AI-Powered Field Service Management Platform
A cloud platform with a built-in AI agent that runs a service company's scheduling, estimates, invoicing, payments and reporting, so crews can focus on the real work
At a glance
I acted as design lead and executive sponsor for an AI-powered field service management platform for a growing home services company. I ran discovery with the owner and crew, researched the market, designed every screen and the AI agent inside it, and used Claude (AI) to help build the platform and its working prototype.
The challenge
A growing handyman business was pricing jobs on paper and juggling texts, receipts and spreadsheets. Estimates were slow and inconsistent, approvals got lost, and the owner had no view of the pipeline. Off-the-shelf software put the features they needed behind expensive tiers.
What I led
- Set the product vision: one cloud platform from the first call to paid, built for crews in the field
- Ran discovery with the owner and crew, and compared Jobber and Housecall Pro pricing and plan limits
- Designed an AI agent into the workflow to draft estimates from photos, check pricing and read receipts, with people always in control
- Turned the owner’s price sheet into a 90-job price book
- Designed every screen phone-first: bottom tab bar, finger-sized controls, works offline
- Architected a serverless AWS back end with signed customer links and private photo storage
- Built reporting on pipeline, close rate, sales by crew member and why jobs are won or lost
- Built a working prototype of the estimate, approval, work-order and manager flows
- Shipped 49 versions with written release notes, known limits and a roadmap
- Led V2: personal sign-in with MFA, user management, leave-by times, crew training, demo mode and a fairer overdue rule
The outcome
The crew now prices jobs on site in minutes while the AI agent drafts line items, checks pricing and files receipts. Customers approve from their phones, and every sale automatically becomes an invoice, a schedule entry, a material list and a job file. Managers see the pipeline, close rate, sales by crew member and why jobs are won or lost in real time, on a platform that costs about a dollar a month to run. V2 gives every person a secure sign-in, tells crew when to leave, trains new crew on day one and only flags invoices overdue after the work is done.
Overview
Service companies run on paperwork: estimates, schedules, invoices, receipts, texts and spreadsheets that pull crews away from the work customers pay for. I designed and built an AI-powered field service management platform to be the backbone of a growing home services business, with an AI agent inside it that takes on the boring back-office tasks. It started as a single page that priced a customer’s to-do list. Forty-nine versions later it is a full cloud platform on the company’s own AWS account.
How I led it: I acted as design lead and executive sponsor. I set the vision, ran discovery with the owner and crew, researched the market, designed every screen, product-managed each release and used Claude (AI) to help build the platform and its working prototype.
What it does:
AI agent: drafts estimate lines from photos and notes, checks pricing, reads receipts and handles routine paperwork, with a person approving every change
Estimates and approvals: about 90 priced jobs, one-tap customer approval with a signed link, and automatic invoices
Work tracking: one board from booked visit to paid that shows who did what and when
Scheduling and communication: online booking, schedules by crew member, Navigate, “On my way” and day-of emails
Receipts and materials: receipt scanning with text reading, stock, audits and paid vs billed on every job
Reporting: pipeline, close rate, sales by crew member, what sells best and why jobs are won or lost
What’s new in V2: V2 makes the platform safe and simple for a growing crew. Once more people used it every day, a shared password, guessed drive times, no onboarding and early overdue flags got in the way, so V2 fixes each one:
Personal sign-in with MFA: everyone has their own account with an authenticator or emailed code, backup codes and a trusted phone for 30 days, so the owner knows who did what and can remove one person’s access
Users as flip cards: managers add, edit, pause or delete people and send setup or reset links without help; the shared crew password is switched off
Leave-by times: crew see when to leave to arrive 10 minutes early, counting the drive from the last job; customers never see it
“Your work” home and first-day training: crew open to today’s visits, and new crew get six short steps once, with Skip
Fair overdue rule: an invoice turns overdue only 5 days after the job is marked done, not before the work is finished
Demo login and clean data: a demo account with its own sample data for sales and training, and test records removed from the live system after a full backup
Services
Customer & Market Research, AI Agent Design, Product Strategy, Product Management, UX/UI Design, Interactive Prototyping, AI-Assisted Development, Cloud Architecture (AWS)
Client name
HoneyDo Crew Home Services
Team
John Karl, Design Lead & Executive Sponsor; the client’s owner and field crew
Prototype
Market research
User research
UX Decision
Go-to-market
This is a working prototype. Follow the guided steps or switch tabs and use the controls. Everything responds live.
Market research
Goal:
Learn where a small crew loses time and money, and why they hadn’t bought software already.
Jobber: plans start around $39 a month for one user and about $169 a month for a five-person team, with marketing and AI add-ons that raise the bill quickly
Housecall Pro: $59 a month for one user and $149 for five, but the price book needs a higher tier and reporting is reserved for the top plan
The gap: the features a small crew needs most, a price book, reporting and help with paperwork, sit behind the most expensive tiers
User research
Goal:
Ground every decision in real input from the people who use it, and say plainly what still needs validating.
The owner and crew (real input): jobs were priced on paper, texts, receipts and spreadsheets were scattered, approvals got lost and the owner had no view of the pipeline
Proto-personas: the owner who needs the numbers, the crew lead quoting in a driveway and the homeowner who wants a clear price fast
Research method: ride-alongs and release-by-release testing with the crew, tracked against time-to-estimate and approval rate
Why Each UX Decision Was Made
Goal:
A crew member should price a job on site in minutes, with one hand, and never enter the same thing twice.
The agent drafts, a person decides: AI speeds up estimates, but nothing is added or sent without approval, which builds the crew’s trust
Ask only what changes the price: the agent asks one follow-up question instead of a long form
The owner’s price sheet became the price book: about 90 jobs, so every crew member quotes the same way
Phone-first: a bottom tab bar, finger-sized controls and offline support for driveways and basements
One record, concept to cash: customer approval creates the invoice, material list and job file automatically
Cost the owner can see: daily AI limits and a serverless back end keep running costs around a dollar a month
V2 decisions
Goal:
Add security and structure for more users without adding steps to a crew member’s day.
Security that doesn’t slow the field: authenticator codes with an email backup and a 30-day trusted phone, because crew sign in with gloves on, in driveways
Settings live on the person’s card: flip the card at the corner instead of opening a separate admin area, so managers manage people where they see them
Guards against self-lockout: a manager can’t pause or demote themselves, and the shared password only turns off once personal sign-in works
Show the time to leave, not a map: crew need one number, with 10 minutes of buffer, and only crew see it
Train once, then get out of the way: six steps, always skippable, remembered on the server so a new phone doesn’t show it again
Overdue means the customer is late, not the crew: the 5-day clock starts when the job is marked done
V2 release and rollout
Goal:
Ship V2 to a working crew with no downtime and no lost data, and give the owner a safe way to demo and sell it.
Release: app versions 60–63 and server versions 35–37, each deployed with a backup first and a written rollback, with automated tests run before every release
Release notes: written for the owner and crew in plain language: what changed, why and what it fixes
Onboarding: managers send setup links; each person sets up their own password and security code in a few minutes
Demo for sales and training: a separate demo login with sample data and a Manager / Crew switch, so live customer data never appears in a demo
Clean data: 12 test records removed from the live system after a full backup; real customers, receipts and scans kept
Next: measure on-time arrivals and time-to-estimate over a full season, then open the platform to more crews
AI Front Door, Rollout and Executive View
Goal:
Let customers serve themselves, roll out without disrupting the work, and give leaders a live view of the business.
AI front door: the website chat assistant books visits, takes call backs and verifies customers with an emailed code
Bookings become estimates: online requests become real estimates, assigned to the first free crew member
Rollout: 49 versions shipped with written release notes, known limits and a roadmap, each tested with the crew
Executive view: converted, pipeline, close rate, unpaid invoices, sales by crew member, what sells best and why jobs are won or lost
Security: crew and manager roles, signed customer links, private storage and daily AI limits
One platform: schedule, contacts, invoices, and receipts & stock.
Final observations
This project shows how I lead with AI, both in how I build and in what I build. I started with the crew, studied what the market charges and where it falls short, and made the calls on what the agent should and shouldn’t do. A growing service company went from a paper price sheet to a production cloud platform for about a dollar a month: the kind of software that usually takes a funded team. Next steps are measuring time-to-estimate and close rate over a full season and opening the platform to more crews.
V2 shows the next stage of leading a product: after proving the workflow, I looked at what broke as more people used it every day and fixed those things first, security, timing, onboarding and trust in the numbers, instead of adding features.
John Karl
19 Years • Product & UX Design Leader Leading Mobile Product for 500K+ Users
If you like what you see or have any questions, feel free to send me an email anytime.
Selected work
[2010 -2025]
