Toast IQ can answer. Can it do the job?

I was planning to write about how much better ChatGPT has gotten. In my experience, GPT-6 Astra is head and shoulders above Claude, which is a fairly significant thing for me to say considering how much of what I’ve shared here was built with Claude.

Then Ruben Hassid’s “How to switch AI” landed in my inbox this morning, and it’s extremely well done. Go read it.

Ruben’s more diplomatic about which one is better, but he does the more useful thing: he walks you through moving your information between them. Your history, your preferences, the skills you’ve built. If you’ve spent months teaching Claude about your business, having to do all of that again is a pretty good reason to put off trying something else. His piece helps you get past that.

I’m not going to rewrite his walkthrough. What I want to do is bring this back to the restaurant.

I’ve been building tools that check invoices, catch bad punches, report labor costs and draft online review responses. Meanwhile, Toast has its own AI product, Toast IQ. It’s reasonable to look at everything I’ve been making and ask how much of it Toast already does. And believe me, many people have asked that.

That’s what I want to explore: how ChatGPT and Claude, connected to the tools I’ve been building, compare with what Toast IQ currently does. Where do they overlap? What does each do well? How much setup does it take before any of this is useful to you?

Because if Toast already solves a problem well, you should know that before spending your day setting up another way to solve it. And if there’s useful work you can do beyond what Toast offers, I want to show you what that looks like.

The useful part of switching AI is being able to take your restaurant information and the work you’ve already done with you.

TL;DR: Toast IQ was useful for quick questions with no setup. In my test, it couldn’t answer check-level combination questions, connect to the outside information I needed, or schedule a daily report. That’s why I’m still building with ChatGPT and Claude.

I asked Toast IQ 32 questions using live data from my restaurants. Some were questions it should be able to answer from Toast alone. Others were the kinds of questions I actually want answered while running the businesses, which often require information from somewhere else.

There’s a lot to like about having this inside your POS. I didn’t have to connect anything. I opened it and started asking questions.

When I was vague, it asked which location and which metric I meant. When I asked it to compare my check average with similar area bars, it said it didn’t have that information. I deliberately gave it an opportunity to make up a plausible number, and it didn’t, which is great.

The setup matters here too. I’ve written entire walkthroughs about getting API credentials and connecting tools. It's definitely a process but it's not too much for the average person, in my opinion. In any case, Toast IQ skips that part, and for somebody who just wants to ask a question about their sales, that’s a substantial advantage.

Toast IQ’s biggest advantage here was convenience: I opened it, asked questions and got answers without a setup project. But you already know that.

I asked what percentage of checks with our most popular individual menu item also had a cocktail.

That’s a useful restaurant question. If people buying a particular entree rarely order a drink, maybe there’s something to try with the menu or with how we sell it. But first I need to know whether that’s actually happening.

Toast IQ could show me item sales and cocktail sales separately. It couldn’t tell me how often they appeared on the same check.

I ran into the same limitation with modifiers. It could rank individual modifiers by popularity, but it couldn’t tell me which combinations people ordered together.

In these tests, it could give me the totals but couldn’t get underneath them to answer questions about individual checks. Connecting an AI to the underlying records gives you a way to investigate those questions (although you should still check its work).

And that last part applies to all of these tools.

It could tell me how much of each item sold. It couldn’t tell me what people bought together.

Then I asked Toast IQ to schedule a daily labor report.

It said it couldn’t. I could come back and ask for it each morning.

I was honestly surprised it couldn’t create a scheduled task. This is where the comparison gets directly into why I’ve been building these tools, and something that clarified a lot for me when I realized it.

Using AI tools is about removing friction and increasing speed.

You can already look up your labor. But if you have to go in every morning and request a report, you’re going to forget. You’re going to get busy doing something else, and you won’t have it when you need it. The fact that you could have asked for it doesn’t help much at that point.

Have that report arrive automatically in your email, in a format you’re actually going to read, and it becomes part of your morning. You’re ready for it. You know where to look and what needs your attention. You can get to the decision faster because you haven’t spent the first part of the morning remembering to go get the information.

That is a world of difference from “yes, you can ask it that question.” A report that arrives first thing in the morning takes one more recurring job off your list.

I also asked about vendor price increases and pour cost. It didn’t have the invoice data, which is to be expected. When I asked whether it could connect to xtraCHEF from the chat, it told me to upload an export.

An upload can work. But if I’m exporting a file every time I want an answer, that’s still a job I have to do. Friction.

The same issue came up with labor budgets and sales targets. Those questions require context beyond the actual sales and hours worked. The numbers need something to be compared against. This, again, is why the MCP connectors that I made available for Toast make all the difference. I also have an MCP for xtraCHEF that I have not made public yet because I'm continuing to develop it. (Are you interested in this? Email me and let me know.)

In this test, I still had to ask for each report and upload outside data myself. Those recurring steps are exactly what I want taken off my list.

Disconnected data and tools make it harder to understand your customers. HubSpot's Agentic Customer Platform brings your data, teams, and tech stack together with AI built in to help your business work faster and create more personalized customer experiences.

Why HubSpot and what's new

Connect more of your business in one place and give every team a smarter way to work. Get set up quickly and start checking off your hardest tasks.

Get started free

Based on this test, I’d use Toast IQ for convenient questions about the information it can already reach. Having that available without a setup project is useful.

For the work I want happening every morning, or questions that need invoices, budgets, manager notes and individual checks brought together, I’m still building.

That also brings me back to Ruben’s piece. I prefer ChatGPT right now, but I’ve built useful things with Claude and I want to be able to use both. The restaurant information and instructions I’ve assembled need to come with me.

I hope this helps answer what has become a popular response to the AI tools for restaurants that I've built so far.

It's pretty incredible the speed at which these things are developing. I know for a fact, directly from Toast, that they are building tools that are very similar to the things that I'm making. I have no idea when they'll come out, what they're called, how much they'll cost, or the exact scope that they'll be able to address but I do know they're coming. Which makes it all the more funny when people say, "Just use Toast IQ," as if there are absolutely no more worlds to be conquered. There certainly are.

What have you built lately?