Small Business AI: How Jen Turned Weekend Cooking Into a Real Income
Jen never planned to run a business. She is a homemaker and a mother of two who loves to cook, and she used to post pictures of her weekend dishes in a friends' group chat. Then the messages started: "Can you make that for my sister's birthday?" A few weeks later she was handling anniversary orders for 20 to 30 people between school runs.
Cooking was the easy part. Pricing, quantities, who paid the advance, what to buy, what to do when a party is cancelled at the last minute, and whether any of it was profitable were the hard part. This guide follows how we helped Jen set up Claude as a back-office assistant for her home food business: the tracker, the costing, the weekly routine, a last-minute cancellation, and how she turned happy customers into referrals. At the end, the same building blocks are laid out as 15 uses for any small business, freelancer or solo professional. Names and numbers are changed or illustrative.
Where Should a Small Business Start With AI? (Quick Decision Guide)
Before the story, here is the short version. Answer the questions in order and stop at the first "yes". Look for the first step that matches your situation, not the most exciting one.
YES → start with a simple tracker (Excel workbook built with AI)
NO
NO → start with order costing and pricing
YES
YES → start with FAQs, reply drafts and weekly announcements
NO
YES → start with a seasonal calendar, a cancellation plan and feedback collection
NO → review your numbers monthly and add one workflow at a time
Takeaway: Jen answered "yes" to Q1 and Q2, which is why her story starts with a tracker and costing, not with marketing.
The Everyday Picture: AI Is the Sous-Chef for the Paperwork
In a restaurant kitchen, the head chef decides what goes on the plate. The sous-chef preps, organizes and keeps the station running. Jen's AI assistant plays the sous-chef role for the office side of her business. It never tastes the food, never meets the customer and never signs off on a price. It prepares, calculates and drafts, and Jen decides.
Takeaway: if a task needs taste, trust or a final decision, it stays with the head chef.
Before and After: What Changed in Jen's Week
The change was not a big technology project. It was a handful of small habits. Compare the two columns below.
Jen's story is also told in video, following her from scattered orders to a back-office routine she controls.
Step 1: Check the Idea Before Spending Money
Jen did not ask AI for 100 business ideas. She already had a skill, a small circle of people who liked her food and a pattern: weekend requests, birthdays and anniversaries. The first job was to test whether that pattern could carry a small business without buying equipment or advertising.
Her first prompt gave Claude her real situation and asked for a small, realistic start.
I am a homemaker and a mother of two. Friends have started asking me to cook for weekends, birthdays and gatherings of 20 to 30 people. I want to explore a small home-based food business without a large investment. Help me build a practical plan covering weekend pre-orders, party orders, pricing, packaging, delivery, marketing and monthly expenses. I have limited time, so suggest a realistic way to start small.
| Decision area | What Jen had to decide | Who answers it |
|---|---|---|
| Business model | Weekend pre-orders, party catering or both | Jen, with AI laying out the options |
| Starter menu | Only dishes she already cooks well | Jen |
| Capacity | How many orders she can handle without wrecking her week | Jen |
| Demand | Will neighbours really pay for this? | Real conversations, not AI |
Apply: AI can organize the thinking, but it cannot know whether enough people near you will buy. That check is still a human conversation.
Step 2: Give the Assistant a Home (A Claude Project)
Re-explaining the business in every chat gets tiresome fast. Jen created one dedicated Claude Project for the food business and wrote standing instructions once. If you have never done that, our beginner walkthrough on setting up a Claude Project from scratch covers every click. Project features and plan limits change often, so check your own screen for what is available to you.
The instructions we used for Jen's project looked like this.
You are my small-business planning assistant for a home-based food business: weekend specials, festive dishes and party catering. Help with menus, ingredient quantities, costs, pricing, order tracking, monthly summaries and WhatsApp messages. Keep costs separate: ingredients, packaging, delivery, labor and other expenses. Never invent supplier prices, customer orders or financial figures. Ask for missing information or clearly label estimates. Do not treat any menu, price or expense as approved until I say so. Keep answers practical and non-technical.
The two lines that mattered most were "never invent figures" and "label estimates". They are the guardrail against the problem we describe in our piece on why AI gives confident wrong numbers.
Step 3: Build the Order Tracker (Our Two Attempts)
This was the most useful part of the whole setup, and it is where we saw the clearest difference between a giant request and a step-by-step one. We built the same tracker twice with Claude, using the same business details.
| Attempt | Prompt we used | What happened |
|---|---|---|
| 1. Giant prompt | "Build me a complete Excel system for my food business with orders, recipes, costs, expenses, payments and a dashboard." | ❌ Looked impressive, but the dashboard counted an advance payment and then the final payment as two separate sales. Order and payment details also had to be typed in more than one place. |
| 2a. Define the job | "List the worksheets this business needs, and what each one should contain. Do not build anything yet." | ✅ Seven sheets agreed on: customers, orders, menu and recipes, ingredient costs, expenses, payments, dashboard. |
| 2b. Build one sheet at a time | "Create the orders and payments sheets first, with dropdowns for status. Enter each order once. Use clearly marked sample data." | ✅ A clean tracker with Order ID, event, delivery date, total, advance, balance and status. |
| 2c. Add the dashboard with a rule | "Build the dashboard. Count revenue once per order. An advance payment is not extra revenue when the balance arrives." | ✅ Revenue, costs, expenses, outstanding balance and estimated profit calculated automatically. |
| 2d. Test it | "Explain each formula, then test the workbook with three sample orders and show me the expected totals." | ✅ Totals matched hand calculations before any real order was entered. |
Lesson: one small instruction, "count revenue once per order", fixed a mistake the giant prompt never thought to ask about. If you want more spreadsheet problems AI can untangle, our look at Excel problems AI can solve in minutes has more patterns.
What the Finished Workbook Tracks
Here is the workbook at a glance. Below the tiles, each sheet is shown the way it looks once opened, with sample data. Every name, price and total in these sample sheets is made up for illustration.
Sheet 1: Orders (sample data)
One row per order. Status comes from a dropdown, so spelling never drifts.
| Order ID | Customer | Event | Dishes | Portions | Allergy note | Delivery date | Pickup / Delivery | Status |
|---|---|---|---|---|---|---|---|---|
| J001 | Customer A | Birthday | Chocolate cake, rice dish, salad | 25 | Nut-free | Nov 8 | Delivery | Confirmed |
| J002 | Customer B | Family dinner | Main dish, two sides | 8 | None | Nov 9 | Pickup | Pending |
| J003 | Customer C | Anniversary | Starter, main, dessert | 30 | Vegetarian | Nov 15 | Delivery | Confirmed |
Sheet 2: Costing (sample data)
One block per dish. The last column is a formula: quantity times unit price. When a price changes, update it once and every recipe that uses it updates too.
| Dish | Ingredient | Quantity | Unit price | Line cost |
|---|---|---|---|---|
| Chocolate cake | Flour | 1 kg | $3.00 per kg | $3.00 |
| Chocolate cake | Sugar | 0.5 kg | $2.00 per kg | $1.00 |
| Chocolate cake | Butter | 0.5 kg | $5.00 per kg | $2.50 |
| Chocolate cake | Chocolate | 0.4 kg | $8.00 per kg | $3.20 |
| Total | $9.70 |
Sheet 3: Payments (sample data)
The balance column is a formula (total minus advance). This sheet is why an advance payment never gets counted as a second sale.
| Order ID | Order total | Advance | Advance date | Balance due | Method | Payment status |
|---|---|---|---|---|---|---|
| J001 | $180 | $60 | Nov 2 | $120 | Bank transfer | Advance paid |
| J002 | $75 | $0 | - | $75 | Cash on pickup | Pending |
| J003 | $240 | $100 | Nov 4 | $140 | Bank transfer | Advance paid |
| Totals | $495 | $160 | $335 |
Sheet 4: Monthly Dashboard (sample data)
Everything here pulls from the other sheets, so nothing is typed twice. Notice that order value ($495) and cash actually received ($160) are shown separately.
| Dashboard line (sample month) | Value |
|---|---|
| Orders received | 3 |
| Order value (revenue counted once per order) | $495 |
| Cash received so far | $160 |
| Outstanding balances | $335 |
| Ingredient costs | $170 |
| Packaging and delivery | $65 |
| Other business expenses | $30 |
| Estimated operating profit | $230 |
| Best seller | Chocolate cake |
Apply: profit here is an estimate on booked orders. Cash in hand is lower until the balances arrive, and that gap is the number Jen watches every Sunday.
Step 4: Know What One Order Really Earns
Sales are not profit. Say Jen quotes $360 for a 30-person birthday order. The table below uses illustrative numbers to show how quickly that shrinks once everything is counted. These figures are examples, not Jen's actual books.
| Cost part (illustrative) | Amount |
|---|---|
| Ingredients | $120 |
| Packaging | $25 |
| Delivery | $15 |
| Jen's time allowance | $60 |
| Left from a $360 order | $140, before overhead and taxes |
Where the $360 goes (illustrative split)
Colors: yellow ingredients, blue packaging, purple delivery, pink Jen's time, green what remains.
To get this from Claude instead of a calculator, Jen supplies her real ingredients, quantities, prices and hours and asks for pricing scenarios.
Here are the ingredients, quantities, supplier prices, preparation hours, packaging and delivery cost for a 30-person order. Calculate the total cost, cost per person and my estimated labor cost. Then show three selling prices that leave a 20%, 30% and 40% margin, and explain the difference between markup and margin. Do not assume any cost I did not give you.
Apply: after her first few orders, Jen could see which dishes looked popular but left little behind once prep time was counted.
Step 5: A Weekly Loop That Does Not Eat the Weekend
The tools only help if they fit a routine. This is the loop Jen settled on. AI does the prep work in each step, and she keeps the decisions.
AI suggests dishes; Jen picks only what she can cook well.
AI drafts the message; Jen checks prices and deadline.
Orders entered once in the tracker; AI builds the shopping list.
No AI here. This is the head chef's part.
AI summarizes revenue, costs, unpaid balances and feedback; Jen decides what to change.
For the announcement step, a prompt like this keeps the message friendly and honest.
Write a friendly WhatsApp announcement for this weekend's menu. Dishes and prices: [paste actual list]. Orders close Friday 5 PM, pickup Saturday 4 to 7 PM. Keep it conversational. Do not invent discounts, ingredients or availability.
Step 6: Seasonal Planning and Extra Income Ideas
Once the basics worked, Jen used Claude to look ahead. A 12-month calendar of occasions, matched with dishes she already makes, showed where demand might spike: Thanksgiving meal packages, Christmas gift boxes, summer party platters, regional festivals. Because Jen asked for ideas that reuse existing recipes, the plan stayed realistic for one kitchen.
| Extra idea | What AI helps with | Honest catch |
|---|---|---|
| Recipe collection | Structure, quantities, descriptions | Every recipe must be tested; you need rights to any photos |
| Party planner templates | Quantity calculators, checklists | Nobody buys what nobody finds |
| Small online workshop | Outline, ingredient list, follow-up sheet | Takes real teaching time |
| Repeat-order packages | Package design and reminders | Still active cooking, not passive income |
Apply: test any of these with a small, cheap experiment before building it.
Step 7: When a Party Is Cancelled at the Last Minute
Sooner or later it happens. A 30-person anniversary order is confirmed, the ingredients are bought, and two days before the event the customer cancels. For a home business there is no corporate budget to absorb that. Jen used Claude for three jobs: sorting out what was actually lost, recovering as much as she could, and making the next cancellation hurt less. The numbers below are illustrative, not Jen's real figures.
Part A: Catch Up and Recover the Loss
What is bought, what is prepped, what is sunk?
Freeze, reuse in another order, or shift to the weekend menu.
A short-notice pre-order for smaller packs, announced the same day.
A polite message that applies the written policy.
Mark the order Cancelled and record the cost for the monthly review.
The first prompt Jen used was a triage prompt. She pasted in her real list of purchases and prep, and asked Claude to organize it, not to decide for her.
A customer cancelled a 30-person order two days before delivery. Here is what I already bought or prepared: [paste list with costs and storage notes]. Sort every item into: can be frozen or stored, can be used in my regular weekend menu, must be used within two days, and cannot be reused. Show the cost of each group so I can see the real loss. Do not assume I can reuse anything I did not list.
| Group (illustrative) | Example items | Amount | Plan |
|---|---|---|---|
| Store or freeze | Flour, sugar, butter, chocolate, packaging | $45 | Keep for next orders |
| Use in weekend menu | Rice, spices, vegetables | $30 | Fold into this weekend's specials |
| Use within two days | Fresh produce, dairy | $25 | Short-notice sale or family meals |
| Cannot reuse | Custom-printed boxes, prepped items | $20 | Real loss; check the advance policy |
Apply: a $120 purchase did not turn into a $120 loss. Once sorted, only a fraction was truly gone, and that was the number Jen compared against the advance.
For the surplus, Jen asked for a quick, honest offer she could post the same day.
Write a short WhatsApp message offering a limited short-notice menu for this weekend, using these leftover ingredients and dishes: [list]. Make it conversational, mention the pickup window and order deadline, and do not invent discounts or availability beyond what I listed.
And for the customer, a calm reply that sticks to the policy Jen had already written down.
Draft a polite, short reply to a customer who cancelled two days before their event. My written policy is: [paste policy]. Thank them, explain how the advance is handled under that policy, and leave the door open for a future order. Do not promise a refund or credit that my policy does not state.
Refund and deposit rules depend on where you live and what you agreed with the customer in writing. Jen checked her local requirements and kept the policy simple and visible. AI drafts the wording, not the legal position.
Part B: Plan Ahead for Uncertainty
The real fix was not a better reply. It was a plan made before the next cancellation. Jen asked Claude to stress-test her setup with "what if" questions and turned the answers into rules.
Here is how my party orders work today: [describe deposit, buying schedule, menu]. List the five most likely ways an order could be cancelled or changed late, estimate which ones would cost me the most using the cost sheet I gave you, and suggest practical rules to reduce the damage. Separate rules I can adopt immediately from ones that need customer agreement.
These are the rules Jen ended up writing into her project instructions and order confirmations.
| Risk | Rule | Where AI helps |
|---|---|---|
| Late cancellation | Advance required to confirm; written cancellation window | Drafts the policy wording and the confirmation message |
| Perishable waste | Buy fresh items as late as possible; prefer freezable dishes for big orders | Builds a shopping schedule split into "buy early" and "buy late" |
| Headcount change | Final count due a few days before; small changes allowed, big ones re-quoted | Recalculates quantity and cost for a new headcount |
| Surplus after a cancellation | Keep a standby list of regular customers for short-notice offers | Drafts the short-notice message in minutes |
| Cash-flow gap | Keep a small buffer instead of spending every payment | Shows how many cancelled orders the buffer could absorb |
Lesson: the cancellation cost her real money once. After that it became a rule, a template message and a line on her dashboard. In the tracker, the order status simply changes to Cancelled, and the dashboard excludes it from revenue while still counting the unrecovered cost.
Step 8: After the Party, Turn Feedback Into Better Menus and Referrals
Jen noticed something simple. After a party, guests said nice things to the host, the host said nice things to Jen, and then it all vanished into chat history. She started collecting that feedback on purpose and using it two ways: to improve the food, and to politely ask for introductions.
Mark the order Completed in the tracker.
A short, friendly message a day or two later.
Positive, negative and suggestions, grouped by dish.
One row per comment, with an action to take.
Fix what is fixable, then invite happy customers to share.
Collecting Feedback Without Being Awkward
Write a short, warm WhatsApp message to thank a customer for their birthday order and ask three quick questions: what they and their guests liked most, anything that could be better, and whether portion size and timing were right. Keep it under 80 words, casual, and easy to answer in one reply.
Sorting the Replies
Customers rarely answer in neat bullet points. Jen pasted the replies, with names removed, and asked Claude to organize them.
Here are customer replies after recent events, with names removed: [paste]. Sort each comment into positive, negative or suggestion, and note which dish it mentions. Then list recurring themes. Do not treat one comment as the view of all customers, and tell me when a theme has only one mention.
The Feedback Sheet (sample data)
Claude added a Feedback sheet to the workbook. The rows below are made up to show the structure.
| Order ID | Dish | Type | Comment (summary) | Action | Status |
|---|---|---|---|---|---|
| J001 | Chocolate cake | Positive | Guests asked who made it | Feature on weekend menu | Done |
| J001 | Rice dish | Negative | A bit spicy for children | Offer a mild version | To do |
| J003 | Dessert | Suggestion | Smaller portions for a big party | Add mini-size option to costing | Test first |
| J003 | Delivery | Negative | Arrived a little late | Add buffer time to the schedule | Done |
Each month Jen asked Claude to summarize the sheet: which dishes were praised, which complaints repeated, and which fixes were still open. One mention is a note. Three mentions is a pattern worth acting on.
Turning Happy Customers Into Referrals
Referrals worked best when they were asked for gently and only after a good experience. Jen only reached out to customers whose feedback was positive, and she never asked for anything that was not true.
Write a friendly message to a customer who gave very positive feedback after their anniversary order. Thank them, and gently ask whether they would be comfortable sharing my menu with friends planning a gathering. Offer to send a short menu they can forward. Do not invent discounts, rewards, reviews or testimonials. Keep it warm and under 90 words.
Apply: feedback tells you what to fix, and referrals only work when the food earned them. AI organizes both, but the cooking still has to be good.
Where AI Helped Most (Relative, Illustrative)
These bars reflect our own impressions from Jen's setup. They are relative and illustrative, not a benchmark and not measured time savings.
Order tracking and payment status
Costing and price scenarios
Sorting feedback and spotting themes
WhatsApp, referral and short-notice message drafts
Seasonal menu ideas
Knowing local demand (still needs real people)
Reading it: AI was strongest where the work was structured and repeatable, and weakest where the answer lived in other people's heads.
What to Hand to AI and What to Keep: A Stoplight
Not every task deserves the same trust. We sorted Jen's tasks into three lights. This idea matches what we cover in deciding what to delegate to AI.
What Broke, and What We Would Not Trust
Jen's setup worked, but not without bumps. This is the failure log from our build.
| What went wrong | Why it mattered | Fix |
|---|---|---|
| Advance and final payment counted as two sales | Revenue looked higher than it was | State the rule in the prompt and test with sample orders |
| Sample prices treated as real | Costing looked precise but was invented | Mark samples clearly; replace with real supplier prices |
| Menu ideas too ambitious | Some dishes were beyond one home kitchen | Ask for "prep complexity" and "advance preparation" for every idea |
| Cancellation reply sounded too generous | An unplanned promise of a refund or credit | Paste the written policy and say "do not promise anything it does not state" |
| One loud comment treated as a trend | Menu changed over a single opinion | Ask AI to flag themes with only one mention |
| Customer details pasted into chats | Unnecessary privacy risk | Use first names or order IDs; check your tool's privacy settings |
When Not to Use AI This Way
- When you have no real records yet. AI cannot organize numbers you have not collected.
- When the output affects health, money owed or legal duties. Food safety, allergens, refunds, taxes and permit rules vary by location and need official sources or a professional.
- When checking the result takes longer than doing the task yourself. We saw this in our own look at when AI creates extra work.
- When you are about to let AI send messages, refunds or payments on its own. Start with drafts and human approval; our guide to running AI agents with permissions and monitoring explains why.
15 Ways Small Businesses Can Use AI (Jen's Kitchen and Beyond)
Jen's story is one example. The same building blocks fit a consultant, a landscaper, an online store or a tutor. Here are fifteen practical uses, grouped by the job they do. Each card shows the idea and a small example.
Two of these have deeper walkthroughs on our site: handling a busy inbox with AI and turning meeting notes into action lists.
What Should You Automate First? A Scoring Checklist
Score each candidate task from 0 to 2 per question (0 = no, 1 = partly, 2 = yes). Tasks scoring 8 or more are good first experiments.
| Question | Why it matters |
|---|---|
| Does it happen often? | Weekly tasks repay setup effort faster than yearly ones |
| Is the process predictable? | Clear inputs and outputs give better results |
| Can you check the result quickly? | Easy review means safer use |
| Is a mistake cheap to fix? | A wrong draft is cheaper than a wrong payment |
| Does it really save time? | Heavy monitoring can cancel the benefit |
Jen's First 30 Days: A Plan You Can Copy
If Jen were starting again today, this is the order we would suggest. Swap the food details for your own skill and the plan still works.
| Week | What Jen does | How AI helps |
|---|---|---|
| 1 | Lists her best dishes, her likely customers and her weekly free hours; sets up the Claude Project | Turns the list into a small plan; she validates demand with real people |
| 2 | Costs her three most-requested dishes with real supplier prices; builds the orders and payments sheets | Builds the sheets step by step and tests them with sample orders |
| 3 | Announces a small weekend menu; takes real orders; writes down her deposit and cancellation rules | Drafts the announcement and the written policy; she verifies prices and wording |
| 4 | Sends feedback requests, reviews the dashboard and decides what to change or keep | Sorts feedback, summarizes the month and suggests improvements to test |
Final Thoughts
Jen's cooking did not change. What changed was that she could see her orders, her costs and her week clearly, and she had a plan for the bad days as well as the good ones. AI did not give her a business. It took over the preparation work around a business she was already building.
That is the pattern worth copying: start with a real skill, keep a human at every decision, test small, plan for the cancellation before it happens, and listen to what customers say afterwards. Your first useful step is probably smaller than you think.
Frequently Asked Questions
What is the easiest way for a small business to start with AI?
Pick one repetitive, low-risk task with a clear process, such as drafting customer replies or tracking orders, and test it on real examples before adding more.
Can someone without technical skills do this?
Yes. Describe what you need in plain language, test the result with sample data, and ask the assistant to explain anything unclear.
Can Claude really create an Excel tracker?
Claude can help design and generate spreadsheet files with formulas, depending on the features available on your plan and screen. Always test formulas with sample orders before relying on them.
What should I do when a customer cancels at the last minute?
List what you already bought and prepared, separate what can be stored, reused or sold quickly from what is truly lost, and apply the cancellation terms you agreed in writing. AI can organize the list and draft the messages, but check local rules and your own policy before deciding anything about refunds.
How can AI help me get customer feedback?
It can draft a short request, group replies into positive, negative and suggestions, and flag which themes repeat. Remove names and contact details before pasting replies, and treat a single comment as a note, not a trend.
Can AI help me get referrals?
It can draft a polite message asking happy customers to pass your menu on to friends. Never use it to invent reviews, testimonials or rewards.
Is AI reliable for business finances?
It can organize and summarize numbers you give it, but it can also make confident mistakes. Verify totals yourself and talk to a qualified professional about taxes and compliance.
Is it safe to share customer information with an AI tool?
It depends on the service and its settings. Avoid pasting names, phone numbers, addresses or payment details unless necessary, and read the privacy controls of the tool you use.
Can AI replace employees in a small business?
It can take over parts of repetitive workloads, but the effect depends on the business and the amount of human oversight. A better starting question is which tasks it can assist with.
Does AI create passive income?
Not by itself. Digital products and templates may reduce repeated effort, but they still need marketing, updates and customer support.
Do I need several paid AI subscriptions?
No. Start with one assistant and one spreadsheet, and add tools only when a clear need appears.
Comments
Post a Comment