Your leads are not junk. They are stuck. An SMS AI agent texted 14,000 of them and 4,200 wrote back.

Tuesday, 10:14 AM. Two minutes after signing up, a lead gets a text that names the lender who will call. They write back. A few messages later: “Not yet.” They have not talked to the lender. Within a minute, the agent answers with the lender’s approved number and application link.
Friday, 11:02 AM. A check-in: were you able to finish the application? The lead: “Yes! They said I’m pre approved.”
Nobody on the team typed a word.

What is an SMS AI agent? An AI that holds text-message conversations with your leads the way a good team member would. It starts the conversation, answers questions, follows up, tracks where each lead stands, and brings in a person only when one is needed.

This is for you if you buy leads, sell leads or run ads to get them, and you have ever said “these leads are junk.” Or your own team follows up when it has time. A local service business, a national lead generator, any product or service where a person has to take steps: reply, apply, send a document, book, pay.

The short version: the lenders who bought these leads could not reach many of them, and that looked like a lead quality problem. The leads were fine. They were stuck between the sign-up and the application, and nobody was walking them forward. An SMS AI agent now texts every new lead within minutes and walks them, step by step. In 20 days: 14,000 texted, 4,200 wrote back, 900 kept talking, about 60 sent a pre-approval letter. The lenders’ contact rates went up sharply.

Why it works: lead quality is not a traffic problem. It is a follow-up problem. A lead is only worth what happens in the days after the sign-up.

The company is a mortgage lead generator. Home buyers sign up for a low down payment program, and each sign-up is sold to a lender, who calls to start the application.

The lenders could not reach the leads.

The home buyer signed up under one brand. Then a stranger from another company called, from a number they did not know. Many did not pick up.

The lenders called a few times and gave up. People who needed a week because of a move, a sick kid or a slow month went quiet, and nobody asked again.

The lenders complained that they could not reach people, or that people went dormant. To a lender, that looks like a lead quality problem.

The leads were not the problem. The gap was.

Between the sign-up and the application there are steps: apply, get pre-approved, send documents. Every step is a place to get stuck, and a lead that is stuck looks exactly like a bad one.

Following up is boring and it never ends. It is also where the money is. So the team decided to follow up directly, in the brand people already knew, starting the minute someone signs up. We built the SMS AI agent that does it.

Meet Alex. It walks every lead, one step at a time.

It texts under the first name Alex, the way a short, friendly person would: casual, one question at a time. If anyone asks, it says it is an AI assistant. Here is the walk.

Step 1. Sign up: the first text, within minutes

A few minutes after someone signs up and agrees to be texted, they get one personal text. It names the lender who will contact them, so the phone call is not a stranger’s, and asks one question: when are you hoping to move?

Step 2. Apply: questions answered, the link shared

When the person writes back, the agent reads the whole conversation and their file, works out what they mean (applied, not yet, stuck, needs time, wants a person) and replies, usually in under a minute. The answers come from a program guide the team wrote, not from guesses. It shares only approved links and phone numbers.

Step 3. Pre-approval: it checks in, then asks for the letter

The lead’s status changes to waiting on pre-approval. A scheduled check-in asks how it is going. When the lead says they are approved, the agent asks for a photo of the letter, and a person confirms it.

Step 4. Documents: it reminds, and a person sees the list

When a lender asks for documents, the agent reminds the lead. Anything that needs a human lands on one Priority list: pre-approval letters to confirm, documents a lender asked for, leads who say nobody has called them, and leads who ask to talk to someone.

Step 5. The lender takes it from there

Every lead carries a status: needs to apply, waiting on pre-approval, documents needed, long term, not interested, opted out. The lender’s team starts from a lead that has a file, not a stranger.

Why leads do not stay stuck

The check-ins keep coming for anyone who goes quiet, spaced out and friendly, in daytime hours only. The schedule bends to what people say. “We’re both sick this week” pushes the next check-in out a week and off the weekend. A move in five months gets a check-in about once a month. “Stop” ends everything, immediately.

And when a lead does come back, the agent picks up the same thread, in their language, with the right lender and the next step. About 100 leads came back after a week or more of silence. One wrote back after six weeks, in Spanish, and got a Spanish answer in under a minute.

Who checks the agent

After every text goes out, a second AI re-reads it against the rules and the conversation and marks it OK, Minor or Flag, with a reason. It runs after sending, so it never slows a person down. 98% of texts came back OK or minor. The rest are flags: a text that asked two questions at once, or a “no late payments, no collections” answered with “sorry about all the messages!” A flag lands in a QA tab, the fix becomes a rule in the agent’s instructions, and the flag is marked fixed.

The guardrails are built in, before anything sends:

• Only people who agreed to be texted.
• Only approved links and phone numbers. If a text has anything else, it is blocked.
• Daytime hours only, and at most 2 texts per person per day.
• “Stop” or “remove me” is respected immediately.
• If someone asks whether it is a bot, it says it is an AI assistant.
• It never quotes a rate or promises an approval. The lender does the lending.
• If it cannot answer from the program guide, it says a team member will follow up and puts the lead on the Needs reply list.
• Every decision is logged, and each feature has an on/off switch.

In the first 20 days
Leads texted	14,000
Wrote back	4,200 (30%)
Kept talking (3 replies or more)	900
Sent a pre-approval letter	About 60
Came back after a week or more of silence	About 100
Typical reply time	Under 1 minute
Texts that passed the second AI’s check	98%
Said thank you	750
Asked if it was a bot	0
Said stop (and it stopped)	2,000 (14%)

On the lender side, contact rates went up sharply, more people moved through the process, and more sent their documents. The leads did not change. The follow-up did.

What it changes for you

If you sell leads: your buyers judge your quality by what happens to each lead. When more of them answer, apply and send documents, the complaint changes.

If you buy leads or run the ads: before you cut spend or change vendors because the leads are junk, check who followed up with them, how fast, and how many times.

If you have a sales team: your people only see the few conversations that need a person, sorted by type, and every automatic text shows why it was sent.

It fits anywhere people have to take steps: a roofing estimate, a dental booking, a demo, an application, a first payment. People get stuck between steps. The fix is the next text.

A lead is an expense until someone texts back. Then it is cash flow.

What stays human: the lender’s call, the final yes, every pre-approval letter someone confirms, and anything that needs a person.

Want to build this yourself? Here’s how it was built, in plain words. It works for any business with inbound leads that need to move through steps.

The piece	What it does
Claude Code	Builds it and tunes it
A texting service with an API	Sends and receives the texts and handles “stop”
An AI model	Reads the thread and the file, works out what the person means and writes the reply
A second AI model	Re-reads every text after it is sent and marks it OK, Minor or Flag
A database and a simple admin screen	The lead file, status, thread, decision log, Priority list, QA tab and on/off switches
A short program guide	The only source the agent answers offer questions from, written by your team

Mortgage and texting rules vary by state and by business. Check with your own counsel before copying any of this.

Prefer not to build it? Have us do it for you.