Optimizing Agent Performance
1. Optimizing Agent Performance
Your agent works, but we can make it better. Small refinements to instructions can dramatically improve response quality. Let's optimize it. I'm in the Agents tab. Click on Sales Intelligence Agent. Then we're going to click Edit. Now we're going to enhance the orchestration instructions. Let's refine these instructions to make the agent more effective. Think about who will be using the agent and what they will need. In this case, it's sales teams who need quick insights to make decisions. When you define your agent's audience and their goals, the agent can tailor its responses appropriately. At the top of the instructions, we're going to replace what we have. You're going to add, you are a sales intelligence agent for a B2B software company. Your primary users are sales representatives, sales managers, and revenue operations teams. They need fast, accurate, data-driven insights. This gives the agent role context and audience awareness. Notice how we're being specific about the users and what they value. Our users don't want lengthy explanations, they want fast insights they can act on. Add this about the metrics. When querying metrics, always provide context with your numbers. Don't just say the win rate is 67 percent, say that win rate for Enterprise Suite is 67 percent based on two wins out of three closed deals. This ensures complete responses. Add response format guidance. Structured response for quick scanning, lead with direct answers to the question, then provide supporting details. Write naturally without bullet points. This makes responses more readable. Why does this matter? Numbers without context are just data. Your sales team needs to understand what those numbers represent. Think about what additional information would help someone make a decision and instruct your agent to include that context. Then we're going to change the boundaries. Do not speculate about whether pending deals will close, only report on factual data from the metrics and actual statements from conversations. If you don't have enough information, say so and suggest what additional information would be needed. This prevents unfounded predictions. Click 'Save. Now let's test the difference. Type what's our pipeline by stage. Compare on the response. It says, 'The data shows we have strong closed deal performance with 630,000 in total value from six successful deals, while only losing one smaller deal of 25,000. Our average closed deal size is 105,000, indicating healthy deal quality across the pipeline.' Notice how it's more narrative and contextual? That's the improved instructions working. One more optimization, add data limitation indicators. When analyzing patterns or making recommendations, indicate if the data is limited. For example, if you're seeing a pattern across only two deals, note that the sample size is small. This prevents overconfident conclusions from limited data. Your agent should tell users when it's working with limited information so they can interpret this insight appropriately. Click 'Save'. Your agent is now optimized for use. Instructions provide clear guidance on response structure, context, and boundaries. Now let's test these optimizations. In the agent playground, type, 'What patterns do you see in our lost deals?' Notice how the agent responds. It identifies the small-biz solutions was lost due to price concerns. But then it mentions that with one lost deal in that dataset, it's difficult to establish a reliable pattern. This is exactly what we wanted. The agent is being transparent about data limitations rather than making overconfident claims from limited evidence. Let's recap what you've learned in this video. You refined orchestration instructions by defining your agent's audience and their needs. You added response format guidance to make answers more readable and actionable. You set clear boundaries to prevent speculation on uncertain data and you added transparency requirements, so your agent indicates when it's working with limited information. These refinements transform a working agent into a production-ready one. In the next video, we'll wrap up what you built.2. Let's practice!
Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.