AI implementation for small and mid-sized businesses
We configure ChatGPT or Claude around your company’s information, train your team, and automate specific repetitive workflows — using the tools you already pay for.
For small and mid-sized businesses. Based in Tacoma. Working nationwide.
01 Work in
02 Your configured setup
ChatGPT or Claude workspace + workflow
03 Useful output
01What changes
ChatGPT or Claude set up with your service terms, procedures and reference material — the versions your business has approved — plus instructions for tone, limits and when to say “I don’t know.”
What employees will be able to do
Ask a question about how the company works and get an answer that points to the source, or draft a customer reply that starts from your policy instead of a guess.
Two live sessions built around your three chosen tasks, a recorded walkthrough and a short guide. The goal is a team member completing the task without someone coaching every step.
What process will change
Recurring drafting and look-up work moves into a shared, tested setup instead of living in one person’s head or in scattered personal chat histories.
Where a process repeats — an inquiry, a booking, a standard document — a defined workflow creates the record, notifies the owner and prepares the draft, with alerts when something fails.
What stays under human control
Anything consequential or unclear. Missing information goes to a person rather than being guessed, and external messages beyond pre-approved text are approved by someone on your team.
02Workflow walkthrough
Illustrative workflow — sample data. It runs in your browser, connects to no tools, calls no AI service and sends nothing.
A website form submission arrives.
Form submission · sample
Design noteOne defined trigger. The workflow does not watch inboxes or guess which emails count as inquiries.
Routing rules and an approved acknowledgment are the only references used.
Routing table · approved
Design noteThe business approves the owner for each service line and the exact acknowledgment wording before anything is built.
Check for an existing contact, create the record, classify the request.
CRM record · created
Design noteOrdinary rules handle the record and duplicate check. AI is used only to read the free-text request and suggest a service line.
Not needed for this standard case; required for anything outside the rules.
Review rule
Design noteA clear request with a matched rule can proceed. Pricing questions, complaints or unclear requests go to a person.
Owner notified; approved acknowledgment sent; record ready.
Result
Design noteThe acknowledgment is the pre-approved text, not a freshly generated reply. Failures raise an alert instead of disappearing.
Select a stage, or walk through the example, to read why each step is designed the way it is.
03Engagements and prices
| Engagement | Price | What you get | Start here when… |
|---|---|---|---|
| AI Workspace Setup & Team TrainingOne-time setup | $2,500per standard implementation | ChatGPT or Claude configured around your approved material, three bounded uses, two training sessions and a 30-day stabilization period. | Your team keeps asking the same questions or drafting the same kinds of documents. |
| Workflow Automation SprintOne-time project | $2,500–$5,000per standard scoped workflow | One defined process automated across up to three existing applications, with alerts, an exception path and a pause switch. | A handoff repeats every week and the steps are already clear enough to write down. |
| Workspace CareMonthly, optional | $350per month | A monthly check-in and up to one hour of small updates or training for a workspace we delivered. | You want someone keeping the setup current after the stabilization period. |
| Automation CareMonthly, optional | $750per month | Failure notifications reviewed in business hours, periodic checks and minor maintenance for up to three delivered workflows. | Workflows we built are in daily use and you want them watched. |
Not sure which fits? That is what the assessment is for. Every engagement has written caps — see exactly what is and isn’t included on the pricing page.
Full pricing and scope limits04Evidence of practical building
Podcast, website and guest operations
An interview podcast for active real-estate investors, hosted by founder Gabe Petersen since 2020, with a website that organizes hundreds of episodes by topic and turns guest booking into a single self-serve step.
What it showsRunning a recurring production operation — guests, scheduling, publishing and an archive — and building the website that holds it together.
Web calculator and iPhone app
A fee calculator that helps mobile notaries quote a job by separating state-capped notarial fees from travel and service fees, with state fee references and loan-signing, travel and profit calculators.
Two iPhone apps from real-estate operations
Two published iPhone apps shaped by the founder’s real-estate operating background: a self-storage underwriting calculator and an offline logbook for water-system readings at mobile home parks, RV parks and campgrounds.
Interactive explanation on this site
Three common administrative workflows — inquiry routing, booking handoff, and document draft and approval — shown stage by stage with sample data, including what happens when information is missing.
9 iPhone apps published under the founder’s developer account — calculators, logbooks and planners.
See the app catalogueApps and owned-business projects demonstrate building and operating experience. They are not presented as client results, paid adoption or measured savings.
05 How an engagement runs
A first conversation about fit and one candidate team or workflow.
A written scope with tasks, sources, applications, caps, exclusions and price.
You approve the scope
Set up in accounts your company owns, with named users and limited permissions.
Representative cases, including missing, conflicting and out-of-scope information.
You review the test results
Training sessions, a recorded walkthrough, a short guide and named ownership.
Handover accepted
Bounded monthly support if you want it. Never required.
06 Specialty
Our founder owns and operates self-storage facilities and mobile home and RV parks, and has spent years interviewing investors and lenders. That shows up as practical judgment about where administration can be simplified — and where decisions must stay with people.
Same packages, same published prices. No automated screening, credit or eviction decisions, ever.
Real-estate operationsPotential scoped applications we can assess — not plug-and-play integrations.
07 Who you work with
Fieldwork was founded by Gabe Petersen, who runs real-estate businesses, hosts The Real Estate Investing Club podcast, builds and ships software, and previously worked in management consulting.
The point of view is simple: AI earns its place when it removes a specific, repeated piece of work, and the people doing that work understand it. Everything else is a demo.
The prompts are the smallest part. The work is choosing three tasks that are worth doing well, organizing the approved material the assistant should rely on, writing instructions that say what to do when information is missing, testing it against your real examples, and teaching your team when to trust it and when to check.
A workspace that knows your service terms, your tone and your escalation rules behaves very differently from a blank chat window.
No. We configure an existing assistant — ChatGPT or Claude — with project instructions and approved reference material. That is not training a model, and we do not build custom models.
Whether the vendor may use your business data to improve its models depends on your plan and settings; we review those settings with you rather than assume them.
Into the third-party tools you choose, under accounts your company owns. The assistant and any automation platform are hosted services, so we never claim your data stays on your own servers.
We keep the first engagement to everyday business material, use named users and limited permissions, and check both the original folder permissions and who can see the AI project — a restricted document uploaded to a widely shared project is no longer restricted.
No. Setup and workflow projects are one-time fees, and everything runs in your accounts. Care plans start at $350/month and are optional — useful if you want someone checking and adjusting things, never required to keep access.
No. The $2,500 one-time setup covers one team or department, one platform and up to 5 people. Many mid-sized companies start with the team that has the clearest repetitive work, then decide what to extend.
Sometimes a bounded assistant or an approved tool-using step is the right answer, and we will say so when it is. But most administrative work is better served by a predictable workflow with a person approving anything consequential.
We sell a specific outcome with a stated scope — not open-ended autonomy. If “agent” comes up, we will spell out exactly what it can see, what it can do and who approves it.
Every workflow we build includes error alerts, a documented exception path, duplicate prevention and a way to pause it. Failures should be noisy, not silent. You also get a handover so your team knows how to check and pause it.
No. We work with small and mid-sized businesses across industries. Real-estate operations is a specialty because of our founder’s operating background, not a limit on who we serve.
Next step
The assessment is a short first conversation about fit: what your team repeats, what tools you already use, and whether a workspace setup or a single workflow makes sense first. It is useful whether you want help with ChatGPT or Claude, or have one specific automation in mind.