The first AI use cases a financial services company puts into production set expectations for every one that follows. Choosing them is harder than it looks. Once a central intake process is in place, requests come in from across the business, each with a sponsor and a reason to go first. In our previous piece, “Establishing Control,” we covered how to set up that intake process and the governance around it. This piece covers the next step: deciding which use cases to pursue and in what order.
Organizations that struggle here rarely lack budget, executive interest, or tools. What they lack is a consistent way to compare competing ideas. Without one, early investments often go to the most visible ideas or the strongest advocate rather than those with the best balance of value, feasibility, and risk. The first wave of use cases sets the pace for the program. A strong first use case builds credibility for the next one. A stalled one makes the next one harder to fund.
The Difficulty of Choosing Where to Start
Most financial services companies have more potential use cases than they could reasonably pursue. The challenge is deciding which ones deserve investment first. In our experience, four patterns account for many of the prioritization challenges organizations encounter.
- Solving the Wrong Problem. Organizations identify a visible pain point and immediately begin discussing solutions. But a frustrating process is not automatically a good AI use case. The visible symptoms and the underlying operational problems are often different. Without understanding the root cause, organizations risk investing in a use case that never delivers meaningful value.
- When Everything Is a Priority. Many organizations begin by collecting ideas from across the enterprise. The result is often a list containing dozens of potential use cases, each with a plausible business case and an interested sponsor. Without a consistent way to evaluate them, the list grows while decisions stall. A long list of ideas is not a strategy. Prioritization requires narrowing the field.
- Starting Too Big. Executives naturally gravitate toward the use cases with the largest projected benefits or the highest visibility. The challenge is that these use cases are often the most difficult to implement. They involve multiple systems, significant process redesign, and more complex governance requirements. When early deployments are also the hardest, early setbacks can slow momentum across the broader program.
- Overlooking the Cost of Waiting. Deferral can feel like the low-risk choice, but delaying a strong use case has a cost. The manual effort, errors, and risk exposure it was meant to address continue every month it waits. Organizations that measure today’s baseline can compare the benefit of acting now with the cost of deferring and use that same baseline to measure results after deployment.
While these challenges appear different on the surface, they share a common cause: the absence of a structured process for identifying and evaluating opportunities. The first step is finding the right candidates before attempting to prioritize them.
Where to Find Use Cases
Identifying strong use cases starts with understanding where work is creating cost, delay, or risk. Rather than collecting ideas through open-ended brainstorming, financial services companies should examine how work is performed across the organization and identify the processes where AI could have the greatest impact. This requires two steps: mapping the value chain and locating the operational friction within it.
This work does not need to cover the entire organization before anything moves forward. Many organizations start with one or two business areas, advance the strongest candidates, and expand the assessment as early results build momentum.
Map the Value Chain First
Begin by mapping the organization’s major business functions and the processes within them. Depending on the organization, the map may include:
- Product lifecycle: Product design, pricing, contract or policy development
- Distribution and sales: Customer targeting and acquisition, advisor and agent support, front-office activities, sales, commissions
- Customer servicing: Origination, underwriting, onboarding, servicing, claims
- Operations: Transaction processing, reconciliation, reporting
- Risk and control functions: Risk management, compliance, legal, internal audit, security
- Enterprise support functions: Technology, finance, treasury, human resources, procurement
The goal is to understand how work, data, and decisions move through the organization before deciding where AI fits.
Find the Friction
Once the value chain is mapped, examine the underlying processes for signs of operational friction. The considerations below are not mutually exclusive. They often appear together and processes that show several of them tend to be the strongest candidates:
- High-volume, repetitive work that consumes significant employee time
- Manual gathering, comparison, or reconciliation of information
- Handoffs across disconnected systems or teams
- Rework caused by incomplete information or process exceptions
- Delays created by queues, reviews, or approval steps
- Reliance on a small number of employees with specialized process knowledge
- Processes or functions that are currently outsourced or are candidates for outsourcing
These conditions do not automatically make a process suitable for AI, but they provide a focused starting point. They show where employees spend time assembling information, applying defined standards, routing work, and resolving recurring exceptions.
Putting It Together
The example below maps the business lifecycle of an insurance company, from product development and distribution through onboarding, billing, and claims. The shading illustrates where AI is likely to have the greatest impact. Activities with high volumes, heavy document handling, repeated data gathering, and defined steps tend to rank higher. Activities that depend more on relationships and judgment tend to rank lower. The activities will differ by business and industry, but the same approach applies across the organization. This provides a starting point for a focused discussion about where AI could make the biggest difference.
Illustrative Insurance Business Lifecycle Value Chain
Together, these steps produce a focused set of candidate use cases grounded in how the organization actually operates. And of course, supported by enabling functions like finance, human resources, legal, compliance, risk, corporate development, and technology, which also will benefit from AI. The next step is to screen out candidates that are not ready and evaluate the rest consistently.
Screen Before You Score
Before scoring begins, each candidate should clear a short set of hurdles. A use case that fails any of them should be set aside, however attractive it looks on paper:
- Meets a minimum impact threshold. The use case should be large enough to justify the effort of building, governing, and maintaining it. Ideas that save a few hours a month or affect a small number of transactions are rarely worth a place in the portfolio, even if they are easy to build.
- The upstream process works. If the work arrives incomplete, inconsistent, or late because of a problem earlier in the process, AI will only move the same problem faster. Fix the upstream process first.
- The regulatory treatment is clear. If it is not yet clear how regulators will view an AI-supported decision in a given area, the cost of getting it wrong outweighs the benefit of moving early. These are better revisited once expectations are clearer.
- Human judgment is not the core value. Some activities are valuable because of the experience, relationships, or discretion a person brings. AI can support that work, but it should not be the target of automation.
Use cases that do not clear these hurdles are not permanently off the table. A broken process can be redesigned, and regulatory expectations may become clearer. Setting them aside for now keeps the scoring focused on candidates that can realistically move forward.
The Prioritization Framework: 4 Dimensions
Once candidate use cases have been screened, the next step is to determine which opportunities should be pursued first. Each candidate should be evaluated across four dimensions:
- Automation Potential
- Business Value and Operational Impact
- Technical Feasibility and Data Readiness
- Risk and Regulatory Exposure
Organizations should define clear scoring criteria within each dimension and apply them consistently across all candidates. The dimensions can then be weighted to reflect the organization’s strategic priorities and risk appetite. The result is a ranked portfolio that provides a consistent basis for investment decisions.
Automation Potential
The first dimension evaluates whether AI can realistically perform the work and whether AI is an appropriate solution. Some processes can be entirely automated. Others are better suited for augmentation, with AI preparing or completing part of the work while a person remains responsible for review or decision-making. Some require process redesign before AI can be introduced, and others may not be feasible with current capabilities.
AI is best suited to work that involves gathering, reading, comparing, and organizing information. In financial services, that may include:
- Aggregating data from multiple systems and documents into a single view
- Researching markets and customers such as competitor products and pricing, industry trends, and prospect or client information
- Extracting information from unstructured content such as applications, contracts, statements, and emails
- Comparing information against a standard, such as a policy, checklist, regulation, or prior record, including quality assurance reviews
- Summarizing case histories, files, and long documents
- Managing workflow including triaging, routing, and tracking work based on defined criteria
- Drafting recommendations, case narratives, and standard communications for review
Applied to the right work, these capabilities improve speed and consistency. They also make it practical to review full populations rather than samples and let employees spend more of their time on judgment, exceptions, and customer relationships.
Processes are generally strong candidates when they follow repeatable steps, use accessible information, apply documented rules, and produce outputs and exceptions that can be checked and routed for review. Processes that depend heavily on undocumented judgment, individual experience, or frequent exceptions may be better suited for augmentation or redesign. That difference affects investment: A process that needs redesign before development costs more and takes longer than one where AI can fit into an established workflow.
Business Value and Operational Impact
The second dimension evaluates the value the use case could create. That value may come from reducing cost, increasing employee capacity, improving customer experience, accelerating cycle times, generating revenue, or reducing operational and compliance risk.
The assessment should begin with the current performance baseline. Depending on the use case, relevant measures may include:
- Processing time or average handle time
- Cost per transaction or case
- Employee effort required
- Error, rework, or exception rates
- Customer response and resolution times
- Sales conversion, revenue per employee, or advisor and agent productivity
- Customer retention, satisfaction, and complaint volumes
- Control coverage and effectiveness
- Financial losses or risks avoided
From there, organizations should define the expected improvement and how it will be measured after deployment. They should also distinguish between theoretical efficiency and realizable value. Time saved only creates value if the capacity can be redeployed, operating costs can be reduced, or service and control outcomes improve. Value should also be assessed net of the cost to operate and maintain the solution, including licensing, usage costs, monitoring, and the staff needed to review outputs and keep rules current.
A use case with a clear baseline, measurable target, and defined source of value is more likely to sustain executive support through implementation than one supported primarily by qualitative benefits.
Technical Feasibility and Data Readiness
The third dimension evaluates whether the use case can be built, integrated, and sustained within the organization’s existing environment. A technically possible solution may still be a poor investment if the data, integration effort, or ongoing operating costs outweigh its expected value.
The assessment should consider these areas:
- Data: Is the required information available, accessible, complete, and reliable? Does the process depend on unstructured information such as scanned documents, PDFs, emails, or free-text notes?
- Integration: Which internal and external systems must the solution access? Are reliable integration paths available, and can the organization support them over time?
- Build complexity: How variable is the workflow? How many decision points and exceptions must be addressed?
- Resource availability: Are the people needed to build, test, and implement the solution available, including business subject-matter experts, technology teams, and risk and compliance reviewers?
- Ongoing operation: What will the solution cost to run, monitor, maintain, and update? How frequently do the underlying data, rules, or processes change?
Each additional dependency increases implementation effort and introduces another potential point of failure. The assessment should therefore consider the full cost and complexity of operating the solution, rather than focusing only on whether a prototype can be built.
Risk and Regulatory Exposure
The fourth dimension evaluates the consequences if the AI produces an incorrect result or acts outside its intended boundaries. As discussed in our previous article, AI applications should receive oversight proportionate to their potential impact.
Risk generally increases when a use case:
- Interacts directly with customers or employees
- Uses sensitive or protected information
- Influences a consequential decision
- Communicates externally
- Operates without review before action is taken
- Produces outcomes that are difficult to explain or reconstruct
- Could create material financial, regulatory, legal, or reputational harm if it fails
- Falls outside existing policies and procedures, or requires exceptions to the AI governance framework
The risk classification established during the governance stage should feed directly into this dimension. A higher-risk classification does not automatically remove a use case from consideration. It does, however, increase the controls, testing, documentation, human oversight, and ongoing monitoring required to deploy and operate it.
Applying the Framework
Scoring works best when the people closest to each dimension contribute to it. Business owners are typically best placed to assess value and automation potential. Technology teams assess feasibility and data readiness, and risk and compliance assess regulatory exposure. A central function, such as an AI transformation office, should facilitate the process: It applies the scoring definitions consistently, gathers the supporting evidence, and brings the results to a steering committee for review. That way, no single function controls the outcome, and differences in judgment get discussed openly.
Each dimension should be scored on a simple scale, such as one to five, with a short description of what each score means. For example, a high score for data readiness might mean the required data is available in structured form from a single system, and a low score might mean it is scattered across documents and inboxes. Clear definitions make scores comparable across use cases and reviewers, and every score should be backed by evidence such as process data, system documentation, or input from the employees who do the work.
Leadership should set the weights before any candidates are scored, so the weights reflect strategy instead of being adjusted to favor a particular idea. An organization focused on near-term efficiency may give more weight to business value and feasibility. One operating under heightened regulatory scrutiny may give more weight to risk. The resulting ranking supports leadership’s decision rather than replacing it. It makes trade-offs visible and gives the discussion a common starting point. In some cases, the outcome will be a build decision. In others, it may point to process redesign, data remediation, or conventional automation before AI becomes practical.
The example below shows how four candidates at a bank might compare. Each is scored from one to five on every dimension for a total out of 20. For simplicity, all four dimensions are weighted equally. In practice, weights should reflect the organization’s priorities.
Illustrative use case prioritization. Each use case is scored 1–5 on all four dimensions, for a total out of 20.
Takeaways and How Ankura Can Help
Effective prioritization gives organizations more than a ranked list. It creates a clear basis for deciding where to invest, identifies the work required before development begins, and establishes the measures that will be used to evaluate results. The strongest candidates combine meaningful business value with a realistic path to implementation and a level of risk the organization is prepared to manage.
Ankura helps financial services companies move from a broad set of ideas to a focused portfolio of build-ready use cases. Our support includes:
- Identifying opportunities: Mapping the value chain and working with business and operations teams to identify processes where manual effort, fragmented data, delays, and recurring exceptions create meaningful opportunities for improvement.
- Evaluating and prioritizing candidates: Applying consistent criteria across automation potential, business value, technical feasibility, and risk to help leadership compare opportunities and agree on sequencing.
- Measuring value: Establishing current-state baselines, defining target outcomes, and setting up the reporting leadership needs to track results after deployment.
- Preparing for execution: Identifying process, data, technology, and control gaps that should be addressed before development begins.
The output is a practical roadmap that identifies which use cases should move forward, what must be addressed first, and how progress and value will be measured. The next article in this series will cover the third stage of the AI Deployment Playbook: building, deploying, scaling, and measuring prioritized use cases.
To learn more about how our Financial Services Advisory practice can help your organization move from AI experimentation to governed, scalable deployment, visit our dedicated Financial Services Advisory page: Financial Services Advisory – Ankura.com.
Related Insights
Establishing Control: How Financial Services Companies Build AI Governance That Gets AI Into Production – Ankura.com
From Pilot to Production: A Practical AI Deployment Playbook for Financial Institutions
© Copyright 2026. The views expressed herein are those of the author(s) and not necessarily the views of Ankura Consulting Group, LLC, its management, its subsidiaries, its affiliates, or its other professionals. Ankura is not a law firm and cannot provide legal advice.
