AI pitfalls
What to avoid when designing AI solutions.
When designing AI solutions, it’s essential to be aware of common pitfalls that can negatively impact user experience, security, and trust.
Recognising and proactively addressing these pitfalls helps maintain high-quality AI solutions that are accurate, transparent, secure, and user-friendly.
The references below connect each example to the relevant experimental AI UX Playbook principles and guardrails. Use the status shown in the Playbook to understand the maturity of each item.
| Pitfall | Example | Prevention | Playbook reference |
|---|---|---|---|
| Hallucinated information | AI shows incorrect vessel arrival times without actual port data. | Always validate AI outputs against authoritative backend data sources, clearly marking or excluding uncertain data. | P3: Be honest about what it knows; G6: Validate factual outputs or clearly mark uncertainty. |
| Hidden automation | AI automatically approves shipment rerouting without clearly informing the logistics coordinator. | Require explicit user confirmation for critical automated actions and follow clear authorisation policies. | P4: Determine the appropriate level of human involvement; G4: Allow confirmation or overriding; G5: Require explicit confirmation. |
| Unexplained AI decisions | AI automatically approves shipment rerouting without clearly informing the logistics coordinator. | Include clear and accessible explanations for AI-driven decisions, linking to detailed reasoning and underlying data. | P3: Be honest about what it knows; G7: Make AI decisions explainable in one action. |
| Persistent biases | AI repeatedly recommends certain suppliers due to biased historical purchasing data. | Regularly audit training data for fairness, diversify inputs, and transparently disclose recommendation criteria. | G13: Assess bias before deployment and document the criteria. |
| Overgeneralization | AI suggests identical inventory replenishment levels regardless of seasonal demand fluctuations. | Contextualise AI recommendations based on roles, logistics tasks, real-time data, and specific contexts. | P1: Earn its place; P3: Be honest about what it knows. |
| Data privacy ambiguity | AI uses customer data for personalized tracking without clearly communicating privacy implications. | Explicitly request user consent, transparently communicate data usage, and provide clear privacy preference settings. | G14: Verify permissions before revealing sensitive data. |
| Unrecoverable errors | AI misinterprets a user’s voice request without clearly guiding the user on how to correct it. | Design clear interactions with explicit prompts and easy correction or error recovery paths. | P5: Know its limits, fail safely; G11: Provide a recovery or correction path. |
| Friction in escalation paths | Users struggle to smoothly transition from AI chatbot interactions to human support. | Clearly and proactively present easy escalation options and ensure smooth hand-off experiences. | G12: Allow human escalation without losing context. |
| Security shortcuts | AI chatbot inadvertently reveals sensitive shipment information without proper authentication. | Enforce secure, role-based authentication and clearly indicate when sensitive actions require additional security steps. | G14: Verify permissions before revealing sensitive data. |
| Inconsistent AI personality and tone | AI conversations abruptly shift from formal to overly casual, confusing users and reducing trust. | Define and consistently apply a conversational style guide aligned with user expectations and company tone. | G15: Keep the AI identity consistent and never impersonate a human. |
Please write to us on our Teams Channel. We encourage and welcome any type of contribution and feedback.