Method & limits

What the data is

Every Texas holder of a mixed-beverage permit files a monthly return with the Comptroller showing gross receipts from liquor, wine, beer and cover charges at each location. The state publishes these returns as open data. This tool reads every Harris County location (plus any Houston-addressed location outside it) from August 2024 onward: filings from locations. The Data tab walks through one filing field by field.

These figures are alcohol sales, not total revenue. Food is not included. A restaurant doing $1M in alcohol might do $3M+ in total sales. The value of the data is that it is monthly, per location, public and current, so it shows momentum that brokers usually only hear about secondhand.

How a candidate is scored

• Momentum (35%): last 12 months versus the 12 before, blended with the last 3 months versus the same 3 a year earlier. Growth of +60% or more scores full marks; −20% or worse scores zero.
• Expansion readiness (20%): one location statewide scores highest. A strong single unit is the classic second-location candidate.
• Scale fit (15%): how close annual alcohol sales sit to the middle of the range you set.
• Durability (15%): years under permit, minus months with no filing in the last year.
• Location (15%): 1 to 5 miles away scores best. They are close enough that their customers know the area, and far enough that a second unit won't take sales from the first. Within ¾ mile is marked down as the same trade area.
• Concept fit: when you pick a concept, operators whose liquor/wine/beer mix doesn't match are marked down. Operators of the same type as the property's current tenants are marked down slightly, so the list favors variety.

Business kind and drink profile

Each card carries two tags that answer different questions. The first, business kind (restaurant, bar, hotel, nightclub, shop or salon...), is judged by an AI model: TypeSafe's Jev reads the trade name, operator, permit and drink mix of each location and picks one kind, with a confidence. It is shown only when that confidence is 0.7 or higher; otherwise the tag says "Unsure", and hovering it shows the model's two leading answers. Restaurant and bar are shown together as "Restaurant or bar": on the top-ranked candidates the model called several chef-driven restaurants (The Palm, Ume, The Grove) bars with confidence up to 0.90, because a short name can't settle it. The model's own pick stays in the tooltip and the CSV. The Data tab scores what is shown against locations a person labeled from the name alone before the model answered. With "Hide hotels, clubs, venues, concessionaires" on, a location the model calls a hotel, venue, private club or concessionaire at 0.85 or higher is hidden too. That catches airport gates, cinemas, veterans' posts and hotels whose names the word list misses, and also a few bars that operate on private-club permits; turn the filter off to see them.

The second tag, the drink profile, is computed, not judged. Drink profile is a guess from each location's mix of liquor, wine and beer sales over the last 12 months. Cover charges over 5% of sales mean live music or a club. Beer at 50% or more means a beer-forward spot. Wine at 40% or more means wine-forward dining. Liquor at 60% or more means cocktail-forward. Anything else is full-service. It is a guess, not a label, and a name check settles it.

Restaurant groups

Restaurant groups often file each unit under its own company, so one taxpayer ID can look like a single-location operator. Every permit holder also gives the state a mailing address, and a group's companies usually share one: the group's office or the owner's home. So do unrelated clients of one accountant, attorney or registered agent. The tool links permit holders that share a mailing address statewide and counts them as one group when the names say so (most share a distinctive word, like the Little Woodrow's and Fast Eddie's companies, or several are numbered sister companies), or when Jev, reading every member's company and trade names, judges them one company with confidence 0.7 or higher. The card then says "same mailing office as ..." with the group's location count, and "Count sister companies as one operator" (on by default) uses that count for the locations-statewide limit and expansion readiness. Turn it off to see a group's single units as candidates. A multi-concept owner whose restaurants share no name, like 917 Franklin and La Colombe d'Or, is often missed: Jev tends to read different brands at one office as an accountant's clients, which is why its "shared office" answer never overrides the names. The Data tab scores both methods against clusters a person labeled before the model answered.

What it can't see

• Anyone without a mixed-beverage permit: coffee, retail, fitness, BYOB spots, and beer-and-wine-only permits (a different permit class).
• Ownership itself. Groups are inferred from shared brand names and shared mailing addresses, not from ownership records. A group whose companies use different mailing addresses, or an old company still using a former accountant's address, looks like several independents; a shared office can occasionally join two owners who are not related. Check ownership before the first call.
• Who owns what today. Each firm's property list comes from its own website, state filings and press, and those include partnerships, managed assets and recent sales. The Portfolio tab shows how far each address could be checked. Confirm the current owner and leasing contact before outreach.
• Addresses the U.S. Census geocoder could not place: . Suburban addresses place less often than inner-Loop ones. Another landed implausibly far from their own ZIP code and were removed. Locations without a position are left out of radius searches.
• Late filers. The most recent month () is still filling in and is shown in grey, not scored.
• Lease terms, credit, or whether an operator wants to expand. This tool decides who gets the first call. A broker's judgment decides the rest.

Next steps in a full engagement

• Add TABC new-permit applications and City of Houston permits, which show who is opening, about 3 to 9 months ahead.
• Add review-count growth and Instagram activity for operators who aren't licensed to sell alcohol.
• Confirm restaurant groups from officers and registered agents on Texas Secretary of State filings, which catches groups the mailing address misses.
• Extend to Fort Bend and Montgomery counties, and fall back to the ZIP-code center for addresses the geocoder misses.
• Send a monthly alert whenever an operator within 5 miles of a vacancy crosses a growth threshold.