An AI assistant cut a second location's license paperwork from days to hours. What the case study shows

What does the case study say?

On October 1, 2026, OpenAI published a customer story about The Den Family Social. It is a social club in Denver for parents, with space for their children too. It runs food and drink service, so it needs a liquor license. The club was building a second location, and it used ChatGPT Work to handle the extra paperwork.

The page makes three headline claims:

• Liquor-license materials took 3 hours, not 4 days. OpenAI puts that at 91% less time.
• Grant applications took 2 hours, not 3 days. That is 92% less time, by the page’s count.
• The leadership team saves 10 to 15 hours a week. The founder estimates she alone saves about 7 hours a week of exploratory talks.

Read it as what it is: a vendor’s case study. OpenAI wrote it to sell its product. The page does not say how the hours were tracked, and we found no outside check of them. We take the numbers as the customer’s own estimate.

What did the assistant actually do?

The detail is more useful than the headline. The club connected its Gmail, Slack and Google Drive to the assistant. It then asked it to pull together what the team knew from the first location and to propose next steps.

The liquor-license job is the clearest example. An earlier hearing slot opened up at short notice. The team used the assistant to gather the documents, find what was missing and convert files into the required formats. The page quotes the club saying four days of “searching, sorting, formatting, and assembling” became three hours of team review.

Three other uses appear in the story:

• Partnership planning. A manager described which partnerships had worked and why, and the assistant drafted a structure for leaders to review.
• Inventory and purchasing. The general manager combined the accountant’s reports with sales and inventory data from the point-of-sale systems. The assistant organized it into a view used to forecast inventory across both locations.
• Email on the move. The founder talks through an email while traveling, lets the assistant draft it and sends it after review.

The page says the team still reviews the output for accuracy and makes the final decisions. The club also plans to try Codex to build a member app and link its point-of-sale systems.

Why is paperwork the slow part of a second store?

Paperwork scales badly. A restaurant’s first opening teaches the steps. The second one needs the same steps, but the knowledge sits in old emails and one person’s head.

The City and County of Denver’s page on moving a liquor license shows how many parts there are. It covers moving an existing license to a new address. We did not check the rules for a brand-new license, so read it as an example of the parts:

1. State and city applications. Both are required, and Denver’s licensing department handles the process.
2. A zoning use permit. The city says to check first whether one exists for the address.
3. A deficiency window. An incomplete file gets a notice by email, with 14 days to fix it.
4. A time limit. Applications that are not completed within 12 months are closed.
5. A public hearing. Nearby residents and businesses can sign petitions and testify. A sign must be posted at the premises for at least 20 days before the hearing.
6. Inspections. These start only after the hearing approves the license.

An assistant can’t skip any of these steps. What it can do is shorten the gaps between them. Finding the lease, the floor plan and the old approvals is the slow part. So is spotting a missing document before the city sends a notice.

What can a restaurant group take from this?

The lesson is about where to point the tool. These five steps follow from what the club did:

1. Pick one repeated task. A license checklist or a permit file is a good start. It is the same for every store, and the steps are written down.
2. Connect only where the documents live. The club linked email, chat and a shared drive. Limit access to the people who handle permits.
3. Ask it to find gaps, not to decide. “What is missing from this file?” is a good question. “Should we open here?” is not.
4. Keep a person on every filing. The club reviewed the work. A wrong address or an old floor plan can cost a hearing date.
5. Track your own hours. Time one filing before and one after. Then you have your own number, not a vendor’s.

Front-of-house work is a separate job. A phone agent such as Slang AI or Smith.ai answers guest calls and takes bookings. Our AI voice agent rankings compare them on published plans. A connector tool such as Zapier moves data between apps on a fixed rule, with no chat in the loop. A guest platform such as SevenRooms keeps one guest profile across stores. None of these replace the paperwork help in the story, and the assistant in the story does not replace them.

What would we not read into it?

Three cautions apply.

The numbers are self-reported. A 91% cut is a big claim from one customer. It covers one application, after an earlier hearing slot opened. Your own result will depend on how scattered your files are to begin with.

A first store is different. The club had a first location’s files to draw on. A brand-new owner has nothing for the assistant to gather.

The city does not move faster. The hearing notice, the 20-day sign and the inspections run on the city’s clock. The case study saved time on the team’s side of the counter, not the city’s.

That still leaves a real gain. A second store costs a group many hours of searching for things it has already done once. A tool that finds those files and lists what is missing is worth a trial on one task. Measure it, keep a person on the final filing, and decide from your own numbers.