Good prompts help Anaplan agents deliver optimum results. Follow the recommendations below to prompt effectively.

  • You don't need to state a role or persona. Agents know their role. Focus on what you want to do. Avoid phrases such as Act as an Anaplan Model Builder... or You are a Senior Financial Analyst....
  • Anaplan AI will direct your prompt to the relevant agent. You don't need to tell it which agent to use.
  • Be upfront with context. If you need to give a lot of context, consider using a specification file. This approach has the added advantage of then being available for use in new conversations. Learn more about specification files for CoModeler.
  • Be clear on inputs and outputs. For outputs, include details that’ll help shape the answer to your needs. For example, ask the agent to explain its reasoning and any assumptions made.
  • With CoModeler build structure first, formulas last.
  • Rather than asking for everything at once, for anything sizable, consider a systematic, ordered approach where you verify as you progress.
  • Consider how you can effectively validate outcomes. Do you need to take a more measured, phased approach?
  • Structure requests as logically as possible using Anaplan or application terminology. For example: Add a list named Size and populate it with 5 items from 100g to 500g. Add All Sizes as the top-level. is a prompt that includes the top-level list concept.
  • Be clear and explicit, and as specific as possible. For example, if asking a sales-related question in Analyst, be clear on the exact time period, customer, location, or product name. Vague detail leaves room for interpretation and guesswork.
  • Use agents to understand your starting point, and to provide recommendations.
  • Use agents to explore options before you commit to anything.
  • Consider including an example. An example can save you from having to provide the details in your prompt. For example, in CoModeler, you could include Build it like ModuleD.
  • Try different vocabulary if what you are using isn’t giving results.
  • If you have a preferred or better method of implementing a task, include it in your prompt.
  • When faced with a limitation, be prepared to do some things manually. However, you can ask the agent about what steps you need to take.