Tony Aiazzi didn't set out to build a restaurant technology company.He just wanted to stop doing invoices.
At the time, Tony was a chef, ten years into his career with Charlie Palmer and planning to open a restaurant with his co-founder, Xavier. They knew there were plenty of difficult parts of running a restaurant. Manually processing invoices was one they had no interest in accepting.
Their invoices were being faxed, FedExed overnight to a back office in Las Vegas, and re-entered by someone who had never seen or handled the products being purchased.
The restaurant Tony and Xavier were planning never opened. The invoice system they built to support it did.
That system became Chouxbox and eventually Over Easy Office, a Manila-based operation processing tens of thousands of restaurant documents every week.
"I hated doing invoices so much as part of my restaurant life and saw it as such a ridiculous use of my time that I developed the system. Now it's my full-time job. I just do it for everybody else."
But Tony's story isn't only about restaurant invoice processing.
It's about what happens when technology automates the easy 90% of a job, and everyone discovers that the remaining 10% is where the real work begins.
What Restaurant Invoice Processing Looks Like in Real Life
On paper, processing an invoice sounds straightforward.
- A restaurant orders a product
- The vendor delivers it
- The invoice is scanned and categorized
- The invoice is sent to the accounting system
That is how the process works when everything goes right. Restaurants rarely have that luxury.
Ice cream melts on the truck and gets returned. A vendor substitutes a product without asking. A delivery arrives short. Someone writes a credit in the margin. Meat and seafood come in at different catch weights. An invoice number that already exists gets recycled.
Then there are handwritten tickets, damaged paperwork, changing vendor item numbers, and notes that only make sense to the person standing at the back door when the delivery arrives.
AI invoice automation is increasingly good at handling clean, predictable documents.
meez's own guide to restaurant invoice management covers the same ground: automating data capture and categorization only helps if the output still reconciles with what actually landed on the loading dock.
Clean invoices are not the reason restaurant operators struggle with invoice processing. The problem is everything that happens when the invoice doesn't match what actually arrived.
Tony estimates that most of the value his team provides sits inside the 5–10% of invoices with notations, credits, substitutions, and other exceptions.
That may be a small percentage of the total invoice volume. It's also the portion most likely to affect whether the restaurant's numbers are right.
AI Can Read an Invoice. Can It Understand What Happened?
Tony compares fully automated invoice processing to self-driving cars. Restaurants don't operate on roads designed for automation.
"They're only going to be fully autonomous when there are roads built specifically for robotic cars to drive on exclusively. As soon as humans are in the mix, it all kind of only gets to a certain point."
Food isn’t standardized inventory. A ten-pound case of produce is fairly predictable. A ten-pound piece of meat might arrive at 9.4 pounds. A vendor may change an item number without warning or substitute a different brand.
The invoice might say one thing while a handwritten note says another. Someone still has to understand what physically happened.
That doesn't make AI invoice automation ineffective. It means restaurants need to be more thoughtful about what they expect it to do.
Modern Restaurant Management's 2026 restaurant tech trends coverage points to the same shift: AI already powers invoice-processing engines alongside drive-thru order models and predictive sales forecasting tools. It automates repetitive work and hands operators back real hours in their week, but the piece is careful to frame that as task automation, not judgment automation.
AI is extremely useful when it removes repetitive data entry, recognizes familiar line items, and routes clean information into the correct system. It becomes less reliable when the work requires context, judgment, or knowledge of restaurant operations.
The goal shouldn't be to remove humans from the process entirely. The goal should be to stop wasting human attention on the parts a machine can handle safely.
The Last 10% Is the Part That Protects Your Margin
It's tempting to judge invoice automation by the percentage of documents it can process without intervention. But a 95% automation rate doesn't necessarily mean 95% of the problem has been solved.
"I would say most of our value is in those like five to 10 percent of invoices that do have those notations. There's a lot of variability there."
The last 5% may include the situations that have the biggest effect on restaurant margins:
- A missing vendor credit
- An incorrect catch weight
- A product substitution billed at a higher price
- A duplicate invoice
- A shorted delivery
- An ingredient mapped to the wrong category
- A vendor price change that no one noticed
Those are not simply data-entry errors. They're situations that can change what a restaurant paid, what it received, and what its food actually costs.
That's why the human reviewer still matters.
Their job is no longer to enter every line manually. Their job is to investigate the exceptions, understand what happened, and make sure the clean-looking data reflects reality.
Good automation doesn't eliminate that work. It makes the important work easier to find.
Restaurant Invoice Processing Is a Food Cost Problem
Invoice data is often treated as an accounting concern.
But every food cost number starts with what the restaurant paid for its ingredients.
When invoice data is delayed, duplicated, or mapped incorrectly, the problem doesn't stay inside accounts payable. It moves downstream into recipe costing, theoretical food cost, purchasing reports, and cost variance analysis.
A chef may think a dish costs $8.20 to produce when the current ingredient prices put it closer to $9.10. An operator may believe a location is running a 29% food cost when its actual cost is several points higher.
By the time the monthly P&L exposes the difference, the restaurant may have spent weeks selling an item at the wrong margin.
That's why restaurant invoice processing belongs inside the broader recipe costing workflow, not off to the side of it.
When invoice line items map accurately to ingredients, recipe costs can update as prices change. Operators can see margin drift earlier. Culinary teams can investigate unusual costs before they become month-end surprises.
Clean invoice data isn't a back-office convenience. It's the foundation every other food cost number is built on.
Better Data Creates Better Restaurant AI Insights
Restaurants are being promised more AI insights than ever.
Tools can identify purchasing trends, flag cost increases, recommend menu price changes, and surface potential margin opportunities.
Those recommendations are only useful when the data underneath them is accurate. meez has written before about the gap between AI-washed tools and AI-real ones: the restaurant groups getting genuine ROI aren't running the flashiest software, they're the ones who fixed the underlying data first.
Restaurant Dive's reporting from this year's National Restaurant Association Show found something similar, tech leaders at the show repeatedly named AI as the thing the industry is getting most wrong right now, even while enthusiasm for the technology stayed high on the floor.
If an invoice line is mapped to the wrong ingredient, the recipe cost will be wrong.
If a vendor credit is missed, the restaurant's actual spend will be overstated. If a catch weight is entered incorrectly, every yield-adjusted cost connected to that ingredient may drift.
AI doesn't know that the loading dock and the invoice disagree unless someone, or some well-designed process, captures the exception.
This is where invoice data and recipe data have to work together.
Invoice automation captures what the restaurant paid. A structured recipe system shows where those ingredients are used, how yields affect their usable cost, and what changing prices mean for each menu item.
Together, those systems can produce restaurant AI insights an operator can actually act on.
Without that foundation, AI may produce an answer that looks sophisticated but is based on numbers that were never right in the first place.
The Feedback Loop Matters More Than the Dashboard
Tony remembers a common restaurant accounting rhythm: sitting down in March to discuss why January wasn't profitable.
By that point, there wasn't much anyone could do about it.
The ingredients had already been purchased. The menu items had already been sold. The labor had already been scheduled. The restaurant was looking backward at a problem that may have continued for another two months.
That lag is still common. National Restaurant Association data from earlier this year found only about a quarter of restaurant operators currently use AI-related tools at all, with administrative tasks ranking second behind marketing as the top use case, which leaves most back offices still running the same lagging, after-the-fact cost conversations Tony was trying to eliminate.
Faster restaurant invoice processing shortens that feedback loop.
Instead of discovering at month-end that the cost of an ingredient jumped, an operator can catch the change during the same week. Instead of waiting for the P&L to reveal a margin problem, the culinary team can see the effect inside the recipe, the same rhythm behind good recipe costing practices generally: catch the drift early, not at month-end.
This is where better invoice automation can improve food cost control.
The value isn't simply saving someone from typing numbers into a spreadsheet. It's helping the restaurant recognize what changed while there is still time to respond.
That could mean calling a vendor, correcting a mapping, changing a purchase, adjusting a recipe, or reconsidering a menu price.
The faster the feedback, the smaller the problem has to become.
Bootstrapping Taught Tony What Venture Capital Might Have Hidden
Tony and Xavier went through 500 Startups early in the company's life. After that, they chose not to continue raising money.
Instead, they built a business that had to support itself.
It wasn't the glamorous version of a restaurant technology startup. They couldn't spend years chasing growth while ignoring whether customers would pay for the service. The business needed to solve a real problem and generate enough revenue to keep solving it.
More than a decade later, Over Easy Office is still standing.
During that time, better-funded competitors in the restaurant invoice automation space have folded, been acquired, or disappeared.
Tony describes what he built as closer to a lifestyle business than a venture-scale software company.
That distinction isn't only relevant to founders. It matters to restaurant operators choosing technology partners. A vendor can have an impressive funding announcement and still build a tool that doesn't fit the daily reality of a restaurant.
The more important questions are often less exciting:
- Does the product solve the messy parts of the job?
- Will the company still support it after implementation?
- Does the business work when the demo ends?
Tony built around those questions because he didn't have the luxury of ignoring them.
The Better Question to Ask About AI Invoice Automation
Every restaurant technology company is talking about automation. A recent FSR magazine survey on restaurant technology adoption found invoicing automation among the fastest-growing categories year over year, with faster service and time savings cited as the top reasons operators moved to automate it.
The better question isn't whether a platform can read an invoice. Most modern tools can read a clean one.
The better questions are:
- What percentage of this workflow is repetitive and safe to automate?
- What happens when the invoice doesn't match the delivery?
- How are credits, substitutions, catch weights, and handwritten notes handled?
- Who reviews the exceptions?
- How quickly does corrected invoice data reach recipes and food cost reporting?
- Can the system show where a price change affects menu profitability?
Catch weights are a good test case. A misread catch weight throws off ingredient yield everywhere that ingredient shows up on the menu, not just on the invoice where the error happened.
These questions reveal more than an automation percentage ever will.
For accounts payable automation, restaurant operators need both speed and accountability. A fast system that produces unreliable data simply allows incorrect information to travel further, faster.
The best systems automate the routine work while making the exceptions impossible to ignore.
Someone Still Has to Be in the Driver's Seat
AI will keep getting better at reading invoices. Tony doesn't question that.
What he questions is the assumption that restaurants will eventually stop being messy enough to require judgment.
Vendors will substitute products. Deliveries will arrive damaged. Prices will change. Someone will scribble a note in the margin. What was ordered, delivered, invoiced, and used won't always match perfectly.
"Someone still has to be in the driver's seat."
That isn't a criticism of AI. It's the clearest explanation of how restaurants should use it.
Let technology process the predictable work. Let people investigate the exceptions. Then make sure the final data flows into the recipes, costs, and reports operators use to make decisions.
The future of restaurant invoice processing probably isn't a completely autonomous back office.
It's a back office where people spend less time entering information and more time making sure it's true.
See how meez keeps invoice data, recipe costs, and your accounting system in sync with meez x Restaurant365.
Listen to the full episode: Tony Aiazzi on bootstrapping, Over Easy Office, and bespoke chef knives. Plus AI as the last mile and singing Chef David Burke's praise.
FAQ: Restaurant Invoice Processing and AI
What is restaurant invoice processing?
Restaurant invoice processing is the work of collecting, reading, categorizing, approving, and recording invoices from food, beverage, and operating vendors. Modern platforms can capture invoices through photos, scans, email, or electronic vendor feeds and send the information into a restaurant's accounting system.
How does restaurant invoice automation work?
Restaurant invoice automation uses technologies such as OCR and AI to extract vendor names, invoice numbers, dates, totals, and individual line items. The software then maps that information to ingredients, products, or accounting categories, reducing the amount of information managers and bookkeepers have to enter manually.
How much of invoice processing can AI automate?
AI can process most clean, standardized invoices with limited intervention. The remaining exceptions, such as credits, handwritten notes, product substitutions, catch weights, and recycled invoice numbers, often still require a trained person to review what happened.
Why do restaurants still need manual invoice review?
Restaurant deliveries don't always match the original order or the printed invoice. A person may need to confirm a substitution, apply a credit, correct a quantity, investigate a duplicate, or interpret a handwritten note. Manual review is most valuable when the situation requires operational context rather than simple data extraction.
What is the difference between invoice automation and accounts payable automation for restaurants?
Invoice automation focuses on capturing and categorizing the invoice. Accounts payable automation for restaurants extends the workflow into approvals, vendor reconciliation, payment scheduling, and payment execution.
How does invoice accuracy improve food cost control?
Accurate invoice data allows ingredient and recipe costs to reflect what a restaurant is currently paying. When invoices are delayed or categorized incorrectly, operators may not see cost increases until the monthly P&L arrives. Connecting invoices to recipes helps teams identify margin drift sooner.
How does invoice data support cost variance analysis?
Cost variance analysis compares what a restaurant expected to spend with what it actually spent. Accurate invoice quantities, prices, credits, and product mappings help operators determine whether a variance came from purchasing prices, waste, portioning, substitutions, inventory errors, or another operational issue.
Which restaurant invoice details are hardest for AI to process?
Common challenges include handwritten notations, catch weights, vendor-specific item codes, substitutions, credits, damaged documents, and reused invoice numbers. These exceptions often need to be reviewed by someone who understands restaurant purchasing and receiving.
Should a restaurant build or buy invoice processing software?
For most operators, buying an established platform is faster and less expensive than building one internally. Purpose-built providers already have workflows, vendor mappings, and exception-handling processes developed across many restaurant clients.
Does restaurant invoice automation integrate with Restaurant365?
Many restaurant invoice and AP automation platforms integrate with accounting and ERP systems such as Restaurant365. This allows approved invoice data to flow into financial reporting without requiring teams to enter the same information twice.
