Practical AI for the Second Half of Your Career
There is a question every chef and restaurant manager quietly dreads: “What does this dish actually cost us right now?”
Not last month. Not when you first priced the menu. Right now — after three supplier price increases, a protein swap, and a yield issue you’ve been meaning to deal with for two weeks.
For most of my career, that answer lived somewhere between an old Excel file, a stack of invoices on a desk, and a rough number in the back of my head. It worked — until it didn’t.
This is the story of how I stopped guessing, and built a simple AI-powered costing system using two tools I already had: Notion and Claude. No server. No developer. No coding background required.
The Real Problem With Food Costing
The issue was never that I didn’t care about food cost. The issue was access to accurate, up-to-date data at the exact moment I needed it.
Supplier prices change constantly. An invoice arrives by email, gets forwarded to accounting, and by the time it lands in a spreadsheet — if it ever does — the information is already a week old and buried in a folder no one opens.
Meanwhile, you’re repricing a special, training a new sous chef on batch prep, or reviewing a banquet proposal, and you’re working with numbers that may or may not reflect what you’re actually paying today. That gap costs real money.
I knew the solution wasn’t a bigger spreadsheet. I needed a live, searchable record of what I’m paying, connected to something smart enough to do the math on demand.
Step 1: Gmail as the Invoice Capture Layer
The first insight was simple: almost every supplier invoice I receive arrives in my Gmail inbox. PDFs, email attachments, forwarded confirmations — they all land in the same place.
Instead of manually filing them or waiting for accounting to process them, I connected Claude to my Gmail using MCP integrations. If that sounds technical, here’s a plain-language way to think about it: you know how when you install a new app on your phone, it asks permission to access your camera, your contacts, or your location? MCP works the same way — it’s the system that gives Claude permission to access your Gmail and your Notion database, so it can move between them on your behalf.
Once those permissions were in place, I created a dedicated Claude Project inside Claude Cowork, configured to act as a restaurant chef’s assistant. I then set a schedule for the agent to automatically access my Gmail inbox once a week, read incoming supplier invoices, and transfer the relevant line items directly into my Notion database — no manual entry, no trigger needed on my end.
By the time I open Notion, the week’s invoices are already there.
Step 2: The Notion Database — A Tiny Data Center for My Kitchen
The Notion database is the heart of the system. It is not complicated. Each row is one invoice line item, and every row captures:
- Supplier name — who sent the invoice
- Category — Meat, Produce, Dairy, Dry Goods, etc.
- Item name — exactly as it appears on the invoice
- Quantity — pack size or unit count
- Price — what I paid, on that date
That’s it. No formulas, no pivot tables, no macros. Just clean, consistent records that stack up over time into something genuinely useful.
Because every item is its own row, the database becomes searchable in ways a spreadsheet never was. I can ask Claude: “How much is the striploin portion right now?” and get a current price pulled directly from my Notion data in seconds — not because I built a search function, but because Claude can query the database in plain English on my behalf.
Step 3: Real Costing in Action — From Batch Prep to the Plate
Here is a real example of how this works in practice.
I gave Claude the full recipe for our garlic aioli — every ingredient, every quantity — and asked: “How much does it cost to make a 10-litre batch of garlic aioli?” Claude pulled the current prices from my Notion invoice database, calculated the total batch cost, and returned the answer in seconds.
But that is only half the story. That garlic aioli is served as a dipping sauce with fries. So I followed up: “If one portion is 2 tablespoons (30ml), what is the cost per portion from that batch?” Claude scaled the batch cost down to the exact portion size I serve — automatically. No formula. No spreadsheet. No manual math.
If I want a record or need to share it with my team, I can export the full costing breakdown to an Excel file for editing. The answer lives on my screen when I need it, and in a file when I want it.
When I need to stress-test a price change, I ask:
*”If my produce supplier raises prices by 10%, what happens to the per-portion cost on this dish? Show me the before and after.”*
That kind of sensitivity analysis used to take 20 minutes with a spreadsheet. Now it takes about 45 seconds.
Step 4: What “No Server, No Code” Actually Looks Like in Practice
The system currently tracks 9 kitchen suppliers, all of whom deliver at least once a week. Every invoice flows into Notion automatically, giving me a running record of what I am spending per supplier — and more importantly, how that spending relates to sales.
But the real test of any data system is whether the numbers mean something in the real world. In this case, they do. I can spot price fluctuations on specific ingredients over time. If spending with one supplier spikes unexpectedly, I can investigate immediately — was it a price increase, a larger order, or a billing error? That kind of visibility used to require a full accounting review. Now it takes one question to Claude.
The system has also grown beyond the chef’s station. I built a simple internal web app — no server, no developer, no hosting infrastructure — that gives the restaurant owner live access to cost tracking and sales data directly on his phone, without logging into the POS. He sees what he needs, when he needs it, in plain language. That is the version of restaurant technology I believe most independent operators actually need.
Step 5: How the System Maintains Itself
Maintaining the system takes far less effort than I expected. Claude flags its own uncertainties during the weekly sync — if an invoice is difficult to read, or a new supplier appears that it hasn’t seen before, it asks me to confirm before recording anything. I don’t check every line item manually; the agent tells me exactly where it needs my attention.
For paper invoices or handwritten delivery slips, I simply take a photo with my phone. Claude uses OCR — optical character recognition, the same technology that lets your phone scan a document or read a sign — to convert the image into a structured database entry. If the image is unclear, it flags it and asks me to verify rather than guessing.
My deeper review only happens when something looks genuinely off — specifically, if ingredient spending as a percentage of sales spikes beyond what I would expect. That is when I go in and investigate. Everything else largely takes care of itself.
What This Changed Beyond the Numbers
The practical impact is real: faster menu pricing, better margin conversations, and less guessing on banquet proposals. But the bigger shift was confidence.
I stopped avoiding the cost conversation because I stopped dreading the data retrieval part of it. When a supplier calls to discuss a price increase, I can pull up current landed costs before the call ends. When ownership asks how food cost is trending, the answer is already organized and waiting.
More importantly, this experiment didn’t just solve a costing problem — it confirmed a business hypothesis I had been sitting on for years.
As a former restaurant owner, I know firsthand that staying on-site every hour to monitor operations isn’t realistic. The operators who survive long-term are the ones who can read their numbers remotely, spot patterns early, and make strategic decisions before small problems become expensive ones.
What this system showed me is that the data was always there — it was just trapped in inboxes, POS reports, and spreadsheets that no one had time to reconcile. Once you organize it and give AI access to it, the insights follow naturally.
That realization is the foundation of F&B Central — a reporting tool built around daily sales, food cost, and labor cost data, designed to surface patterns and support strategic management. The framework isn’t borrowed from restaurant tech. It comes from corporate finance: KPI management, Management by Objectives, performance measured against actual data — not gut feel or vague assumptions. The difference is that now, AI makes that level of analysis accessible to any operator, not just the ones with a finance team behind them.
How to Start This Weekend
You don’t need to replicate the full system immediately. Here is a minimum viable version anyone can build:
- Create a Notion database with five columns: Supplier, Category, Item, Quantity, Price.
- Enter 20–30 line items from your most recent invoices — manually, to start.
- Pick one dish or batch prep you want to cost accurately.
- Copy the relevant Notion rows into Claude with a prompt like:
- Check the math manually once. If it is right, trust it going forward. If it is off, look for a unit conversion error — that is almost always the culprit.
*”Here are my current ingredient prices. Calculate the cost per portion for [Dish Name] using these quantities: [list them]. Show your work step by step.”*
Once you have done this for three or four dishes, you have a working costing workflow. After a month of invoices, you have a data center.
An Honest Word on Limitations
This system is only as accurate as the data going into it. If a supplier sends a late invoice or an item name is inconsistent between deliveries, Claude cannot fill in the gaps — it will ask you to verify instead.
It is also not a real-time POS integration. It is a costing assistant and a spending tracker, not a full food cost management platform. For deeper operational integration, you would need something more robust — which, again, is exactly what I am building with F&B Central.
But for what it is — a fast, searchable, AI-queryable invoice system that a non-technical person built and runs alone, covering 9 suppliers and an entire kitchen operation — it is already paying for itself many times over.
Frequently Asked Questions
Do I need to know how to code to build this?
No. The entire system uses Notion (a no-code database tool), Claude (an AI assistant), and your existing Gmail account. If you can set up a table in Notion and type a question into a chat window, you have everything you need.
Is my invoice data safe if I connect it to Claude?
Always check your organization’s data policy before connecting any AI tool to business email. For independent operators, Anthropic (the company behind Claude) publishes its data usage policies clearly. Use your judgment about what information you share, and avoid uploading anything that contains customer personal data.
What if my suppliers send paper invoices instead of emails?
Take a photo with your phone. Claude can read printed and handwritten invoices using OCR — the same technology your banking app uses to scan cheques. If the image is unclear, it will ask you to clarify rather than guessing.
Can I use this for labor cost too, not just food cost?
Yes, in principle. Labor cost data can be structured into Notion the same way. The costing logic Claude applies is the same — you define the inputs, and it calculates the outputs. This is one of the areas F&B Central is being built to handle more formally.
What is F&B Central and how is it different from this system?
F&B Central is a purpose-built restaurant management reporting tool currently in development. Where this Notion-and-Claude setup is a lean, personal workflow, F&B Central is designed to be a structured platform for operators who want daily EOD reporting, trend analysis, and strategic KPI dashboards — without needing a developer or a POS upgrade to get started.