How to Build an AI SDR That Replies, Qualifies, and Routes Every Inbound Lead in Under 60 Seconds

Inbound leads die two ways. The first is speed. Someone fills out your form, gets back a flat “Thanks, we will be in touch,” and has already talked to two competitors before a human sees it. The second is quieter and just as costly: the lead reaches the wrong person. It lands in a shared inbox, sits, gets forwarded twice, and by the time the right salesperson picks it up the moment is gone.

Most advice only deals with the first problem. This is about solving both.

I built a system that reads every inbound form fill, enriches it, scores it, works out whether it is even a real prospect, writes back a personalized reply that reacts to what the person actually wrote, and routes the lead to the right salesperson based on where they are and what they want. All in under 60 seconds. It also pings the assigned rep’s phone with a short brief, so the right person knows the moment a good lead lands.

People have a name for this now: an AI SDR. You will also see it called an inside sales rep, an AI sales assistant, or an inbound BDR. The broader category goes by several names: AI lead response, AI lead qualification, lead response automation, and speed to lead are all different labels for the same underlying problem, the gap between when a lead arrives and when a human engages.

The build below is deliberately generic. I used Make for this one, but the same pattern runs in n8n or whatever connector you prefer, and it works with any CRM or with no CRM at all. I have redacted my keys and IDs, but the pattern is the point, and it applies to almost any company that takes inbound through a form. The cost to run it is roughly $9 to $29 a month for a Make or n8n subscription, the Apollo free tier for enrichment, and a few cents of AI costs per lead.

Two Ways Inbound Leads Die

The research on speed to lead is brutal and consistent. The MIT Lead Response Management Study found that responding within 5 minutes makes you 100 times more likely to reach a lead than waiting 30 minutes, and 21 times more likely to qualify it. Harvard Business Review found lead quality drops about 80 percent after those first 5 minutes. Velocify clocked a 391 percent jump in conversion when the first response lands inside one minute. And the kicker: 78 percent of customers buy from whoever responds first, yet the average business takes around 47 hours to reply. That is not a typo. Forty-seven hours. Every minute that passes, the opportunity evaporates.

If you want to see what that gap is costing you in real dollars, our Speed-to-Lead Revenue Calculator will put a number on it.

Alex Hormozi puts it bluntly. Get your first response under 60 seconds and, in his words, you could “4X your business tomorrow.” Worth the 40 seconds:

The half nobody talks about is ownership. Even teams that respond quickly often respond from a generic inbox with no owner attached. The lead gets a reply, but it does not get a person. Nobody feels responsible for it, follow-up is inconsistent, and a lead that should have gone straight to your strongest closer in that region or that specialty ends up in a queue.

A good AI SDR fixes both at once. It replies instantly and personally, and in the same motion it decides who owns the lead and tells that person directly.

What an AI SDR Actually Does

Here is the full flow, before we build it. Every step is tool agnostic.

• A visitor submits your contact form
• Your automation platform catches it through a webhook
• It pulls the email domain and flags free inboxes like gmail and outlook
• An enrichment service fills in the company: industry, size, revenue, description, location, tech stack, founding year
• The automation scores the lead for fit using AI lead scoring, deterministically, from that data
• An AI step handles lead qualification: it classifies intent, detects what the person is interested in, and writes a tailored reply plus a short internal brief
• The automation routes the lead to the right salesperson based on location and interest
• It sends the reply, notifies the assigned rep, and creates the CRM record with that rep as the owner
• Junk is filtered out before any of that happens

Three jobs, kept separate on purpose. Scoring is math, so the automation handles it. Spam filtering, interest detection, and writing are judgment, so the AI handles those. Routing is a lookup, so it gets its own clean step. And one rule that simplifies everything: location decides who owns the lead, not whether the lead is any good. Fit and ownership are different questions.

For enrichment we use Apollo, which returns the company data the whole system runs on from nothing more than an email domain. Any decent enrichment provider works, but Apollo is the one we reach for.

The Build, Step by Step

Capture the Mobile Number. Even If You Reply by Email First.

The system above replies by email. That is a solid starting point, and for most B2B businesses it is where you start. But a lot of people would rather be texted than emailed, and capturing the mobile number on your form costs nothing. The data is valuable the moment it exists.

Add a phone field to your form, alongside or just below the email field. Make it optional. Add a short checkbox: “I’m happy to be contacted by text.” That checkbox is what creates consent, and consent is what separates a compliant SMS follow-up from a cold text nobody asked for. The Pushover alert to your own rep needs no consent because you are texting your own team. Texting the prospect does.

Once you have a mobile number and a checked consent box, the automation can branch. Email goes out as the primary reply. A parallel SMS goes out via Twilio carrying a shorter version of the same message, a one-liner that reconfirms you saw their inquiry and tees up the meeting link. Most people see a text before they see an email. An SMS arriving in seconds while the email sits in an inbox is a meaningful lift in response rate, especially for leads that came through on mobile.

WhatsApp works the same way via Twilio’s WhatsApp channel, and for international leads or audiences that live in WhatsApp it is often the higher-performing branch. The full SMS and WhatsApp build with Twilio is a separate post, because there is enough to it (compliance, message templates, opt-out handling) that it deserves its own treatment. But the form field is where it starts. Add the phone capture now and the SMS branch is a 30-minute add-on when you are ready for it.

How the Routing Works

Routing comes down to two questions: where are they, and what do they want. You can lead with either, depending on how your team is organized.

Route by territory. Map the location from enrichment to the rep who owns that region. Use city or state, whichever fits how you have carved up the map.

Route by service interest. If the form has an interest dropdown, use it. If not, the AI infers it from the message. Map each interest to the rep who specializes in it.

Pick a primary axis and let the other refine it. If your team is built around geographic territories, route by region first and use interest to flag a specialty handoff. If your team is built around specialist pods, route by interest first and use location only when two specialists could both take it. Either works. What matters is that the rules are mutually exclusive, so no lead matches two of them, and that a catch-all owns everything that falls through. An unowned lead is the whole problem you are trying to solve, so never let the system create one.

The assignment then flows into the steps that already exist: the notification goes to that rep, and the CRM record is created with that rep as the owner. The lead is answered, owned, and logged before anyone has touched a keyboard.

Assignment without enforcement still leaks. Routing sets the owner. What it does not do, by itself, is stop the owner from sitting on the lead for a day. Add a timer: if the rep has not actioned the lead within your defined window (15 minutes for a top-fit lead is a good starting point), the automation fires a second alert or reassigns to a backup. That escalation step is the difference between a routing system that works in theory and one that works in practice.

The Fit Scoring Model

The automation handles the numbers. The AI handles the one fuzzy call, industry fit, by judging the company against your ideal customer profile rather than matching brittle strings.

Bands: 9 or more is Top, 6 to 8 is Strong, 3 to 5 is Good, 1 to 2 is Low fit, no enrichment match is Needs a look.

Two overrides sit on top. A company with 10 or fewer employees lands at Low fit no matter how good the rest looks, because if they cannot afford you, nothing else changes that. And no enrichment data lands at Needs a look, since you are scoring blind.

On revenue: where Apollo returns a real number, use it. Where it does not, which is common for small private firms, estimate from headcount. A rough 150,000 dollars per employee gets you in the right range for whether a business can afford a real engagement.

The AI Step That Runs It

This is the genericized instruction set. The bracketed values get mapped to your form and enrichment fields.

You are writing a warm, human reply on behalf of [YOUR NAME] at [YOUR COMPANY],
to someone who just filled out the contact form. Do NOT assume the person came for
any one specific service. Sound like a sharp salesperson who did 30 seconds of
homework, not like a marketing email.

FIRST classify the inquiry. Set lead_type to one of: prospect, job_seeker, vendor,
spam, unclear. Set should_reply to true only when lead_type is prospect or unclear.

THEN detect interest. Set service_interest to the single best match from your list,
for example: getting_found, lead_response, automation, advertising, general.

LEAD
First name: [first_name]
Their message: [message]

COMPANY (from enrichment, may be blank)
Company: [company_name]   Industry: [industry]   What they do: [description]
Location: [city], [state]   Size: [employees] employees

WRITING RULES

- Read and react to their message FIRST, before anything else.
- Plain, spoken English. Short sentences. No em dashes. No marketing filler.
- Warm and low pressure. Never name a service they did not mention.
- Never use brackets or placeholders. If a value is blank, write around it.
- Ask exactly one short, specific question that helps you learn what they need.
- If the lead looks qualified, offer the meeting link to book a call.

Return ONLY this JSON, no markdown:
{"lead_type":"...","should_reply":true,"service_interest":"...","subject":"...","body_html":"<p>...</p>","internal_brief":"..."}

Two rules earn their place. “React to their message first” stops the AI from ignoring what the person wrote and pitching something generic. And the classification up top is what keeps your CRM clean and your replies away from the wrong people, which brings us to the next part.

Keeping the Junk Out

Not everything that hits your form is a lead. You know the ones: the cold pitch offering to redo your bookkeeping, the “I noticed your website could rank higher” SEO blast, the agency promising 30 qualified meetings a month, the LinkedIn outreach company that found you through, of course, LinkedIn outreach. Some days that is half the inbox.

The last thing you want is your shiny new AI SDR firing a warm, personal reply at a cold pitch, or a recruiter, and creating a CRM record for them. So the AI classifies intent before anything sends. Real prospects and genuine but unclear messages get the full treatment. Vendors and job seekers get a quiet internal heads up to you and nothing automated. Outright spam ends silently, no reply, no record. Your pipeline stays clean and you never auto-court someone trying to sell you backlinks.

A Fast Reply Is Not a Follow-Up

Here is the thing I keep running into, and it is the most important part of this whole post. Responding fast does not win the deal on its own. If your AI SDR fires back in 30 seconds but the actual human takes a day to follow up, you are right back to the original problem and you have just taught the prospect to expect speed you do not deliver.

Two things make the fast reply actually pay off.

First, the reply has to provide value, not just acknowledge receipt. That is exactly why the AI reads the message, qualifies the lead, and, when the lead looks qualified, offers a meeting link. Not just any meeting link. Point qualified leads at a senior person who is positioned well, because the goal is not a reply, it is a conversation. You want them on the phone, in a Google Meet, or on a Zoom call. Lining that meeting up with a calendar booking link is the single most valuable thing the email can do. Any calendar system works, and Calendly works brilliantly for this.

Second, a quiet truth worth knowing: most people do not realize the first reply was written by AI. They believe it came from the human, which is a good thing, and it means your sign-off and email signature should look as authentic as a real person’s. The flip side is the trap. The moment they reply and a real human goes silent for 24 hours, the illusion and the goodwill both evaporate.

Add a short timed sequence to the prospect too, not only a nudge to the rep. If the lead does not respond to the first email within 24 hours, the automation sends a brief second touch: a one-liner that reconfirms you saw their inquiry and checks whether they had a chance to look at the booking link. Nothing pushy. A single touch rarely closes anything; a simple two-touch sequence meaningfully improves the chance a conversation actually happens. Stop the sequence the moment they reply.

You can close the human follow-up gap too. The same automation can watch your inbox, notice when a prospect replies, and nudge the rep if they have gone quiet. And if you captured a mobile number and consent, you can branch the flow to text the lead back within seconds, which gets seen before almost any email. You can even hand a qualified lead to an AI voice agent that places a real follow-up phone call. Those are bigger builds with their own moving parts, so I will walk through them in their own posts. The point for now is that speed is the start, not the finish.

And to be straight about it: plenty of CRMs do a version of this natively. I built the platform-agnostic illustration on purpose, so you can see the moving parts no matter what you run. In practice we customize the workflow and the AI to the specific business, because the real work is figuring out which problem you are actually solving before you automate anything.

Where AI Agents and MCP Fit, and Where They Do Not

In June 2026, Make added MCP tool support to its AI Agents. An agent can now connect to external MCP servers, read the tools each server exposes, and call those tools alongside native Make modules in a single run. The pitch is tempting: one agent, every tool, let it figure out the rest.

So should you rebuild the system above as an agent? No. And the reason matters, because it applies to almost every automation decision you will make this year.

Agents reason. Pipelines execute. An agent looks at a goal, considers its tools, and decides what to do next. That deliberation is the feature, and it is also the cost. It adds seconds of latency and a margin of variance to every run. The first 60 seconds of lead response is the one place you can afford neither. You want the same eight steps to fire in the same order every single time someone hits submit. That is a pipeline job.

Where agents earn their keep is everything after that first reply. Once the instant response is out and the rep has been alerted, the stopwatch stops and the judgment work begins. An agent with MCP tools can research the company properly, beyond an enrichment lookup: their site, their reviews, their recent news. It can draft the personalized follow-up sequence. It can check the rep’s calendar through an MCP server and propose meeting times. It can watch the thread and nudge a rep who has gone quiet. None of that is time-critical, and all of it benefits from a model deciding which tool to reach for.

A simple test settles it: the stopwatch decides. If the task has a deadline measured in seconds, build a pipeline. If the task benefits from reasoning across several tools and a minute of thinking costs nothing, give it to an agent. MCP did not move that line. It made the agent side of the line far more capable.

Six Principles That Hold on Any Platform

Skip the tool-specific quirks. These are the design choices that make the difference no matter what you build it in.

Reply first, store last. Send the autoresponder and fire the notification before you write to the CRM. The conversation and the alert matter more than the database row, so a storage error should never block them.

Filter junk before you reply. Classify intent up front and only let real prospects through. This is what keeps your CRM clean and protects you from auto-replying to a sales pitch.

Make routing rules mutually exclusive, and add a catch-all. Every lead should match exactly one rule, never two, never zero. The most common bug in lead routing is a lead that matches no rule and silently vanishes. A default owner prevents it.

Plan for the no-match. Free email, unknown company, thin data. It will happen often. Those leads should still get a warm general reply, still get an owner, and get flagged so a human can decide whether to chase them.

Keep math in the tool and judgment in the AI. Do not make the AI count points, and do not make an if-statement guess whether something is spam. Each does the job it is good at, and the whole system gets simpler and more reliable.

Assign and enforce. Routing sets the owner. Enforcement closes the loop. Add a timer that fires a second alert or reassigns to a manager if the rep has not actioned the lead within your window. A top-fit lead sitting untouched for 24 hours is just a slow reply with extra steps.

A Generic Autoresponder vs an AI SDR

This Works With Any CRM

The build above does not care what CRM you use. The create-contact step simply swaps for whatever you run. And the webhook trigger is not limited to web forms: the same pattern handles leads arriving from Meta lead ads, Google Local Services Ads, Angi, Thumbtack, or any other platform that can fire a POST request or trigger a webhook. The enrichment, scoring, qualification, and routing logic stays exactly the same regardless of source. People searching for speed to lead usually search it next to their own CRM or platform, so here is how some common ones stack up, and where a platform gives you more than a do-it-yourself build.

Capabilities vary by plan and change often, so treat this as a starting point. The pattern in this post layers on top of any of them. Where HubSpot and Zoho pull ahead is the depth of native workflow, multi-step qualification, and AI, which is why those are the two we build full managed AI SDRs on. If you want the deeper version, see the 60-second AI lead response build inside a CRM and the HubSpot vs Zoho AI SDR comparison.

Frequently Asked Questions

The Bottom Line

You can build all of this yourself. The steps are here, the AI instructions are here, and the routing logic is the part most teams are missing. If you take inbound through a form, it is worth doing, because the moment after someone hits submit is the most valuable and most wasted moment you have.

You could also just have us build it. We have implemented this across multiple platforms and multiple businesses, so you skip the trial and error, it gets done fast, and you get the version that already accounts for everything we learned the hard way. We ran a version of this engine on 1,000 dormant leads in our own CRM. The AI SDR booked 24 meetings and helped generate over $750,000 in proposals, including a $200,000 deal closed in the first month. Read the full story. For most businesses a system like this pays for itself inside the first 90 days, because winning even one or two extra deals you would have lost to a slow reply covers it many times over.

Want to know what slow response is costing you first? Run the numbers in our Speed-to-Lead Revenue Calculator. And if you would rather have a more advanced version running natively in your CRM, tuned to your business and monitored as real leads come through, that is exactly what we set up with our AI Lead Response system on HubSpot and Zoho.

---

Rivetline builds AI SDRs and speed-to-lead systems on HubSpot and Zoho, the Convert tier of our model. See how the setup works, run the free AI Visibility Report, or book a call to talk through your inbound.