Hey, it is Vlad. Voice AI is one of the features I am most excited about in Exoserva, and Voice Analytics is the dashboard that shows you exactly how well it is working. Here is the thing – most voice AI systems just throw every single customer call at an expensive AI model and hope for the best. That gets really expensive really fast. So I built a smarter approach: a three-tier system where simple questions get instant answers, medium questions use pre-built templates, and only the truly complex conversations go to full AI processing. Think of it like a restaurant: not every order needs the head chef. The sous chef handles most plates, the prep cook handles the basics, and the head chef steps in only for the special dishes. This dashboard shows you how that system is performing and how much money it is saving you – and trust me, the savings add up fast.
Estimated time: 15 minutes
Before You Begin
- An active Exoserva account with Owner or Admin role – Voice Analytics shows cost data and system performance metrics that are typically reserved for business owners and administrators. If you are a manager or dispatcher and cannot see this page, ask your account owner to grant you Admin access.
- Voice AI enabled and configured – this is a separate setup process. If you have not done it yet, start with the “Setting Up Voice AI” guide first. Voice Analytics only has data to show if Voice AI is actually running and handling calls.
- At least a few days of voice message activity – just like with email tracking, the analytics need real data to be meaningful. If Voice AI has only handled 3 calls, the percentages and trends will not tell you much. Give it at least a week of normal call volume, and ideally two weeks, before using this dashboard for serious decision-making.
- A basic understanding of what “latency” means – in simple terms, latency is how long it takes for the system to respond after a customer says something. Think of it like the pause between asking someone a question and getting their answer. Lower latency means faster responses, which makes for happier customers. We will explain the specific thresholds for each tier in this guide.
Step 1: Navigate to Voice Analytics
Look at the left sidebar of your Exoserva dashboard and find the section labeled Analytics. Click “Voice Analytics” underneath it. If your sidebar is collapsed to just icons, look for the phone icon with a small chart symbol. The Voice Analytics page will open and you will immediately notice several things in the header area.
First, there is the page title with a phone icon next to it. Right beside that, you will see a small green pulsing dot with the word “Live” – this is an important indicator that tells you data is being captured in real time. Every time a customer interacts with your Voice AI, the analytics update automatically. You do not need to refresh the page or wait for a daily report. Next to the Live indicator, you will see a Period Selector with three buttons: 7d (last 7 days), 14d (last 14 days), and 30d (last 30 days). Click any of these to change the time window for all the charts and metrics on the page. In the top-right corner, there is a Refresh button – you normally do not need this since data updates live, but it is there in case you want to force a complete recalculation of all metrics.
Below the header, you will see two tabs: Overview (selected by default) and Logs. The Overview tab is the analytics dashboard with all the charts, cost savings, and AI insights. The Logs tab shows individual voice message records so you can see exactly what customers said and how the system responded. We will cover both tabs in this guide. If your Voice AI has not handled any calls yet, you will see a placeholder message encouraging you to configure your Voice AI agent – do not worry, once calls start coming in, this dashboard will come alive with data.
Tip: The green “Live” indicator means new voice events are reflected on this page as they happen – you do not need to manually refresh. However, if you have had this page open for a very long time (several hours) and want to make sure everything is fully up to date, clicking the Refresh button will force a complete data reload. This is rarely necessary but good to know about.
Step 2: Understand the Three-Tier System
Before we dive into the charts and numbers, I need to explain the concept behind everything on this page. The Voice AI system does not treat every customer message the same way. Instead, it uses a three-tier architecture – think of it like a triage system in a hospital emergency room, where minor issues are handled quickly by nurses, moderate cases go to doctors, and only the most complex cases go to the specialist surgeon. Each tier has a different speed, capability, and cost.
Tier 1: Instant (shown in green with a lightning bolt icon) – These are the fastest responses, with a target response time of under 50 milliseconds (that is 0.05 seconds – literally faster than a blink of an eye). Instant responses handle common, simple questions that have pre-configured answers stored right on the server. Questions like “What are your business hours?”, “Where are you located?”, or “What services do you offer?” do not need AI to answer – the answer is always the same. By handling these locally, the response is instantaneous and costs you absolutely nothing. Zero API fees.
Tier 2: Template (shown in blue with a document icon) – These responses use pre-built templates and take under 200 milliseconds (0.2 seconds – still very fast, barely noticeable). Template responses handle structured requests like appointment scheduling (“I need to book a service for next Tuesday”), service inquiries (“How much does a furnace tune-up cost?”), or status checks (“Is my technician on the way?”). The system recognizes the intent, fills in the right template, and responds. Very small API cost – pennies.
Tier 3: AI (shown in purple with a computer chip icon) – These are full AI-powered conversations using Claude, our advanced language model. Response time is under 2000 milliseconds (2 seconds). AI responses handle complex, nuanced situations where the customer says something unexpected, needs detailed troubleshooting help, or the conversation requires understanding context and making judgment calls. This tier is the most capable but also the most expensive per message.
The magic of this system is that it automatically routes each incoming message to the cheapest tier that can handle it well. If a simple “What are your hours?” question can be answered instantly at zero cost, why would you pay for full AI processing? The analytics dashboard shows you how well this routing is working and where you might be able to save even more money.
From Vlad: Let me share the math that convinced me to build the three-tier system instead of just sending everything to AI. A typical home service company gets 50-100 customer interactions a day through voice. At $0.03 per AI call, that is $1.50 to $3.00 daily – seems small, right? But multiply that by 365 days, and you are looking at $550-$1,100 per year for one company. Now multiply that by the thousands of businesses on our platform, and it becomes millions of dollars. The three-tier approach handles 60-70% of all messages instantly at zero cost, and another 15-20% through cheap templates. Only 10-25% of messages actually need the full AI brain. That saves our customers enormous amounts of money while keeping response quality exactly the same – because the AI still handles everything it needs to.
Step 3: Review the Distribution Ring and Cost Savings
Now that you understand the three tiers, let us look at how YOUR specific messages are distributed. At the top of the Overview tab, you will see a Hero Section with two key visualizations sitting side by side.
On the left is the Tier Distribution Ring – this is a donut-shaped chart (a circle with a hole in the middle, like a bagel) that shows what percentage of your messages are handled by each tier. The ring is divided into colored sections: green for Instant, blue for Template, and purple for AI. If you see a lot of green, that means most of your messages are being handled instantly at zero cost – great! If you see too much purple, it means a lot of messages are going to expensive AI processing, and you might want to create more Instant or Template responses to bring that cost down. In the center of the ring, you will see the Total Messages count – the total number of voice interactions during your selected time period.
On the right is the Cost Savings card, and this is the number that will make you smile. It shows two amounts: AI Cost Saved (how much you WOULD have spent if every single message had been processed by the expensive AI tier) and Actual Cost (what you actually spent, since most messages were handled by cheaper tiers). The difference between these two numbers is your savings – real money that stayed in your pocket thanks to the tiered system.
A healthy distribution typically shows 60% or more of messages handled by the Instant and Template tiers combined. If your AI tier is handling more than 40% of messages, that is a signal to review which common questions could be moved to Tier 1 (Instant) with pre-configured answers. We will talk about how to identify these opportunities in the AI Insights step (Step 4).
Step 4: Read AI-Generated Insights
Below the hero section, you will find the AI Insights Card – this is where the system analyzes your voice data and gives you specific, actionable advice in plain English. Think of it as having an operations consultant who reviews your call patterns every day and tells you exactly what to optimize. The card displays up to three contextual insights at a time, each labeled with a type and severity level.
Here are the types of insights you might see and what each one means: Excellent Efficiency or Good Efficiency (shown in green) means your instant response rate is meeting or exceeding the target – your system is handling most simple questions automatically and cheaply, which is exactly what you want. Optimization Opportunity (shown in amber) means the system has identified messages that are going to the AI tier but could potentially be handled by a cheaper tier – it is suggesting you create new templates or instant responses for these patterns. Template Suggestion (shown in blue) gets even more specific – it names particular customer intents (like “pricing inquiry” or “schedule request”) that appear frequently in Tier 3 AI calls and would be better served by pre-built templates. High AI Latency (shown in red) is a warning that AI response times are getting too slow, which could frustrate customers on live calls. And Cost Savings (shown in green) summarizes how much the tiered system has saved compared to running everything through AI.
These insights are not static – they update dynamically as your data changes. They compare your metrics against configurable thresholds: for example, an instant response rate above 60% earns an “Excellent Efficiency” badge, while an AI cost per message above $0.02 triggers an “Optimization Opportunity” suggestion. The insights are calculated from your actual performance data, so they are always relevant to your specific business, not generic advice.
Step 5: Monitor Today’s Live Stats
Below the AI Insights, you will find the Today’s Live Stats section. While the hero section and insights show data across your selected time period (7, 14, or 30 days), this section focuses exclusively on what is happening RIGHT NOW, today. Think of it like a sports scoreboard that updates in real time during the game – except this game is your daily business operations.
You will see four metric cards showing today’s activity. The Instant card shows how many Tier 1 messages have been handled today, what percentage of today’s total they represent, the average response time (latency), and an “On Target” badge when that latency is within the 50ms threshold. The Template card shows the same information for Tier 2 messages, with a 200ms latency target. The AI card shows Tier 3 messages with a 2000ms (2 second) target. And the Total Today card shows the aggregate count across all three tiers – this is your daily call volume at a glance.
Each card has a latency indicator that changes color based on performance: green means the tier is responding within its target time (great!), amber means it is slightly above the target (worth watching), and red means latency is significantly above the threshold (investigate immediately, because slow responses frustrate customers). Below the live stats, a Summary Stats section provides the full-period tier distribution as a stacked bar chart (a horizontal bar divided into green, blue, and purple segments) and a detailed cost savings breakdown comparing what you spent on local processing versus AI processing.
This section is most useful during business hours when calls are actively coming in. You can watch the numbers climb throughout the day and see in real time which tiers are handling the load. If you notice the AI tier card suddenly spiking in the late morning, it might mean customers are asking questions that your Instant responses do not cover yet.
Step 6: Analyze Response Latency by Tier
Speed matters in customer service. When someone calls your business, they do not want to wait 5 seconds for a response – by then, they are already frustrated. That is why we built detailed latency monitoring into Voice Analytics. The Response Latency section shows three cards, one for each tier, and each tells you exactly how fast (or slow) that tier is performing.
Each latency card displays the average response time in milliseconds compared to the tier’s target. A progress bar fills up based on how close the latency is to the target – if the bar is less than half full and green, you are well within the target. If the bar is full and turning amber or red, latency is approaching or exceeding the threshold. Here are the target thresholds you should know: Instant tier should respond in 50 milliseconds or less (that is 0.05 seconds – nearly instantaneous). Template tier should respond in 200 milliseconds or less (0.2 seconds – so fast the customer barely notices a pause). AI tier should respond in 2000 milliseconds or less (2 seconds – noticeable but acceptable for a thoughtful response to a complex question).
The latency values throughout the entire dashboard are color-coded consistently so you can spot problems at a glance. Green means 50ms or below (excellent). Yellow means 51-200ms (good for Template tier, a bit slow for Instant). Orange means 201-1000ms (acceptable for AI tier, too slow for everything else). Red means above 1000ms (over 1 second – needs attention).
If any tier consistently shows red latency indicators, it could mean several things. For the Instant tier, it might mean your pre-configured answers database is getting too large and needs cleanup. For the Template tier, it could mean a template is too complex. For the AI tier, it usually means the system prompt (the instructions you gave the AI about your business) is too long and needs to be shortened, or the AI is handling more calls simultaneously than the infrastructure can support.
Warning: Sustained AI latency above 2000ms (2 seconds) seriously degrades the customer experience on live calls. Customers will notice the pause and may get frustrated or hang up. If you see red latency on the AI tier card, take action quickly. The most common fix is shortening your system prompt – the set of instructions you give the AI about your business. Every extra paragraph of instructions adds processing time. Keep your system prompt focused and concise.
Step 7: Explore Intent and Category Trends
Now we get to some really interesting data. The Intent Distribution and Category Distribution sections show you WHAT your customers are actually talking about. This is incredibly valuable information that goes beyond just “how many calls did we get” – it tells you what your customers need, what questions come up most often, and where you should focus your resources.
The Intent Distribution chart shows a ranked bar chart of the top 5 intents detected across ALL voice messages (across all three tiers). An “intent” is what the customer is trying to do – for example, “schedule_appointment” means they want to book a service, “pricing_inquiry” means they want to know how much something costs, “service_status” means they are asking about an existing job. Each bar shows the intent name, the count of messages with that intent, and the percentage of total messages. This tells you what customers care about most – and that should inform how you run your business.
Next to the intent chart, the Category Distribution shows the top 5 categories handled specifically by the Instant tier (Tier 1). Categories are broader groupings like “business_hours”, “location”, “greeting”, or “service_areas”. This chart tells you which types of questions your Instant responses handle most frequently – essentially, which pre-configured answers are getting the most use. If a category appears here with a high count, it means that Instant response is saving you money every day.
Below the distributions, the Daily Trend chart shows a stacked bar graph with one bar per day across your selected time period. Each daily bar is divided into three colored segments: green (Instant), blue (Template), and purple (AI). This chart reveals patterns you might not notice otherwise – like higher AI usage on Mondays (when customers call with problems from the weekend), or seasonal spikes in certain types of inquiries. Understanding these patterns helps you plan ahead and optimize your Voice AI configuration for peak periods.
Step 8: Search and Filter Voice Logs
The analytics dashboard gives you the big picture, but sometimes you need to see the actual conversations – the exact words customers said and exactly how your Voice AI responded. That is what the Logs tab is for. Click the Logs tab at the top of the page (next to “Overview”) to switch to the individual message view.
The Logs view shows a chronological list of every voice interaction your system has handled. At the top, you will find a search bar where you can type to filter messages by content – for example, type “appointment” to see only messages where the customer mentioned appointments. Next to the search bar, you will see tier filter buttons labeled All, Instant, Template, and AI. Clicking one of these buttons narrows the list to only show messages handled by that specific tier. This is extremely useful for auditing – for example, click “AI” to see only the expensive AI-processed messages and check whether they actually needed AI, or could have been handled by a cheaper tier.
Each log entry shows you several pieces of information: a tier badge (color-coded green, blue, or purple to match the tier system), the detected intent (what the customer was trying to do) and category (the broader topic), the actual user message text (what the customer said), the response generated by the system (what the AI or template said back), a latency value (how long the response took, color-coded from green to red based on the thresholds we discussed), and the timestamp (exactly when this interaction happened).
Reading through the logs is the best way to understand how your Voice AI is actually performing in real conversations. You might discover that the AI is giving excellent answers to complex questions (confirming that the cost is worth it), or you might find that it is spending expensive AI processing on simple questions that a template could handle. Both discoveries are valuable.
Common Mistakes to Avoid
- Not reviewing the tier distribution after the first week of using Voice AI – many businesses leave the default configuration unchanged and end up routing 50% or more of messages through costly AI processing when simple templates would handle them just as well. Set a calendar reminder to review this dashboard one week after enabling Voice AI, and then weekly after that.
- Ignoring high AI latency warnings until customers start complaining about slow responses on calls – by that point, you may have already lost customers who hung up in frustration. Monitor the latency cards daily (or at least every few days) and investigate any sustained red indicators immediately. A 3-second response time might not sound bad, but on a live phone call, it feels like an eternity.
- Focusing only on cost savings while completely ignoring response quality – saving money is great, but not if customers are getting wrong answers. The cheapest tier is not always the best. Make a habit of reviewing the Logs tab (Step 8) weekly to verify that Instant responses and Templates are actually answering customer questions correctly and helpfully.
- Forgetting to update Instant responses when your business details change – this is a subtle but costly mistake. If your business hours change for the summer season, your phone number changes, or you raise your prices, the Instant responses will still give the OLD information. They will give WRONG information at lightning speed. Every time you change something about your business, check your Instant responses and update them to match.
- Never looking at the Daily Trend chart – this chart reveals patterns that are invisible in the aggregate numbers. For example, you might discover that AI usage spikes every Monday morning (customers calling about weekend emergencies) or that Template usage drops on Fridays (fewer appointment scheduling calls). These patterns help you optimize staffing and Voice AI configuration for different days of the week.
- Creating templates that are too rigid and miss customer intent variations – if your template for “pricing inquiry” only triggers when someone says “how much does it cost” but misses “what are your prices” or “is it expensive,” those alternate phrasings will fall through to the expensive AI tier. Review the AI-tier logs to find different ways customers phrase the same question, and make sure your templates cover all common variations.
What’s Next?
Now that you’ve completed this guide, check out:
Need help? Post in the Tech Support category or contact support@exoserva.com.