Case studies

Real problems. Real systems.

From a café and a guided tour company, both of them my own businesses: what hurt, why we pulled the pieces together, what we built, and how it works.

“You’re not really clocked in until the truck checks out.”

Why bring it all together

One place for the data is the whole point.

Most small businesses already have the information. It sits in a point of sale, a booking system, a spreadsheet, a text thread and somebody’s memory. Every project below starts by putting the pieces in one place, because of what that makes possible.

Typed once

A fact entered at the source reaches the schedule, the website and the office on its own. Retyping is where errors and delays come from.

Questions across systems

Sales next to staffing. Bookings next to guides. Counts next to orders. Answers that were out of reach when each sat in its own system.

Each new piece costs less

Once the data is together, the next tool does not start from scratch. It reads what is already there.

AI can help

An assistant can only work with what it can see. Data in one place is what lets it look things up, draft the work and flag what needs you.

A café and bakery

One app for the whole crew

Problem: Shifts, swaps, messages and the handbook lived in five different places.
Why: When shifts, swaps and messages share one record, the schedule, the hours and the clock-ins stop disagreeing. It also gives the next job, such as checklists or training, a place to live that is not a sixth app.
Built: A staff app on each person’s own phone.
How: Next shift → clock in → pick up an open shift → message a manager single source of truth

Staff app home screen showing the next shift, a clock-in button, open shifts and quick help — Staff app design, sample data

Labor that follows sales

Problem: Schedules came from habit. Rushes ran short, slow hours ran heavy.
Why: Sales sat in one system and schedules in another, so nobody compared them. Once register history sits beside the schedule, staffing becomes a question the numbers can answer.
Built: A schedule builder that reads sales by the quarter hour and suggests who works when.
How: Register history source → demand by hour forecast → staffing need → draft schedule a manager approves human in the loop

Illustrative day, not real figures

Ordering from the main kitchen

Problem: A satellite café ordered by text message. The kitchen guessed.
Why: Counts and orders lived in text messages and in people’s heads. In one place, the kitchen can see what the café actually uses and prepare for need, not for a guess.
Built: A count screen with a par level for every item. The gap becomes the order.
How: Count on a phone → compare to par rule → pick sheet prints → order emailed on schedule trigger

Illustrative count, not real figures

Hiring from the front window

Problem: Applications arrived on paper, or not at all.
Why: A paper application ends at a desk. When it lands in the staff system, nothing is retyped and a new hire’s details are already there on day one.
Built: A QR code in the window and a phone-friendly application.
How: Scan → apply → interview sheet prints → a hire lands in the staff system single source of truth

A guided tour company

Tour itineraries, kept in one place

Problem: Trip details lived in Word files and got retyped for the website, the guides and marketing.
Why: One record for each trip day feeds the website, the guides and marketing, so a change is made once. The same structured data can answer new questions, such as which days climb the most.
Built: An itinerary editor with maps, daily miles and climbing.
How: Upload a GPS track → miles and climbing fill in → staff edit each day → approve → it’s on the website structured data · approval

Tour map with a route drawn and trip totals for miles, climbing, descent and hours

List of GPS tracks for a tour, each with buttons to show on the map or put on the website

Screens from the working system

The truck checks out before you clock in

Problem: Vehicle checks depended on memory and reminders.
Why: A check tied to clocking in becomes a record, not a habit. A record of defects lets the office see a pattern before it becomes a breakdown.
Built: The inspection is part of clocking in, so it happens at the same moment every day.
How: Clock in → inspection → all good, or a defect report → manager notified → truck held until cleared status · exception

UncheckedCheckedDefect?ClearedOut on tripReturned

Schedules straight from bookings

Problem: Every week, trips were retyped from the booking system into the schedule.
Why: Bookings are the demand and the schedule is the response. Connecting them removes the retyping, and a change to a booking reaches the guides without a phone call.
Built: Guide scheduling reads trips directly from the reservation system.
How: New booking → trip appears on the schedule → guides assigned → changes reach guides integration · trigger

Trip reports while it’s fresh

Problem: Trip notes arrived days later, if they arrived at all.
Why: Notes written while a trip is fresh, in the same place as the trip, add up over time. The office can see what goes wrong, and what guests ask, across many trips instead of one.
Built: A short trip report as part of clocking out.
How: Clock out → a few required questions → notes reach the office the same day required field

More ideas, same principle

There’s so much more we can do than we knew.

Your history, finally usable

Six years of roasting logs pulled out of the roaster’s software and into one place. Now they can predict what to roast and when.

data extraction · forecasting

Eight years of sales you can talk to

Every sale from the point of sale, stored in a place you own. Ask it questions in plain English instead of digging through reports.

data ownership · AI over your data

The mail opens itself

Scanned letters and bills are read, sorted and turned into draft entries for the books. A person still approves each one.

OCR · human in the loop

A phone assistant

Calls a business for you, gets through the phone tree, waits on hold and books within limits you set. It always says it’s your assistant.

voice agent · guardrails

New brains for an old machine

A production machine kept running after its maker stopped supporting the computer that controls it. Watch first, change nothing, then replace.

reverse engineering · safety first

Recipes that know their cost

Supplier invoices feed ingredient prices, so every recipe shows what it really costs this week.

structured data · single source of truth

Waivers before anyone arrives

Guests sign on their phone ahead of time, and the signed copy is filed with their booking.

e-signature · integration

An inbox that sorts itself

Notices and newsletters filed away, real questions from real people raised to the top.

rules + AI · triage

The honest part

What I’d tell you before you start.

The first project took a year. Early AI tools fell short, and so did a freelancer. I’ll tell you what I learned.

Since December 2024, what a person without a coding background can build has changed enormously. The tools are no longer the main thing holding you back.

That does not remove the need for a mental framework. You still have to direct a developer or an AI. Programming organizes data, runs on hardware, and has to be kept secure. You have to know how to ask for each of those, and the tools won’t do that part for you.

The real skill is describing your approach clearly. The tools change every month. That skill does not.

Two limits on this page. The charts and the staff app screen use sample data, and each is labeled that way. And I can’t promise you a result. What you get depends on your data and how clearly you ask.