Training AI
QuickBooks
Content design | New functionality | In-product
We added just enough friction at the right moment to inspire confidence in automation. We call it Train QB—a simple setup process that invites the user to answer a few quick questions, sort some expenses, and teach our AI how to categorise.
Context
Smart automation works behind the scenes to match a customer’s bank transactions to their chart of accounts instantly and error-free. This process is essential for maintaining business finances, which was previously manual, time-consuming, and error-prone.
Challenge
Technically, our AI worked perfectly, almost entirely eliminating the categorisation process for our customers. However, during early testing, we found that customers were not confident that their transactions had been categorised correctly, and we observed them reviewing each transaction manually, even after AI had done its thing.
UX writing for confidence building
Our goal was to provide reassurance throughout the process. Copy was written to anticipate hesitation, such as “you can always change the category later” to mitigate anxiety about the AI making mistakes. Clear, instructional language helped to build trust step by step, ensuring users felt in control without needing to manually verify every decision.
Solution
To add just enough friction at the right moment to inspire confidence. We call it Train QB—a simple setup process that invites the user to answer a few quick questions, sort some expenses, and teach our AI how to categorise.
UX writing solution
The writing here focused on emphasising collaboration between the user and the AI. By using phrases like “show us how you like to organise your expenses, so we can do it for you,” we positioned the AI as a tool that adapts to the user’s preferences. The conversational tone helped to make the interaction feel personalised and approachable, enhancing the user’s willingness to trust the tool.
Discovery process
-
Early empathy, discovery and prototype testing of AI with customers in multiple markets.
-
Design for delight guiding principles used to uncover a solution to the customer problem.
-
Design, test and iterate cycle, including multiple rounds of prototype testing.
-
Build, and gradual roll out to new users.
UX writing process
-
​Discovery and research: conducted interviews with customers to understand anxieties and needs, which shaped the tone and style of the microcopy.
-
Collaboration with design: worked closely with designers to ensure that the copy supported the visual layout and flow of the experience.​
-
Prototyping and testing: integrated UX writing into prototypes, iterating based on real user feedback to ensure clarity and ease of use.
-
Finalisation and iteration: continuously refined the copy based on evolving user interactions, ensuring ongoing improvements in customer experience.
Results
​The AI-powered 'auto-categorisation' helped users:
-
Cut transaction categorisation time, from hours per month to seconds.
-
Reduced categorisation errors.
-
Reduced accountant intervention is required.
​
​The one-off setup process helped us:
-
Boost trust and acceptance of the auto-categorised transactions.
-
Understand that a little friction at the right time can be a powerful tool.
Microcopy and tone impact
The conversational microcopy played a direct role in these outcomes, as it eased user interaction with the AI, minimising hesitation and confusion. The clear and direct tone of the messages ensured that users felt supported throughout the process, resulting in a smoother setup and fewer mistakes.