Getting the most out of your 7en agent comes down to a few core areas: clear instructions, high-quality data, and continuous improvement. This page covers the key practices that separate great agents from mediocre ones.
Your agent's System Prompt is the single biggest lever for improving response quality. A well-written prompt produces consistent, on-brand answers. A vague one produces unpredictable results.
Do:
Avoid:
Your agent answers from what you give it. Garbage in, garbage out.
Use accurate, up-to-date sources. Stale documentation causes the agent to give outdated answers. Set a regular review schedule for your knowledge sources — especially for pricing pages, product specs, and policy documents.
Be specific, not broad. A focused help center article outperforms a 200-page PDF where 90% is irrelevant. Break large documents into topic-specific chunks where possible.
Remove conflicting information. If two sources say different things about the same topic, the agent may hallucinate a blend of both. Audit your sources regularly from the Knowledge Management section.
Prefer structured content. Numbered steps, bullet lists, and clear headings are easier for the AI to extract and present accurately than dense paragraphs.
Different tasks benefit from different models. As a starting point:
| Use case | Recommended setting |
|---|---|
| Customer support / FAQ | Lower temperature (0.2–0.4), faster model |
| Sales or lead generation | Medium temperature (0.5–0.6) |
| Creative or long-form content | Higher temperature (0.7–0.9) |
| Technical documentation | Lower temperature (0.1–0.3) for precision |
You can change the model and temperature at any time from Model Settings without retraining.
Never deploy an untested agent. Use the Playground to run your agent through real scenarios before going live.
Test for:
Keep a short list of 10–15 test questions that cover your most common and most critical use cases. Re-run them after every major knowledge base update.
Launching your agent is the beginning, not the end.
Review conversations regularly. The Conversations section shows what real users are asking. Look for patterns — repeated questions with poor answers signal gaps in your knowledge base.
Update sources after product changes. Whenever your product, pricing, or policy changes, update the relevant knowledge sources and retrain the agent the same day.
Refine your system prompt over time. If you notice the agent going off-topic or using the wrong tone, tighten the instructions rather than adding more data.
Start building intelligent AI agents to engage with customers.
