What are some common pitfalls to avoid when implementing OKRs in generative AI initiatives?

Other than general guidelines around good OKR's, for Gen AI in particular, the space is moving very rapidly. We've found that quarterly KR's can change wildly as we experiment and iterate. Keep things flexible and fluid and expect to end somewhere that may be very different than where you started in terms of results.

- Avoid nested OKRs. Keep OKRs simple and measurable. Striving for simple OKRs ( e.g : Enhance customer engagement through personalized marketing content keep the OKRs on click rate , % growth or no. of click rate on engagements.) so we know model is working.
- Ensure OKR align to broader org vision & strategy
- Unrealistic OKRs – e.g not able to source data or not enough data for the AI use case to be trained with.
- Bias and Accuracy – Ensure Data is diverse and accurate to dial down biases and reduce hallucinations by LLM models
- Always strive for transparency in AI models. ( e.g three explainability tools (e.g., SHAP, LIME) for AI models in next 3-6 months )

When implementing OKRs in generative AI initiatives, a few common pitfalls can derail progress. One is setting vague or unrealistic objectives—AI capabilities evolve rapidly, so OKRs must be ambitious yet achievable. Another is focusing too much on output metrics (e.g., number of models trained) rather than impact-driven outcomes (e.g., improvements in content quality or user adoption). Misalignment between technical and business teams can also lead to disconnected goals, so cross-functional collaboration is key. Finally, failing to iterate on OKRs as AI models learn and improve can limit adaptability. The best approach is to keep objectives flexible, impact-focused, and aligned with broader organizational goals. Curious to hear what challenges others have faced.

I've been reflecting this last quarter on the overall concept of AI literacy and transforming information technology management in general. This by observing our multiple pilots, especially LLM model usage, as well as digital literacy campaign discussions.

I am developing a theory that I am not yet convinced of that we need to think about objective setting for these models more following an HR practice and general goal setting.

That would imply that one should set the same objectives and key results for the model as you would for the team of people, and the model and the people share the OKRs.