Why you need to design failure modes before success scenarios
Gideon Awolesi ·
A demo is a simulated environment. No matter how much you try to simulate, you choose the input, and you know the expected outcome based on the inputs you created. On the other hand, real users paste in messy data and take answers at face value, thinking it's the correct one.
01Start with the failure modes, then the happy path
A lot of teams spend all their design time on the happy path, where it's all peachy and input and output correlate beautifully. For me, I start with the failure modes: pitfalls the user can fall into, recovery steps, and how they can get help if needed.
The happy path is the flow everyone on your team has already seen working in the demo. The failure modes usually get far less attention, and they're the screens real users will run into in production.
02Your users just froze on your AI chat screen. What do you do?
So a client we worked with had this content creation tool, AiListz, that opened up to a blank prompt box, and new users didn't know what to do to start using it. A blank prompt box is a failure mode too, even though nothing technically went wrong.
We designed the onboarding, the conversational guidance, and the controls, so people could understand what an AI action would do before they used it. We also designed the guidance and chat assistance to answer common questions, so new users didn't need to contact support to get started.
03Telling people they're dealing with AI
AI Act Article 50 has applied since 2 August 2026, so people have to be told when they're interacting with an AI system. From 2 December 2026, generative systems that are already on the market also have to mark the content they produce.
So you need to design how that disclosure looks in your product. It has to be clear, but it shouldn't make people distrust a feature that works. It's also the natural place to let people know how they can check an answer, and what to do when the model gets it wrong.
04How you can map the failure modes in your own AI feature
Here's how YOU can do this for one AI flow in your product:
- Collect example inputs from real users, including the messy ones, and run the flow with them instead of the inputs you used in the demo.
- Write down every point where the output is wrong, unclear, or in a format the user can't use.
- For each one, design what the user sees, how they can recover, and how they can get help if they need it.
- Check that the AI disclosure is clear at the points where people act on the answer.
It's important that you test this with people from your user group, not with your team. Your team already knows what a good answer looks like, so they'll fill in gaps that your users can't.
05What you get in our free AI audit
If you've got an AI feature in your product, send us one AI flow and one question. Within 5 business days, you'll get a free recorded walkthrough and a one-page fix list, and we'll tell you straight if a bigger project makes sense.