PART 2: How to Improve the Performance of Your Sales & Support Agent

Modified on Wed, 15 Jul at 7:20 PM

Launching your Sales or Support Agent is only the beginning. To ensure your AI agents continue delivering accurate, helpful, and high-quality responses, you should regularly review their conversations and refine their knowledge base.

In Part 1, you rehearsed with your new hire before their first shift. Part 2 is the ongoing coaching after they start. Even a great employee gets better with regular feedback — a manager who listens in on calls, spots the questions that tripped them up, and fills those gaps in training. Unlike a one-time setup, optimization is an ongoing process: as your customers ask new questions and your business evolves, updating your Product Knowledge Base continuously improves your agents' performance.

Why Continuous Optimization Matters

Every customer interaction provides valuable insights into how your AI agent performs. By reviewing conversations regularly, you can:

Identify questions the agent couldn't answer accurately
Discover missing information in your Product Knowledge Base
Improve response quality and consistency
Increase customer satisfaction
Ensure your AI agent stays aligned with your latest business information

? Best PracticeSchedule regular reviews — weekly or bi-weekly — to keep your AI agent performing at its best. Put it on your calendar like any other business routine; 20-30 minutes per review is usually enough.

The Ongoing Optimization Loop

Step 1: Review Customer Conversations

Navigate to the Sessions tab in your FlexiAgent dashboard. The Sessions page contains the complete history of conversations handled by your AI agents. Here, you can review:

Customer questions
AI-generated responses
The agent's reasoning process
Complete conversation history

Reviewing these sessions helps you understand how your AI agent interacts with real customers — which is often quite different from your test questions.

The Sessions tab showing the searchable history of conversations handled by your AI agents

Step 2: Identify Areas for Improvement

As you review conversations, pay attention to responses that are:

Incomplete
Incorrect
Unclear
Missing important business information
Not aligned with your preferred communication style

Create a list of questions where the agent's response could be improved. These questions will tell you exactly what information should be added to your Product Knowledge Base.

Step 3: Update Your Product Knowledge Base

When you identify gaps in the agent's responses, update your Product Knowledge Base by adding:

New frequently asked questions
Improved answers
Updated pricing
New services or products
Business policies
Additional product details

After updating the knowledge base, replace the existing file in the Vault with the latest version. Future conversations will automatically use the updated information.

⚠️ Reminder: Replacing means delete first, then uploadFlexiAgent doesn't support editing an uploaded file directly. To replace your knowledge base: delete the old file from the Vault, then upload the updated .md version, and let the enrichment process complete. See the full walkthrough in the Related Articles below.

Build Your Knowledge Base Over Time

Your Product Knowledge Base doesn't need to be perfect on day one. Instead, treat it as a living document that grows alongside your business. Over time, continue adding:

Frequently asked customer questions
Better responses
Common objections
Product updates
Support scenarios
Business processes

The more comprehensive your knowledge base becomes, the more accurate and helpful your AI agents will be.

✅ Why this mattersThink of it as compound interest for your support quality. Each review adds a few answers; each added answer removes a future failed conversation. A knowledge base that grows by just 3-4 real customer questions every week becomes a comprehensive business brain within a few months — one that answers questions your day-one version couldn't have imagined.

Create a Question & Answer Library

One effective way to improve your AI agents is by maintaining a structured Question & Answer (Q&A) library. For example:

Customer Question	Recommended Answer
What are your business hours?	Provide your operating hours and contact information.
What are your pricing plans?	Explain available pricing options and packages.
Do you offer refunds?	Describe your refund policy.
How can I contact support?	Provide available support channels.

As new questions arise, continue adding them to your knowledge base. Over time, this creates a comprehensive resource that helps your AI agents answer customer queries more effectively.

Review Conversations Regularly

Make reviewing AI conversations part of your regular workflow — for example, a weekly, bi-weekly, or monthly review. During each review:

Read recent customer conversations.
Identify weak responses.
Update your Product Knowledge Base.
Upload the latest version to the Vault.

Small, consistent improvements will significantly enhance your AI agent's performance over time.

Use Customer Ratings to Prioritize Improvements

If your Sales or Support Agent is integrated with a customer messaging platform such as:

Freshchat
Intercom
Other live chat or customer support platforms

Enable customer satisfaction (CSAT) ratings whenever possible. Customer ratings help you quickly identify conversations that require attention.

Focus first on chats that received:

⭐ 1 Star
⭐⭐ 2 Stars
⭐⭐⭐ 3 Stars

These conversations often reveal areas where your AI agent could provide clearer, more accurate, or more helpful responses.

✅ Why this mattersRatings turn your customers into your quality-control team. Instead of reading every conversation hoping to find problems, low ratings point you straight to the interactions that disappointed someone — so your limited review time always goes where it improves the experience most.

After reviewing these interactions:

Identify what information was missing or unclear.
Update your Product Knowledge Base.
Replace the existing knowledge base file in the Vault.
Verify the improved response in future conversations.

Optimization Checklist

Use the following checklist as part of your ongoing optimization process:

Review recent conversations in the Sessions tab.
Identify questions with unsatisfactory responses.
Add missing information to your Product Knowledge Base.
Expand your Q&A library with new customer questions.
Replace the updated knowledge base in the Vault (delete old → upload new).
Monitor future conversations for improvements.
Prioritize low-rated customer chats for review.
Repeat this process regularly.

Common Situations & Quick Fixes

❓ "There are too many sessions — I can't review them all."You don't need to. Prioritize: (1) low-rated conversations if CSAT ratings are enabled, (2) the most recent sessions, and (3) unusually long conversations — length often signals the customer had to ask the same thing multiple ways.

❓ "I keep fixing the knowledge base, but the same question still fails."Verify the replacement actually took effect: the old file was deleted before the new upload, the enrichment process completed, and you're testing in a New Chat. Also check that the answer in your file is clearly worded under a relevant heading — vague or buried information is harder for the agent to retrieve.

❓ "My agent isn't connected to Freshchat or Intercom — can I still prioritize reviews?"Yes. Without ratings, use the Sessions tab and focus on recent and long conversations. You can still apply the same loop: spot weak answers → update the knowledge base → replace in the Vault → verify.

❓ "How often should I really do this?"Start weekly while the agent is new — that's when you'll find the most gaps. Once responses stabilize, move to bi-weekly or monthly. The key is consistency, not intensity: a 20-minute review every week beats a 3-hour audit once a quarter.

❓ "My knowledge base file is getting big — is that a problem?"A growing file is a good sign, but keep it organized: clear headings, one topic per section, and a Q&A format for common questions. Well-structured information stays easy for the agent to retrieve no matter how much you add.