LEADERSHIP INSIGHT
How do you decide when to trust AI?
An employee brings you an analysis, a draft procedure, or a recommendation. It is clear, organized, and took a fraction of the usual time.
They used AI to help prepare it.
What would you need to know before approving it?
That question belongs in every department. Accounting may need to verify a variance explanation. HR may need to check an employee communication against current policy. Operations may need to confirm that a suggested shortcut preserves a necessary control.
AI can help a team make progress quickly. Managers set the expectations for what happens between receiving an answer and putting it to work.
A useful place to start is asking what information the tool used, which assumptions shaped the answer, and what someone checked independently. The depth of that review should reflect the consequences of getting it wrong.
You do not need to personally rework every draft. You do need a team that can explain why its work is ready to use.
IN THE KNOW
Your member may arrive with an answer already
A member asking about a loan may have already discussed it with a chatbot. In a September 17 release, PensionBee reported that 57% of the 1,000 U.S. adults it surveyed who use AI for personal finance would act on its guidance without checking it. The online survey used a non-probability sample; it does not measure all Americans or credit union members.
For credit unions, that creates a service opportunity: understand what the member wants to accomplish, examine the assumptions behind the advice, and help them find an appropriate next step. Being helpful starts with being curious about what they were told.
The next card purchase could start with an AI agent
AI is moving into the checkout process. On September 30, Mastercard announced new Agent Pay services, including a probability score entering U.S. testing to indicate whether an AI agent initiated a transaction.
An agent could, for example, make a purchase within permissions a person has given it. That adds another layer to the familiar question of who authorized a payment. For cards, operations, and fraud teams, this is a development worth understanding as providers build the infrastructure. The announcement describes testing and new capabilities, rather than universal availability across credit unions.
Consumer borrowing slowed and the details matter
The Federal Reserve’s October 7 release showed consumer credit growing at a 1.9% seasonally adjusted annual rate in August. Revolving credit declined at a 4.2% annual rate, while nonrevolving credit grew at 4.1%. These figures describe annualized monthly changes; the release excludes loans secured by real estate.
The mix matters for planning. Slower balance growth can prompt a closer look at applications, card usage, repayments, and member demand. It does not, by itself, tell us whether households are becoming more financially comfortable. Your credit union’s own activity will help explain which parts of the national picture matter locally.
ONE MOVE TO MAKE
Agree on what gets checked
Choose one recurring task where your team already uses an approved AI tool, or a task you are considering for an authorized pilot.
Bring a sample to your next team meeting. Identify the information it depends on, the assumptions that need checking, and the person who can verify it before use. Include a condition that would require a second review.
Keep the exercise focused. A draft training agenda and a proposed change to a member-facing procedure deserve different levels of scrutiny.
Use Before You Act on AI: A Manager’s Review Guide to capture your agreement. Apply it to the next example, then ask whether the checks caught something useful and whether the process was practical enough to repeat.
DOWNLOAD THE MANAGER’S REVIEW GUIDE BELOW →
A FINAL NOTE
Make room for better questions
A useful question for your next meeting is, “What did you check before bringing this forward?”
Ask it with curiosity. Give people room to explain what they verified, what they changed, and what they still do not know. That conversation helps you understand both the work and the support your team needs.
Where would a clearer review process help your team most?
Until next time,
Keep Ascending.
Reply and tell us: What is one task where your team would benefit from a clearer way to review AI-assisted work?