discernment help clarify if AI is even the right tool for the job and mitigate the disconnect between the technology and the workstreams it’s trying to optimize.
Why General-Purpose AI is the Problem
When you add a general-purpose AI tool without embedding it into a specific workflow, you’re only adding to the burden. The team member is required to learn something new, draw conclusions around where and how to use it, navigate multiple systems to inform context, and then synthesize the findings to inform decisions. It also contributes to what many of us refer to as the swivel chair effect: constantly moving between platforms and systems to build a complete picture.
Comparatively, AI tools designed for specific workflows meet employees where they work instead of asking them to introduce a new habit or system. Because these tools are purpose-built, they understand the workflow nuances, have access to data sources, and understand what can and can’t be done.
If the goals are ROI and adoption but no change management plan or budget exists, the difference between “one more tool to learn” and “a purpose-built addition to your workflow” is the difference between seeing results and not.
General-Purpose AI
Purpose-Built AI
Approach
Deploy the tool, then find use cases
Identify the workflow, then build around it
Integration
Sits outside existing workflows
Embedded directly into workflows
Experience
Adds another tool and more context switching
Reduces friction and surfaces relevant data
Data
Employees gather context manually
Relevant data is accessed automatically
ROI
Broad gains are difficult to measure
Measured against workflow-specific baselines
Adoption
Depends on employees forming new habits
Fits existing habits and solves a defined problem
What This Looks Like in Practice
The use cases are familiar. A call is routed to a customer service representative requesting support for an account holder needing to make a larger payment than typical and wondering if the transition limits will stop it from happening. Without an embedded AI tool, the CSR is forced to bounce between multiple systems: one to review recent transactions, one to evaluate limits, and potentially a third to request any adjustments, as well as a policy review to confirm what’s authorized. This all happens while the account holder is on hold and sometimes takes up to thirty minutes.
When AI is purpose-built into the workflow, the same task can take closer to ten seconds. Relevant transaction details are surfaced instantly, limits are compared, adjustments are made that align with the institution’s policies, and everything can be viewed in one place. Rapport is built between the CSR and account holder, and the institution can dedicate what could have been thirty minutes to more meaningful work. The difference is a quantifiable improvement within day-to-day workflows that directly impact customer and member experience.
A Foundation for Measurable Outcomes
Purpose-built AI simplifies the ability to quantify and measure success because it’s constructed for the very workflows that were identified, measured, and selected as an opportunity for optimization with AI. This means a baseline already exists, and you can then measure against those baselines after the fact. This matters because AI is expensive. Usage costs are ongoing and need to be justified. The ability to clearly communicate measurable improvements is critical to ensure budgets are approved and accounted for.
The Community Institution Competitive Advantage
I work closely with many community institutions, and they have an advantage that often gets overlooked. These banks and credit unions know their customers and members personally. They understand the operational nuances that make them unique, and can quickly identify where workflows break down, and where bottlenecks and inconsistencies exist. That proximity matters. When purpose-built AI is implemented thoughtfully, there’s often an immediate impact. Teams show up more consistently and better informed, and institutional knowledge becomes something the whole organization can access and build on.
When AI solves real customer problems and begins automating what’s repetitive, a 20-person team can deliver the output and consistency of a 35-person team. The experience gets better for customers and members, and employees get their time back to focus on what humans do best, like judgment calls, kindness, and the empathy that keeps people coming back and sets community institutions apart.
Where to Begin
If AI investments feel overwhelming, begin with the problem, not the AI tool. Take the time to sit with your teams to learn what their challenges are and where the opportunity lies to improve their working experience. Understand where the friction exists, what’s requiring more output than it should, and what information is constantly a challenge to uncover. The answers will help inform where AI can create real value.
The financial institutions making the most progress with AI are not the ones who moved fastest, but the ones who built with purpose.