Sultan’s AI Agents Pay Off for FavTrip

After one of his artificial intelligence (AI) agents catches a major error, Sultan shares his playbook and future goals.

One of Babir Sultan’s artificial intelligence (AI) agents recently caught a $25,000 error.

Earlier this year, Sultan, founder and CEO of FavTrip, a three-store convenience chain in Missouri, identified four pain points he needed to solve for his business. He assigned an AI agent to each issue. One tracks snacking trends and makes suggestions for which snacks to stock. The second drafts posts for FavTrip’s three social media channels. The third is a billing agent that reconciles weekly gas invoices, and the fourth AI agent messages job applicants for open interviews.

Sultan said it all comes down to efficiency. The AI agents free employees from time-consuming tasks, allowing them to focus their energy on other responsibilities in the store.

“A lot of small operators that I talk to, they think that these things are meant for big chains, and I always say, ‘if I could do it, you guys could do it,’” he said.

The implementation of the AI agents recently paid off in a big way when the billing agent, which the chain calls “Mary,” made the $25,000 find.

“When it comes to reconciling 4,550-plus gasoline invoices, it’s a tedious process,” Sultan said. He struggled to find an employee that wanted to handle the task.

Enter AI agent “Mary.”

Mary not only flagged the $25,000 error in real time, it also reached to the vendor to resolve the problem.

“(It) emailed the vendor that we have on file and said, ‘Can you please clarify? My name is Mary. I am an AI bot, but I need clarification.’ And lo and behold, they said, ‘Sorry, that was meant for another store.’ But imagine not catching that and then having to go back and forth. For a small operator, $25,000 is a lot of money. It’s not a rounding number, that’s for sure,” Sultan said.

Mary wasn’t flagging problems and messaging vendors on day one, however. Sultan started small. When he originally set up the AI agent, it simply reported on which gas loads had been billed, and which had been drafted and paid for.

**“**It was like connecting to our Gmail Google sheet,” he said.

After a while, satisfied that Mary was working well, Sultan then took the technology further. He set it up so that anytime Mary noticed that the company had been billed incorrectly, it would flag the item and say, “I can’t find this” and stop there.

“Then we would say, ‘Okay, why can’t you find it? Let’s go through a bunch of papers and find that invoice,’” Sultan said.

After a while, Sultan took the technology even further. He gave the agent even more responsibility and armed it with the list of the correct vendors, including their email, contact name and so on. Before taking this step, Sultan spoke with his vendors, so they were aware an AI bot named “Mary” would be emailing them if there was ever an issue. Now, Mary can email the vendor and cc Sultan if an issue arises.

And that’s how Mary was ready to act when the $25,000 error came through. It noticed the problem in real time and immediately sent an email.

“(Mary) said, ‘Hey XYZ, I noticed there’s a load that came in. Doesn’t look like it’s ours. Can you please confirm where it went?’ And they’re like, ‘Oh, sorry, the driver forgot to put in which store it went to. They accidentally billed you,’” Sultan explained.

“You’re dealing with different people,” he added. “You’re dealing with a fuel supplier, a driver, cashiers, managers, billing people. Mary keeps everybody organized.”

The takeaway for Sultan was that this same process would have taken the team months to spot and correct. “You can imagine — and all store operators know this. If that ACH (Automated Clearing House) came out seven days later and then you caught it, a month is out. It’s going to be probably another couple of months out before you tell the bank, ‘Sorry it was the wrong one,’” he said.

Then there are forms to fill out and the vendor to contact.

“(Mary) did it the same day that we got the email. The very same day — within minutes,” Sultan said.

Sultan acknowledged that Mary can also make mistakes, which is why human oversight is still crucial. If Mary flags something incorrectly, Sultan updates the code.

“When I say code, it’s like a chat with your agent back and forward. I say, ‘Can you update skills?’” he explained.

Now Sultan is working with his AI agent to determine pricing strategy.

“You see how it builds up from now. The AI knows the delivery process, but it also knows the cost now too. But now, I want to take it one step — just one — to ‘what should I retail (it for)?’” he said. “I want to know within my zip code what is something going for because that’s more unique to me.”

“The sky honestly is the limit,” he added. “One big thing I’ll tell a lot of people, if they’re thinking about AI: Think of it as your microwave. It’s not going to make your wonderful recipe. It will warm your food for you. So come in problem first.”

For more examples and strategy on how to get started with AI, even as a small operator, listen to the CStoreTalks podcast with Sultan HERE.