Anaplan AI: Give it full model context to stop invalid formula suggestions

Question

Problem: Co-modeler often suggests formulas that don't work - invalid syntax, or timeouts requiring manual copy/paste that still fails. This forces users to debug the suggestion and explain the model's structure back to the tool (often having to leverage another AI like Chat or

Claude) instead of just applying a working solution .

Ask: Make Co-modeler fully aware of the model it's operating in - its lists, modules, existing formulas, and dependencies - so it only proposes formulas verified to work in that context. If it lacks the confidence/context to guarantee a formula works, it should say so rather than

suggest one.

Why it matters: Less time spent debugging bad suggestions or explaining model structure back to the tool. More trust in acting on Co-modeler's output without double-checking it.

Principle: A recommendation tool should limit/avoid guessing - it should either know the answer or be upfront when it doesn't.

Kartik · Answer

I’ve run into a similar pattern where the formula itself can look reasonable, but it doesn’t account for the way the existing model is structured.

I think having access to the model’s modules, line items, dimensions, and dependencies would make a big difference. Even better would be some kind of validation step before the formula is suggested or applied.

If Co-Modeler could check the proposed formula against the actual model context first, it could distinguish between “this should work” and “I don’t have enough context to verify this.

jmeyer · Answer

Agreed - thank you. I have run into it suggesting functions not available, as well as using existing functions incorrectly.

For instance, [LOOKUP: PARENT('Hierarchical Line Item')] when it should have instructed me to have a different line doing the parent calculation first, and then using that line item as the lookup.

jepley · Answer

Confirmed— I have run had this issue as well. Suggesting various ways to get "Random" and it said it was going to try functions not available. If I asked it to double check it would realize that it couldn't run, otherwise CoModeler would try to implement and then come back with the issue after spinning a while.