The Restaurant Owner's AI Dashboard: What You Should Be Tracking
If you're paying for AI features but can't see what they're doing, you're writing blank checks. Here's what a restaurant AI analytics dashboard should track — and why most platforms hide these numbers.
If you're paying for AI features but can't see what they're doing, you're writing blank checks. How many orders did the AI touch today? What did those interactions cost? Did the upsell suggestions convert or just annoy people into closing the tab?
Most restaurant owners running AI tools right now can't answer any of that. The vendor sold them on the magic — "AI will boost your revenue!" — but gave them zero visibility into what's happening behind the curtain.
I built an AI observability dashboard into Deally because this is a fundamental problem. You wouldn't run your kitchen without knowing food costs. So why would you run AI agents — agents interacting with your customers and costing you money per request — without knowing what they're doing?
Why Visibility Into AI Matters More Than the AI Itself
There's a pattern I've seen with every restaurant owner who starts using AI features. Week one: excitement. Week two: curiosity. Week three: vague unease. By month two, it's either "I think it's working?" or "I turned it off because I wasn't sure it was worth it."
Both reactions come from the same place — no data. That's a failure of the platform, not the operator.
AI agents run in the background. They respond to customers, generate upsell suggestions, and process natural language commands. Unless there's a dashboard showing you what happened, you'll never know whether that AI is earning its keep or burning through your budget on failed interactions.
If you've read our piece on using restaurant analytics to make better decisions, this is the same principle applied to AI specifically. What operators need is a screen that answers four questions:
- Is it working? Success rates, error rates, completed vs. failed interactions.
- Is it making money? Conversion tracking on suggestions, revenue attributed to AI.
- Is it costing too much? Cost per request, total spend, trends over time.
- Is it fast enough? Response times — because a customer waiting 8 seconds for a chatbot reply is a customer who's about to leave.
If your AI vendor can't answer those with actual numbers, they're asking you to run a piece of your business on faith.
The Metrics That Actually Matter (And the Ones That Don't)
The tech industry loves vanity metrics — total messages processed, number of sessions, "engagement score." Those numbers are almost useless for an operator trying to figure out whether they should keep paying for this thing.
Here's what matters.
Cost Per Request
This is the single most important number most platforms don't show you. Every time an AI agent processes a request — a customer question, a menu command, an upsell evaluation — it costs something. Language models aren't free.
In Deally's Agent Hub, every agent shows its average cost per request. Right there on the dashboard, updated in real time. If your ordering agent costs $0.03 per interaction and handles 200 conversations a day, that's $6. Totally reasonable. If it starts running excessively long conversations, you'll see that cost climb before it becomes a problem.
I've seen operators unknowingly spending $400 a month on AI features that generated maybe $150 in incremental revenue. Nobody showed them the cost side of the equation.
Success Rate and Error Rate
An AI agent that fails 15% of the time isn't a tool — it's a liability. Every failed interaction is a customer who didn't get their question answered or an order that didn't get placed.
Success rate should be per-agent, not a platform average. Your ordering assistant might run at 97% while your menu management agent sits at 99.5% — and that's fine, because ordering is inherently messier. But if the ordering agent drops to 88%, you want to know immediately, not at the end of the month in a summary email.
Errors have different severities, too. A "couldn't parse that request" error is annoying. A "hallucinated a menu item that doesn't exist" error is dangerous. Good observability distinguishes between the two.
Upsell Conversion Rate
If you're using AI-powered upselling, the dashboard needs to show what percentage of suggestions customers actually accepted. Not "impressions" — conversions. Did the customer add the item and complete the order with it?
A 12% conversion rate might sound low, but if the average upsold item is $6 and you're running 300 suggestions a day, that's $216 in daily incremental revenue. Compare that to the cost-per-request number from above. That's the math your dashboard should make easy.
Average Response Time
Past a certain accuracy threshold, speed matters more. An AI that gives a perfect response in 6 seconds will lose to a good-enough response in 1.5 seconds. If your conversational agent is dragging, that response time metric is your early warning.
We track response time per agent with internal targets: ordering agent under 2 seconds, menu agent under 3 seconds, upsell evaluation under 500 milliseconds. When those spike, something needs attention.
The Deally Assistant: Your AI Co-Pilot for the Dashboard
The Deally Assistant is a persistent AI bot that lives in your dashboard. Not a separate app, not a pop-up widget — it's always there. It has 14 tools and can check today's orders, look up a customer's history, analyze menu performance, or pull up the AI metrics we've been discussing.
What makes it different from a static dashboard: you can ask questions in plain English. "How did the upsell agent perform last week?" "What's my ordering agent's error rate trending?" "Which suggestions had the highest acceptance rate this month?"
The Assistant delegates to specialized agents when needed. Want to understand why the ordering agent's success rate dipped on Thursday? It can pull the trace logs and surface the specific conversations that failed — model issue, menu data problem, or just a batch of confusing customer requests.
Most restaurant owners aren't going to study a metrics dashboard for 30 minutes. They'll glance at it during a lull or pull it up on their phone between deliveries. An AI that summarizes "here's what you need to know today" makes the difference between data that gets used and data that gets ignored.
Keeping Costs Under Control (Without Killing the Feature)
The number one fear I hear from operators about AI is runaway costs. It's legitimate — language models charge by usage, and a busy restaurant generates thousands of interactions a day. Without guardrails, the bill gets ugly fast.
Here's how we handle it in Deally, and what you should demand from any platform.
Rate limiting. Every agent has configurable limits. If your ordering agent suddenly processes 10x normal volume — a bot hitting your page, a loop somewhere — the limiter kicks in before you've burned through your budget. Any AI platform without rate limiting is a financial liability.
Session management. Sessions auto-expire after 30 minutes of inactivity with a hard cap of 100K tokens. That prevents a single rambling conversation from eating disproportionate resources — and keeps the AI's context focused, since hour-long conversations produce worse responses anyway.
Hot-model switching. Different AI models have different price-to-performance ratios. A complex ordering conversation might need a more capable model, while a simple upsell evaluation can run on something cheaper. Deally lets you change the model per agent without redeploying anything. If costs trend up, switch your upsell agent to a lighter model and see if conversion holds.
Trace logging. When something goes wrong, you need to see exactly what happened. Every interaction is logged with a full trace: the input, reasoning, output, cost, and time. The AI equivalent of a kitchen camera — you hope you never need it, but when you do, it saves hours of guessing.
The goal isn't to minimize AI spending — it's to maximize the return on every dollar. You can't optimize what you can't measure.
When AI Isn't the Answer
I'm going to say something that sounds counterproductive from someone who builds AI tools: turn off any AI feature that isn't demonstrably earning its keep.
The dashboard exists to make this judgment call possible. If your upsell agent's conversion rate has been below 3% for three weeks, the suggestions aren't resonating. Don't leave it running and hope. Rethink the strategy or turn it off.
If your ordering agent's error rate is climbing, something is wrong with the menu data feeding it. Fix the data first — the AI is only as good as the information it works with. For a broader look at how AI fits into operations, this piece on AI in restaurant operations covers the full landscape.
Some operations don't need AI yet. A 12-seat counter-service spot with 15 items? It won't move the needle. A 200-seat operation running delivery, pickup, dine-in, and catering across 80 items? That's where AI agents generate serious returns — and where the dashboard becomes essential.
AI is a tool, not a strategy. The strategy is serving great food and growing revenue. AI accelerates that when it's applied to the right problems and monitored with real data. Without monitoring, you're just guessing.
Frequently Asked Questions
Do I need technical knowledge to use a restaurant AI analytics dashboard?
No. In Deally's Agent Hub, metrics are displayed in plain language: cost in dollars, success as a percentage, response time in seconds. If you can read a P&L, you can read the AI dashboard. The Deally Assistant also lets you ask questions in natural language instead of navigating charts.
How much should AI features cost a restaurant per month?
It depends on volume, but most Deally restaurants spend $30-$300 per month on AI compute. The key metric is the ratio of AI spend to AI-generated revenue. If you're spending $150 and the upsell agent alone is generating $800 in incremental revenue, that's a 5x return. The dashboard makes this math visible.
What's a good success rate for an AI ordering agent?
For a well-configured ordering agent with accurate menu data, you should expect 95% or higher. Below 90% usually indicates a menu data issue — missing modifiers, unclear item names, or option groups that confuse the model. The dashboard will show you the specific conversations that failed, so you can diagnose whether it's a systemic problem or just edge cases.
Can I see what the AI said to my customers?
Yes, and you should. Trace logging records every conversation and response. You can review specific interactions, see where things went off track, and understand why a session failed or succeeded. You shouldn't have to blindly trust an AI that's talking to your customers.
A great AI agent paired with zero visibility is just an expensive mystery. A decent one paired with a clear, honest dashboard becomes a tool you can manage, optimize, and trust.
If you're evaluating platforms, ask to see the AI dashboard before you ask about any other feature. If they don't have one — keep looking.
See how Deally's Agent Hub gives you full visibility into every AI interaction.