# How to Use Contextual Memory in a ZazzyAgent AI Agent Customers normally don't communicate in isolated questions. A conversation may look like this: > Customer: I'm interested in the enterprise plan. > AI: Sure. How many users do you have? > Customer: About 50. > AI: ... The last response needs to understand what came before it. That's where **Contextual Memory** becomes useful. ## What is Contextual Memory? Contextual Memory allows an AI Agent to use previous conversation messages when generating its current response. Instead of only looking at: > The latest customer message the AI can also consider: > Relevant messages from earlier in the conversation. ## Why does this matter? Without context, the AI may have trouble understanding short follow-up messages. For example: > Customer: Do you offer annual billing? > AI: Yes. > Customer: How much? The AI needs to understand that: > "How much?" refers to annual billing. ## Enable Contextual Memory Open your AI configuration. Find: **Enable Contextual Memory** Turn it on for the Agent that needs conversational history. ## Configure the amount of context [ZazzyAgent](http://zazzyagent.com) lets you control how much previous conversation context the AI can use. The right amount depends on your use case. A simple FAQ assistant may need relatively little context. A support Agent handling long conversations may benefit from more. ## Why not use the maximum amount? More history isn't automatically better. Older messages may contain: * Outdated information * Irrelevant questions * Changed requirements * Previous decisions that are no longer applicable The goal is useful context, not maximum context. ## Example: Sales conversation Customer: > I'm looking for automation for a company with 50 employees. AI: > What are you looking to automate? Customer: > Customer support. AI: > Approximately how many conversations do you receive each month? Customer: > Around 5,000. Each answer depends on what the customer said earlier. Contextual Memory helps the AI maintain that continuity. ## Example: Support conversation Customer: > My order hasn't arrived. AI: > What's your order number? Customer: The AI needs to understand that: > 12345 is the order number associated with the delivery problem. ## Example: Booking Customer: > I'd like an appointment on Friday. AI: > What time would you prefer? Customer: > Afternoon. Later: > Actually, can we move that to Monday? The AI needs enough context to understand what: > "that" refers to. ## Contextual Memory vs Knowledge These are completely different. ### Contextual Memory Answers: > What are we talking about right now? ### Knowledge Answers: > What does the business know about this subject? For example: Customer: > What's your refund period? Memory tells the Agent that the customer is discussing a purchase. Knowledge contains: > Your actual refund policy. ## Contextual Memory vs Custom Fields These are also different. Memory is primarily about the conversation. Custom Fields store structured customer information. For example: ```text Contextual Memory "I'm currently discussing an enterprise plan." Custom Field Company Size = 50 ``` If the information needs to remain available as structured customer data, save it to the appropriate field. See [ZazzyAgent Custom Fields Guide](https://blog.zazzyagent.com/zazzyagent-custom-fields-guide). ## Memory and lead qualification Customers often provide multiple pieces of information naturally. For example: > I'm Rahul from ABC Ltd. We're looking for support automation and handle about 10,000 conversations a month. The Agent can use the current conversation to avoid asking questions the customer has already answered. ## Memory and human handoff Contextual understanding can also improve human handoff. A customer may explain a problem across six messages. A human shouldn't have to restart the conversation from: > What's the issue? The existing conversation provides useful context. ## Memory isn't permanent storage Don't rely on conversational memory to store important long-term information. If you need: > Budget = ₹75,000 later, save it as structured data. If you need: > Preferred location = Ahmedabad later, save it in a customer field. ## Too little context The AI may repeatedly ask: > What are you looking for? even though the customer explained it earlier. ## Too much context The AI may pay attention to something from an old part of the conversation that is no longer relevant. For example: > Customer initially wanted Product A. Later: > Customer changed their mind and wants Product B. An excessive context window may make the old preference unnecessarily influential. ## Test contextual memory Run a multi-message conversation. For example: ```text I'm interested in your enterprise plan. ↓ I'm looking at around 50 users. ↓ How much would that cost? ``` Check whether the final question is understood correctly. ## Test topic changes Try: > What plans do you offer? Then: > By the way, where are you located? Then: > Can I book a demo? You want enough context for continuity without confusing separate topics. ## Memory and AI Agents Contextual Memory is particularly useful for: * Sales conversations * Customer support * Lead qualification * Booking conversations * Longer troubleshooting sessions It is less important for a bot that only answers isolated one-message questions. ## Common problems ### AI repeats questions Check whether Contextual Memory is enabled and whether the configured context is sufficient. ### AI uses old information The context may be too broad for the type of conversation you're handling. ### AI forgets a recent detail The context may be too limited. ### Important customer information disappears later Store the information in Custom Fields instead of relying only on memory. ## A useful three-layer model ```text Contextual Memory "What are we talking about?" Knowledge "What does the business know?" Custom Fields "What customer information should we retain?" ``` Using the right layer for the right kind of information produces much more reliable AI behavior.