Shane Ian Murphy: Testing AI Explanations with a Cambridge Customer Scenario
A proposed design exercise separates explaining an option from asking permission to act.
I am Shane Ian Murphy, a former co-founder and Chief Business Development Officer of Limitless Mobile and the founder of Global Secure App. This is my affiliated founder view on proposed design principles for helpful AI explanations. I see a useful role for explanations, but I do not think a fluent answer is enough. The customer should understand what the system is explaining, which information supports it, and whether anything will happen as a result of the conversation.
Financial services contain many moments where an explanation can help. Someone may want to understand a transaction status, compare the stated costs of available routes or find the conditions attached to an offer. AI could make that information easier to navigate. The difficult part is preserving the distinction between describing an option and recommending it, or between preparing an action and carrying it out. A friendly tone should not blur those boundaries.
Consider a fictional Cambridge customer asking an assistant to explain an unfamiliar charge in a product demonstration. The customer wants to know what the entry refers to; they have not asked the assistant to change anything. A useful response would distinguish the recorded description, any supporting source and the information still missing. It would leave the transaction unchanged unless the customer separately requested an available action and reviewed its consequences. This example describes a proposed test, not an existing product feature or a real customer's account.
My preference is for explanations that reveal their basis in ordinary language. If a fee estimate comes from the selected route and amount, say that. If a processing time comes from a provider's published expectation, identify its status as an expectation. If the system cannot determine why a payment is delayed, it should not invent a likely cause and deliver it confidently. An honest boundary can be more helpful than a detailed answer that sends the customer down the wrong path.
Customer control means more than adding a confirmation button at the end. Before an action, the person should be able to inspect the relevant amount, destination, conditions and consequences. They should know whether the assistant is still gathering information, has prepared an instruction or is asking permission to execute one. Changing a detail should be possible without restarting an opaque conversation. When a process has real consequences, the review screen needs to carry the meaning of the decision clearly.
I also think people need a route out of the conversational interface. Some customers prefer structured screens. Others need support from a person when a situation is complex or distressing. An assistant should help them reach the appropriate route with relevant context, subject to privacy controls, rather than insist on repeating generic answers. Whether the issue belongs with us or a provider also needs to be clear. A handoff is useful only if the customer understands who can actually investigate.
Data use deserves a similarly direct explanation. A service should be clear about what information is used for a particular feature and what choices a customer has. I would not treat broad access to personal financial information as the default price of receiving a simple explanation. The precise arrangements will depend on the product implemented, but the design question can be asked now: what is necessary to answer this question, and can we provide useful help with less?
I am describing principles for development, not claiming that an AI assistant or any particular integration is live. Before launch, those principles need testing against real product constraints and provider responsibilities. The outcome I want is modest and practical: a customer leaves an interaction understanding more than they did before, with a clear record of what they chose. Assistance should make it easier to exercise judgement. It should not make judgement disappear behind a system that sounds certain about everything.
A useful explanation should also survive a follow up question, allowing the customer to inspect a relevant condition or correct an assumption before making their choice.
About the Creator
Shane Murphy
Shane Ian Murphy: co-founder/partner, Tri-Party Tokenisation; founder/CEO, Global Secure Technologies (planned 2027 launch); co-founder/partner, Dual Party Capital. Partner, RougeN International Corp. Formerly of Limitless Mobile.
Enjoyed the story? Support the Creator.
Subscribe for free to receive all their stories in your feed.
Comments
There are no comments for this story
Be the first to respond and start the conversation.