Instant quotes for trades

Use a strong model, because this pipeline involves judgment calls and a weak model makes expensive mistakes with real customers and real money.

After that, give it the keys for the job. In chat, tell the agent which accounts it's working with and paste the credentials when it asks: a Google Places API key from Google Cloud Console with the places API enabled and a budget cap on it, a postcard API account with Lob, PostGrid, or Stannp, a Vercel AI Gateway key for Jev, which the next step covers, and a Twilio number for the win-back messages. Once a client signs you'll also get their price list, their customer list, and access to put the quote widget on their site. On Grok bot you can also log into a website once in the bot's browser and it stays logged in.

Jev is a decision model from typesafe. The agent writes the messages, and Jev is the part that decides. You give it a situation, like one quote request with the customer's answers and the business's own price list, and a question with fixed answers, and it gives back one of those answers plus a probability for how sure it is. The agent only acts when Jev is sure and sends everything else to you, which is what makes it safe to let an agent near a business's customers and money.

You get Jev through Vercel AI Gateway. Make a Vercel account, open AI gateway, create an API key, and paste it into the agent. The model name is typesafe-ai/jev, and the agent calls it the same way it would call any other model. It costs almost nothing to run. In my testing, Jev decisions came to about $0.054 per 1,000, and a quote landed on screen in around 800ms.

The agent never invents a price. The owner's own rates go into a table, and Jev's job is to decide which line of that table a request falls under and whether it can be quoted at all. That's what makes this safe to put in front of money, because the number on screen is the owner's number.

The table is simple, one row per job type with the base price, what changes it, and what the job can't cover:

Jev answers four things per request: which job type this is, whether it can be quoted from the table, which modifiers apply, and whether anything in the request means a human should look first, like an emergency, a commercial address, or damage the customer describes. Above 0.8 it quotes, anything under that turns into a callback request with the details already collected, so the owner never loses the lead either way.

The widget sits on the business's own site and asks three or four questions in plain words, like what needs doing, how old the place is, and when they want it. Then it shows the price, a time window, and a button to book, and it sends the owner the booking with everything the customer answered.

Keep the questions short, because every extra question loses people. The customer's phone number comes last, after they've seen the price, so even the ones who don't book leave you a lead.

The second pipeline uses a list the owner already has. Every invoicing or booking tool exports past customers with what they had done and when, and most trades have hundreds sitting there. The agent writes one message per person based on their last job, not a blast, so a customer who had a water heater put in four years ago hears about the flush it's due for, and someone whose ac was serviced last spring hears about this year's service before summer.

The rules matter more than the writing. One message per person, never someone who complained or had work go wrong, nothing to anyone who's been contacted in the last few months, and every message comes from the business with a way to book in one tap.

This fits trades where the job has a repeatable price and customers come back, so plumbers, HVAC, electricians, pest control, garage doors, cleaning companies and landscapers. Filter for a rating of 4.0 or above, at least 10 reviews, a phone listed, and a website, because the widget needs a site to sit on, and put the ones whose site only has a contact form at the top.

The demo has to show both halves, so the agent builds one page for each business with a working quote widget at the top loaded with prices it estimated from their listed services and their area, clearly marked as estimates for them to correct, and underneath it a sample win-back message written for their trade with a line saying how many past customers a business their size usually has sitting unworked.

The owner will play with the widget first, which is what you want, because the moment they type their own job in and see a number come back, the conversation stops being about whether this works.

Postcard APIs like Lob, PostGrid, and Stannp mail real printed cards from an API call for a dollar or two each. The front is a screenshot of the quote widget mid-quote with their business name on it, and the back says "we built [business name] an instant quote. Scan and price a job." With the QR and your phone number. The win-back half stays off the card, because one idea per card converts better and it becomes the thing you add on the call.

What you ask for is their real price list and their customer export. The price list usually doesn't exist as a document, so the fastest route is a ten minute call where the owner says their prices out loud and the agent writes the table from the recording, then sends it back for them to correct.

Then you run the win-back on the first fifty names with every message shown to you before it goes, and you put the widget on their site. Bookings from both are counted the same way, and that first batch of bookings is what you get paid on and what makes the owner keep it running.

After the first batch this runs on a schedule. The win-back works through the rest of the list slowly, in small batches, and then keeps going on its own as customers age into being due again, so someone serviced last spring gets a message next spring. The widget keeps quoting, and anything Jev wasn't sure about lands in a list for the owner with a callback ready to go.

Once a month the agent sends the owner a simple count of bookings from the widget, bookings from the win-back, and the jobs that needed a callback, and that same count is what you invoice your per-booking fee from.

Once every step above works on its own, this is the whole pipeline in one message. Paste it in, fill in the brackets, and the agent saves it as one skill you can rerun on a new city with a single line.