How to audit your firm's workflows for useful AI opportunities

Many financial services firms know they should be exploring AI. The harder question is where to begin.

Employees may already be using tools to prepare emails, summarize documents or work through reports. Leadership may have seen convincing demonstrations. Vendors may be proposing agents, assistants and automation before anyone has agreed which part of the business needs attention.

The firm already has a useful starting point: the workflows its people run every week.

Monthly reports, client onboarding, meeting follow-up, reconciliations, document review, information requests and approval processes all contain evidence about where time is spent, where work slows down and where mistakes become costly.

A structured review of those workflows makes the AI discussion more concrete. It helps leadership see what could be improved, what information the work depends on and what would need to remain under human control.

Start with work that already happens

An early AI project is easier to assess when the underlying work is familiar.

Take a monthly client report. Someone collects information from several places, checks which version is current, compares figures with the previous month, follows up on missing items and prepares a draft for review.

The process may already be documented. The useful detail often sits with the person doing it.

They know which spreadsheet tends to arrive late. They know which client uses a different format. They know which source is considered authoritative when two records disagree. They can usually tell when a figure deserves another look, even if it passes a basic calculation check.

Those details determine whether AI can help. They also show where ordinary automation, clearer ownership or better source data may solve the problem more reliably.

Sit with the people who run the process

A workflow audit should include the people who perform and review the work.

Ask them to walk through a recent example from the first input to the final approved output. Watch where they move between systems, rename files, copy information, compare records, wait for another person or make a judgment call.

Useful questions include what starts the process, which files and systems are used, how someone knows which version is current, where the work usually pauses, which inputs are unreliable, what the reviewer checks and which exceptions require experience or escalation.

The purpose is to understand how the work actually runs. A procedure document may describe the intended path. A recent example reveals the handoffs, workarounds and decisions that shape the real one.

Follow one complete cycle

Broad discussions about efficiency can produce long lists of ideas without enough detail to evaluate any of them.

Following one complete cycle keeps the review grounded. For a client onboarding process, that might begin when the firm receives an instruction and end when the approved record, documents and next actions are visible to the right people.

For a reconciliation, it might begin when source files arrive and end when every difference has been resolved, approved or recorded as an exception.

Capture the trigger and expected outcome, the people and systems involved, the information used at each stage, the checks and judgment calls, the common exceptions and the final system of record.

This creates a usable picture of the workflow. It also prevents a team from improving one task while leaving the surrounding handoffs untouched.

Look for repeated effort and decision friction

AI tends to be most useful when the work contains repeated handling of information and a clear way to review the result.

The audit may reveal that someone repeatedly collects the same data, compares similar documents, classifies incoming requests, searches for supporting information, prepares a first draft or flags differences for another person.

Those are credible places to investigate.

The audit may also expose a different problem. Two systems may hold conflicting records. An approval may have no clear owner. A report may depend on files that arrive under inconsistent names. Staff may be compensating for an unclear process through memory and manual checking.

AI could still form part of the eventual solution. Leadership first needs to understand whether the workflow is ready to support it.

Mark where judgment enters the work

Financial services workflows often combine routine handling with decisions that depend on context, experience and responsibility.

An AI system may collect information, compare records, identify missing items or prepare a draft. A person may still need to decide whether a difference is material, whether a client instruction is clear, whether an exception can be accepted or whether another team must become involved.

The audit should make that boundary visible.

For each possible AI task, identify what the system may prepare, what evidence it must show and who decides whether the output can move forward. If the reviewer needs the original document, prior record or calculation, that information should be available beside the proposed result.

Human review becomes far more useful when the reviewer has a defined job. A vague instruction to check everything can leave them reconstructing the process from the beginning.

Check the information and permission boundary

A promising use case can still be unsuitable if the required information is incomplete, poorly controlled or unavailable to an approved tool.

Record which data the workflow uses, where it is stored and who currently has access. Note whether the process includes personal information, client records, confidential commercial material or documents subject to retention and approval requirements.

Then define the operating boundary for an early test.

Which information may enter the system? Which sources can it retrieve? Which records should remain outside the first version? Who can see the output? How long should working material be retained? What should happen if a required source is missing or two records disagree?

These decisions affect the design from the beginning. They should not be left until after a convincing demonstration.

Compare the opportunity with the current baseline

The workflow needs a starting point if leadership wants to judge whether the project helped.

The baseline can be simple. Record how often the process runs, how long a normal cycle takes, where work waits, how much review is required and how frequently items return for correction. Note which delays or errors matter most to the business.

The measure should match the purpose of the workflow. A client-service process may care about response time, record completeness and missed follow-ups. A reporting process may care about preparation time, corrections, late inputs and reviewer effort. A document-review process may care about coverage, false alerts, traceability and time to resolution.

The goal is to give the firm something real to compare. A faster first draft may be useful, but the full result also includes checking, correction, approval and any new administration created by the system.

Rank the opportunities before building

A workflow audit will often produce more ideas than a firm should pursue at once.

Consider whether the workflow affects a meaningful business outcome, occurs often enough for improvement to matter, has dependable source information, produces an output that can be checked, has manageable permission boundaries and can be tested without disrupting important work.

An opportunity may rank highly because it saves repeated effort and has a clear review path. Another may need better data or a more stable process before AI is introduced. A third may be possible but too minor to justify maintaining.

Finding that out is part of the value of the audit.