The Local FDE Playbook
Twenty automations you can sell to local businesses right now.
Twenty AI automations a solo builder can sell to local businesses, modeled on the forward deployed engineer play that Anthropic and Blackstone just funded at $1.5 billion with Ode. Every item names a real business result, a marketplace price, or a published workflow, because a screenshot of a template proves less than a plumber's revenue number. Five are easy enough to start this week. Voice agents pay more and need more testing. I run automations like these inside my own company. The window is the 12 to 24 months before the incumbent software ships this natively.
Read time: 11 min
What do you sell when the AI labs just told everyone where the money is?
Because that's what happened this month. On July 15, 2026, Anthropic and Blackstone launched Ode with Anthropic, a $1.5 billion venture that embeds engineers inside large enterprises to build AI systems against their real data, staying until the systems work. Hellman and Friedman is a named partner, Goldman Sachs is among the backers, and it fields around 100 engineers built on the Fractional AI team. OpenAI launched its own equivalent, The Deployment Company, in May.
The investor thesis is public. Sequoia's Julien Bek published "Services: The New Software" on March 5, 2026, arguing that the next trillion dollar company will be a software company masquerading as a services firm, on the logic that for every dollar spent on software, six are spent on services. Andreessen Horowitz's Alex Rampell made the blunter version in October 2025: global software is roughly a $300 billion market, and US labor alone is $13 trillion a year. Selling the work pays on a scale that selling the tool never reached.
Here's what they left on the table. Ode and The Deployment Company serve enterprises and the mid-market, the clients who can afford one of a hundred elite engineers. Every dental office, HVAC company, law firm, and corner restaurant needs the same work done, and nobody with a billion dollars is coming for them.
The job title is forward deployed engineer, coined at Palantir, where FDEs embedded with clients and built against live data. The barrier used to be the engineering team standing behind that person. That barrier is gone. The solo version is one person, an LLM, a workflow tool like n8n or Make, and the business's existing calendar, inbox, and spreadsheet. Most of what follows is a weekend.
The market, honestly
Two numbers frame this, and they disagree by an order of magnitude for a reason. Self-reported surveys say adoption is nearly universal, with the US Chamber of Commerce claiming 89 percent of small businesses use AI in some capacity. The rigorous figure is far smaller: the SBA Office of Advocacy, working from US Census data, puts production AI use at 8.8 percent of small firms as of August 2025, and JPMorgan Chase transaction data shows 17.7 percent actually paying for an AI tool, at a median of roughly $28 a month.
Read those together and the picture is clear. Everyone is experimenting, almost nobody has AI running operations, and the people who could install it for them barely exist. That's the gap. It also means your buyer has probably never paid real money for software like this, which is why every price below anchors to a salary or a countable leak rather than to technology.
How to read the prices in this playbook
Three kinds of numbers appear below, and confusing them is how people talk themselves out of a good business.
Software prices are what the underlying tools cost you. Template prices are what a packaged version of the workflow sells for on a marketplace, which proves the workflow is buildable and that someone wants it. Managed service prices are what a business actually pays for a working outcome, and those are the ones that matter to your invoice.
The gap between the second and third is your business. A $45 template is a parts list. What a client pays for is the configured workflow, their business rules, the integrations into their actual tools, the testing, the monitoring, and someone answering the phone when it breaks. Every price here was checked on July 30, 2026, and voice rates in particular move fast, so re-verify before you quote anyone. Marketplace figures listed in other currencies are converted to approximate US dollars.
Start with these five
If you want the shortest path to a first paying client, pick one of these: review responses, appointment reminders, lead follow-up, invoice chasing, or support inbox triage. Each one can start as a small workflow with a human approval step, which means your first version is safe to ship before it's perfect. Voice agents command more money, and they need considerably more testing before you put one in front of customers.
The menu, by industry
Home services and trades
- Missed-call receptionist. The agent answers after hours, collects job type and location, handles common questions, and books or routes the lead. Build it on Goodcall or Retell plus a calendar and a CRM or Google Sheet. Bye Junk reported a 25 percent monthly revenue increase and more than $2,500 in bookings within 30 days of running Goodcall alongside Yelp. Goodcall lists $79, $129, and $249 per agent per month, cheaper annually, and its pricing page says agencies can give clients dashboard access.
- Quote and estimate follow-up. When an estimate goes out, the workflow follows up by email or SMS, answers basic objections from an approved knowledge base, and alerts the owner on a reply. Stack is Gmail or Outlook, Sheets, an LLM, and n8n. A packaged version listed on Gumroad at around $15, which tells you the build is trivial and the service is everything around it.
- Lead scoring and instant routing. Score every website, ad, and referral lead on service area, job value, urgency, and fit, then push high-value leads straight to the owner and drop the rest into nurture. A five-star template listed at around $19, and n8n publishes a community workflow that qualifies leads from Sheets with an LLM and alerts Slack.
Salons, spas, gyms, and anything appointment-based
- WhatsApp booking bot. Customers ask about services, check availability, book, reschedule, and get reminders without waiting on staff. WhatsApp plus a calendar plus n8n plus an LLM for intent. A template listed at $39. Brow Arc reported a 6x return using Goodcall across multiple locations.
- Appointment reminders and no-show recovery. Personalized reminder the day before, confirmation collected, follow-up on anyone who doesn't respond, staff alerted. Start with plain email or SMS before adding any AI. A clinic no-show system listed at around $103, with a published stack of n8n, Sheets, Gmail, Groq, and Telegram. For health businesses, keep patient data out of general-purpose tools unless the entire stack is approved for the privacy rules that apply.
The rest of this playbook is below.
Enter your email to read it. You also get one new playbook every week.
One email a week. Unsubscribe any time.
- Dormant-customer reactivation. Pull everyone who hasn't booked in 60, 90, or 180 days, generate a relevant offer, run a short sequence, stop automatically on a reply or booking. The email version is the easy first build; the voice version runs on Retell or HighLevel. HighLevel reports that Billy Gene is Marketing made roughly 800 calls, generated 190 leads, and attributed $40,000 in sales to its Voice AI. Treat that as one named result, not a benchmark you can promise.
Law firms, accountants, and professional offices
- Intake and call screening. Ask the practice-area questions, flag conflicts and urgency, collect contact details, book the consultation, and write the summary into the intake system. Maria Jones Law Firm credits Smith.ai with fewer missed calls and faster intake. Smith.ai lists AI Receptionist Pro at $150 a month and a full-service Enterprise tier at $500 a month that includes custom prompting, integrations, setup, and ongoing optimization. That $500 line is the most useful managed-service benchmark on this page, because it tells you what a business will pay for the maintained version.
- Reception and qualification for tax firms, consultants, and physical therapy. Identical core system, different questions, routing, and calendar. Smith.ai names The Tax Chick.PRO and Native Physical Therapy on its pricing page and says more than 5,000 businesses use it, across legal, home services, medical and wellness, real estate, financial services, and retail.
- Invoice and receipt extraction. Watch an inbox or folder, read invoices and receipts, pull supplier, amount, due date, tax, and line items, then write to Sheets, Airtable, or accounting software for approval. Templates listed at $45 for AI-powered invoice processing and around $76 for Stripe-to-QuickBooks automation. This is an excellent first project precisely because a human approves everything before it touches the ledger. Anchor the price to the $49,210 median bookkeeping clerk salary from BLS.
- Invoice follow-up and collections. Detect overdue invoices, draft progressively firmer reminders, send on a schedule, stop on payment, produce a weekly aging summary. A five-star template listed at around $103. Keep disputes and anything delicate with a human.
Real estate, property, and automotive
- Real-estate lead manager. Capture portal and website leads, qualify budget, area, timeline, and financing, then book a viewing or nurture. A five-star system listed at $270, the highest template price on this list, which is itself a signal about willingness to pay in this vertical. HighLevel ships pre-built real-estate snapshots.
- Voice follow-up for dealerships and high-ticket sellers. Call new leads or an old inquiry list, ask a short qualifying set, book the next step, save the transcript. There's a published n8n workflow for outbound calls from Sheets via Retell. Retell lists $0.07 to $0.31 per minute. Vapi cites Kavak using voice agents for inbound support and outbound reminders. Read section 6 before selling any outbound calling.
Restaurants, cafés, and food businesses
- Phone and catering intake. Answer hours and menu questions, collect catering details, take callback requests, route anything unusual to staff. Midnight Cookie Co. runs Goodcall for support, operations, and catering inquiries. Goodcall's entry plan is $79 a month, or $66 billed annually.
- Newsletter and offer generator. Turn weekly specials, events, and photos into a short email, a draft for approval, and a scheduled campaign. A five-star template listed at around $13. One of the easiest offers on this list, because version one can stop at a draft a human reviews.
Retail and e-commerce
- Abandoned-cart recovery. Detect the abandoned cart, draft a product-aware sequence, send, stop on purchase. A template with two five-star ratings listed at around $70. Use the store's approved discount rules rather than letting a model invent offers.
- Support inbox assistant. Label incoming email, retrieve the relevant policy or order data, draft the reply, escalate refunds, complaints, and anything low-confidence. Listed at $37 and up. n8n publishes a basic Gmail labeling workflow. Run it in draft-only mode until it earns trust.
- FAQ and document chatbot. Load approved documents, answer with citations, hand off anything uncertain. n8n features a named RAG chatbot for company documents built on Drive and Gemini. n8n Cloud starts at about €20 a month billed annually, and the Community Edition self-hosts free.
Marketing agencies and local service brands
- Google Business Profile review responses. Pull new reviews, draft a reply in the owner's voice, flag one- and two-star reviews, publish only after approval. A multi-location system listed at $45. HighLevel markets automated review responses as a reputation feature, which is your warning that this one gets absorbed early.
- Social content queue. Turn one offer, customer story, or weekly update into several channel-specific drafts, route for approval, schedule what's accepted. A full social-media automation system listed at $49, with a narrower Instagram comment auto-response plus an hour of setup help listed around $104. Keep public posting behind approval until it's boring.
- Client onboarding and meeting-to-CRM. After a sale: create the folder, send the welcome, collect missing details, book kickoff, summarize the meeting, create the follow-up tasks. Templates include a $59 client-onboarding concierge and a meeting AI agent listed for a few dollars. Every CRM is shipping meeting summaries, so the summary is not what you're selling. What you're selling is the deletion of all the copying between email, forms, calendar, folders, and CRM.
What the tool economics actually look like
Workflow engine: n8n Cloud from about €20 a month annually, Community Edition self-hosted, and a public library that held 11,043 workflows when checked. Voice: Retell $0.07 to $0.31 a minute, Vapi about $0.05 a minute hosting plus model costs, and Goodcall selling the finished small-business product from $79 a month.
Agency resale, which is the part that decides whether this scales: HighLevel lists AI Employee plans at $50 and $97 per client sub-account, says most add-ons can be resold at your own markup, and its $497 a month plan supports a branded SaaS mode. Goodcall gives agencies client dashboard access. Those are explicit signals that the reseller path exists rather than something you have to invent.
Template market, for calibration: Gumroad showed 2,654 products for n8n and 703 for Make.com automation, ranging from a few dollars to a few hundred. A listing is weaker evidence than a named business result, so weight the ratings, the recorded sales, and the end-business case studies above the price tag.
What to refuse
Unconsented outbound AI voice calling, first and always. The FCC ruled on February 8, 2024 that AI-generated voices count as artificial or prerecorded under the TCPA, which means prior express consent, written for marketing, with statutory damages of $500 to $1,500 per call and no cap. One campaign against a purchased list can end a solo operator. Inbound, where the customer calls the business, is a different and far safer thing.
Also skip: anything requiring custom model training, generic website chatbots, competing with the mature ambient medical scribing vendors, and reselling n8n itself as a hosted product, which its Sustainable Use License forbids. Building client workflows on n8n and charging for that work is explicitly permitted. Hosting n8n and charging users for access is not.
Healthcare work needs a signed BAA. Anthropic offers one on Enterprise and the direct API and explicitly not on Pro, Max, or Team plans, and consumer chat products are never acceptable for patient data. Law firm work has to stay clear of unauthorized practice of law, with a lawyer in the loop on anything substantive. This section is a summary, not legal advice, and the rules keep moving.
And know who eats what. ServiceTitan and FieldEdge are adding native voice, Weave covers dental recall, Clio ships law firm intake AI, and HighLevel already markets review responses. Assume any single-feature offer gets absorbed within 12 to 24 months. What survives is cross-system glue, the client's own accumulated data, and judgment.
The part most coverage skips: selling it
The guru numbers are wrong, and planning around them will kill you. Templates and courses market cold email reply rates of 5 to 10 percent. Independent data: Belkins' 5.6 million email dataset shows a 0.51 percent US reply rate, Instantly's 2026 benchmark averages 3.43 percent, and GrowthFlare's operator data works out to roughly one booked call per 250 to 500 well-targeted sends. Plan on that number.
What converts, in order. Niche to one industry and name one countable leak, because "a specific workflow for five nearby firms" outsells "AI transformation" every time. Go warm first, since referrals, your own network, and industry groups beat cold outreach for a first client, and 82 percent of very small firms tell the SBA that AI simply isn't applicable to their business, which is an objection that specificity beats and volume doesn't. Give them a demo they can touch, ideally a live phone number answering as their business. And charge for pilots, because free pilots don't get used.
Anchor to the salary or the leak, never to the technology. A receptionist costs $37,230 a year base per BLS, around $45,500 fully loaded, to cover 40 of the 168 hours in a week. That comparison closes. A feature list does not.
The three artifacts you need
One, a live demo agent. A working phone number running your receptionist build that you can re-skin per prospect in minutes. It's your portfolio and your closer.
Two, a one-page price sheet. Each offer on a line with the salary or leak it addresses, your monthly price, and the arithmetic between them.
Three, a written pilot proposal template, 300 to 500 words, and almost nobody has one. What you're installing, what it replaces, what it costs, the single 30-day success metric, and what happens when the pilot ends. That success metric line is what converts a pilot into a retainer, so write it once and reuse it forever.
Choosing your first offer, and compounding after it
Five steps, in order. Pick one industry you can speak to credibly. Pick one measurable leak inside it, because missed calls, unconfirmed appointments, stale estimates, overdue invoices, and unanswered reviews are all easy to count and therefore easy to sell against. Build the smallest useful version using a form, an inbox, a spreadsheet, a calendar, and one model call, with no dashboard. Keep a human approval step, starting with drafts and alerts and automating the sending only once the output is stable. Then sell the maintained outcome, because the template is the cheap part and configuration, business rules, integrations, testing, monitoring, and monthly improvement are the service.
The compounding comes from the checklist you write after each install: the questions you asked, the integration that broke, the setting that worked. First client takes a weekend, tenth takes two days. That checklist is also the closest thing to a moat you get, because a prompt is portable but your accumulated knowledge of how a dental office in your city actually runs is not.
Two benchmarks tell you whether to keep going. If you can't close one in roughly five qualified demos, the niche or the offer is wrong, so narrow before scaling outreach. And hold 70 percent gross margin with under 10 percent monthly churn, because worse than that means you're selling a feature the platform is about to ship for free.
Honest limits
Churn is structural, and pretending otherwise is how most automation agencies die. SaaStr's Jason Lemkin documented clients migrating off an agency by copy-pasting the prompt. ChartMogul's analysis of roughly 3,500 companies found AI-native products retaining worse than consumer SaaS. The AI BDR category runs 50 to 70 percent annual churn, and Gartner projects that more than 40 percent of agentic AI projects get abandoned by the end of 2027. The defense is the same as the moat: multi-workflow retainers, the client's own data, monitoring they can't copy-paste.
These systems need babysitting. One public case saw a support agent go from solving 95 percent of issues to failing completely after a silent model change, and the Cursor incident, where a support bot invented a policy and triggered cancellations, is the standing warning for anything customer-facing. That is precisely why the retainer exists. Price it in rather than apologizing for it.
The marketplace prices in section 4 prove buildability and demand. They do not prove that any particular seller earns a living, and the named case studies are single results reported by the vendors involved. Use them to show a prospect the workflow is real, not to promise them a number.
And the honest read on the people selling this dream: analysis of the largest agency guru's revenue suggests most of it comes from teaching the model rather than running the agency. The market is real. The easy version of it is not.
What this changes
For twenty years, selling software to a small business meant building a product and hoping thousands of them bought it. What Ode proved at the top of the market, and what a plumber's missed call proved at the bottom, is that the better trade right now is one person installing working systems inside one business at a time, paid like the labor they replace instead of like the software they run on. The tools finally made that a job one person can do. The labs took the Fortune 500 and left everything else, and everything else was always the longer list.
