This article provides some best practices when using Omni Insights, locate the practices that are most relevant to your workflow!
- Start new AI sessions wherever possible rather than going back to a past session in your history
- Similar to the way it works with Claude, long sessions can consume more credits as the context accumulates
- Omni surfaces a little yellow dot to try to encourage users to start new sessions when the context window starts to get full
- Be specific with your questions/prompts from the start – the more context you give, the better and faster the answer. Instead of asking "show me student data," try "show me a breakdown of student sign-ups by major for the past 90 days, grouped by week." Specificity reduces back-and-forth and gets you to a usable result in fewer exchanges.
- Ask for one thing at a time – It's tempting to ask a multi-part question in one message, but breaking complex requests into sequential steps produces more accurate results and uses credits more efficiently. Build your analysis incrementally: get the base data first, then ask for visualizations.
- Save and create a template of your best reports and dashboards – when you produce the report you need, save it. Reusing and returning to a dashboard/workbook is faster than starting from scratch and produces consistent results.
- Use precise date ranges and filters upfront– asking for data with open-ended time ranges (e.g. "all time") pulls more data than necessary and can produce unfocused results. Always specify the time range, cohort, or segment you care about from the start. This narrows the analysis and reduces unnecessary processing.
- Iterate on an existing report rather than rebuilding– if you have a report you like, use the AI to modify or extend it rather than generating a new one from scratch.
- Know when Omni AI Assistant isn't the right tool
- Omni is best for exploratory analysis, one-off questions, and building new reports. For reports you run on a recurring weekly or monthly basis, it's more credit-efficient to save and schedule them directly in Omni rather than re-prompting the AI each time. Build it once with AI help, then schedule it.
- Omni is a reporting tool, not a stand-in for ChatGPT or Claude. It does not scrape the web or other resources beyond your data and any additional context you give it. Additionally, it should not be used for questions unrelated to data analysis.
- Review before you iterate– before sending a follow-up prompt, read the full response. Many users send a correction immediately after a response without fully reading it, only to find the AI already addressed their concern. A quick read before each follow-up avoids redundant exchanges.
- Be explicit about the output format– tell the AI how you want the output presented: "as a table," "as a bullet summary," "as a chart," or "in plain text." Without this, the AI may produce a format you have to ask it to redo, costing an extra exchange.