FieldFlow

FieldFlow is a field service management platform for small trades businesses of one to five people. It covers the whole job cycle (estimates, scheduling, invoicing, payments) plus an AI assistant, for a flat $19 a month with the whole team included. I founded the company and I am its only designer and its only engineer, which means every screen, flow, and word in the product is mine, along with the pricing and positioning decisions the design has to serve. I also run my own work through it, which is the fastest research loop I've had: every rough edge shows up in my own week.

The big platforms compete on breadth. Modules stack on modules, add-ons multiply, pricing runs per seat and per feature, and the product ends up needing a sales call to buy and an onboarding program to learn. For most small trades businesses, that means paying for software they mostly don't use and working around a UX that was designed for the demo, not day to day operations.

FieldFlow is built on the opposite premise: everything you need, nothing you don't. It covers what a service business actually does every day (estimates, scheduling, invoicing, payments, getting paid on time) and spends its entire design budget making those feel modern and effortless. That standard holds wherever the work happens, for an office manager running the day from a desk or a solo operator invoicing from the truck between jobs.

The price makes the same argument the product does: $19 a month, flat, whole team included. Not a suite, not a bundle of modules you'll never open. The essentials, nailed, at a price that isn't a decision.

Field service software gets built by addition: every release answers the question of what else the product could do, and that's how the incumbents ended up bloated. FieldFlow's feature set is mostly a record of what got cut. The filter: does this belong to the loop a service business runs every day to get paid? Quote the work, schedule it, do it, invoice it, collect, and nudge the people who haven't paid. Anything on that loop had to be excellent; anything off it didn't ship.

Here's where we landed:

• Estimates. Built in seconds from a reusable price book, then converted to an invoice or a scheduled job in one click. Customers sign on their phone, and every signed document is snapshotted and hashed for tamper evidence.
• Invoicing. A live PDF preview that matches exactly what the customer receives, with selectable styles, custom numbering, and live totals. Credit notes, refunds, and voids all leave an audit trail, and QuickBooks stays in sync.
• Scheduling. Month, week, and day views with drag-to-create and drag-to-reschedule, a full recurrence builder for repeating work, and one-tap timers for time on the job.
• Payments. Customers pay the way they actually pay: card through Stripe, a QR code for Venmo, Zelle, PayPal, or Apple Cash, or a check with mail-to details. Per-channel delivery feedback means a bounced email never masquerades as a sent one.
• Reminders. Separate cadences for unpaid invoices, upcoming appointments, and open estimates, each with editable message templates and a log of everything sent.
• Expenses & reporting. The one deliberate exception to that filter. Expenses with receipt photos and categories, a mileage log, and reports that total income, expenses, and net profit for any period, with CSV export. Most small trades businesses use a sliver of what QuickBooks does, so covering that sliver well lets them drop the subscription rather than add one.
• AI. Flow, the built-in assistant, runs ninety real tools, drafts while leaving consequential sends to a human click, and flags what needs attention, like a completed job with no invoice.

The AI assistant can be massively helpful, but plenty of trades people are skeptical of AI. So Flow is optional: a master toggle removes it from the interface entirely, and every one of its write tools mirrors an action in the regular UI, so nothing important hides behind it.

For the people who do use it, Flow is usually faster than the UI itself. "Invoice the Hendersons for the water heater job and send it" is one sentence, versus a navigation, a form, and a send dialog. On web it also knows what you're looking at, so "send it" means the invoice already on screen. Ninety tools create invoices, start timers, find open slots, and record payments, returning results as cards with deep links, and the model drafts while consequential sends stay a human click.

A proactive triggers engine watches for what a good office manager would notice, like a completed job with no invoice or an accepted estimate never scheduled, and surfaces the top nudge as a one-click prompt. Underneath, requests start on a fast, cheap model and escalate only when a write is attempted, which is what makes an AI employee viable at flat $19 pricing.

A trades business gets judged in about ten seconds, on the customer's phone: the estimate opens, and the customer decides whether this outfit seems professional, whether to say yes, and when to pay. So every customer-facing object in FieldFlow (estimates, invoices, and jobs) has a branded public page opened from a tokenized link, no account or app required.

Saying yes takes about ten seconds: email, a 6-digit code auto-filled from SMS, then sign by drawing or typing. Behind it, the signed document is snapshotted to PDF and SHA-256 hashed into the operator's audit trail, so a disputed signature has tamper evidence. Invoices accept Stripe, a QR code for Venmo, Zelle, PayPal, or Apple Cash, or a check, and afterward a "How did we do?" page routes happy customers to the business's review platforms, turning finished work into the next job.

Underneath is a design system with an unusual constraint: the mobile app, the web app, and the generated PDFs all had to look and behave like one product. A single shared color package feeds all three surfaces, Space Grotesk runs everywhere, and 26 web primitives plus a full mobile kit sit on the same tokens.

It goes deeper than visuals: 225 web files import a shared package of totals math, validation schemas, permissions, and formatters, so an invoice can't total differently on two screens. The system is also role-adaptive, giving operators the full app while crew and subs get a focused three-tab mobile experience backed by a permission matrix that mirrors the database security rules.

I designed FieldFlow and I built it, alone. The style was set in Figma, a handful of screens that fixed the type, the color, and the density, and from there the product was built in Claude Code. The mobile apps are Expo, the web app is Next.js, and both sit on Supabase, with Stripe for payments, Twilio for the messages that go to customers, and an Astro marketing site in front.

Flow, the assistant, is where the build got hard. Every one of its write tools mirrors an action in the regular interface, so the shape of each response has to be right before anything is sent, and a model doesn't return the same shape twice. QA was using it: running my own jobs through it, logging every response that came back wrong, and fixing the tool definition or the prompt until the log stayed empty. Building solo on a $19 price also meant the AI cost architecture, the cheap model first and escalation only on a write, was designed rather than defaulted.