Playbooks
Expert analysis flowcharts Sundial uses for common questions like metric change, funnels, and behavioral analysis.
Playbooks encode best-in-class ways of answering common analytical questions. Instead of letting the model invent a new approach every time, a playbook is a flowchart of the analysis a skilled analyst would run: which checks to make first, which assumptions are safe, and which follow-ups need confirmation.
Sundial ships a growing library of playbooks for the most common analyses. Your team can also create and curate its own. Playbooks are useful for both data scientists and business partners; they do not replace deep, context-heavy work by skilled analysts on the hardest questions.
Where to find them
- At answer time: the Analysis Agent matches your question to a relevant playbook when one exists.
- For curation: create and manage playbooks in the Context Engine, in the UI or via Git.
Example playbook types
| Playbook | Typical question |
|---|---|
| Metric change | What drove the increase (or drop) in a metric? |
| A/B test | How did a change to onboarding (or another experiment) perform? |
| Funnel | Where do users drop off in a conversion path? |
| Aha moment | Which early behaviors best predict later retention? |
| Behavioral analysis | What are users doing on the platform, and how do behaviors relate to retention? |
Metric change (example)
A metric-change playbook follows a structured path, for example:
- Is the move above or below expectations?
- Could seasonality explain it?
- Is there a mathematical relationship (for example revenue as price × quantity)?
- Which dimensions or segments disproportionately explain the change?
- What related metrics and second-order effects should you look at next?
The answer usually leads with a clear summary (for example, a pricing effect concentrated in certain plans), then shows the supporting breakdowns: contribution of price vs quantity, dimensional cuts, and related metrics such as expansion or churned ARR when a pricing change is involved.
Behavioral analysis (example)
For questions like “what are users doing on the platform?”, playbooks distinguish safe assumptions from ones that need confirmation. The agent may ask you to confirm which actions matter most, then run an analysis that typically includes:
- Prevalence vs frequency: what share of active users do a behavior, and how often.
- Change over time: whether behaviors are rising or falling.
- Stickiness: whether doing a behavior today predicts doing it again later.
- Correlation with retention: which behaviors associate with users staying.
Why playbooks matter for trust
Without playbooks, an LLM often invents assumptions, some harmless, some not. Playbooks encode which assumptions are safe to make and which should be confirmed with the user, and they produce repeatable structure so similar questions get similar rigor across your company.