← Back to Blog
The Front Door Problem: One AI Intake Router for Every Client Request

The Front Door Problem: One AI Intake Router for Every Client Request

November 3, 2026·7 min read

Last week I counted the ways a client can reach me. WhatsApp, Instagram DMs, two email addresses, a text message, and a contact form on a page I have not opened since March. Six front doors into one brain, and that brain is also doing the invoicing.

Nobody designed this. It accumulated one channel at a time, each one reasonable on its own. And it has quietly become the most expensive line item in most founder-led service businesses, because a lead that dies in a requests folder never shows up anywhere. You just feel vaguely behind all the time.

The job you never hired for: being the router

Your bottleneck is not the work. It is the sorting. Every message that arrives needs a human judgment before anything can happen to it: client or stranger, urgent or just loud, phone call or paragraph, answer now or answer after the thing I am already late for. That sorting step is the front door, and it is nobody's job in a business with nine people and a founder.

I ran a training company in New York for 18 years before any of this. We trained more than 90,000 people, and the reason it held together at that volume was not talent. It was a front desk. Somebody whose actual job was to receive everything and send it somewhere sensible. Founders running service businesses today have no front desk. They have a phone in their pocket and eleven tabs open.

A brand photographer I talked to in June had 41 unread messages in her Instagram requests folder. Four were from a venue group that wanted a $9,000 shoot. She found them 19 days later. Her marketing worked fine. Her front door did not. That story is the whole problem, and it is why "I need to clone myself" is the most common sentence in my inbox.

What an AI intake router actually does

A CRM is a filing cabinet. It stores what you already decided to store, and it assumes you did the deciding. A router is the mailroom. It receives, opens, reads, decides, and sends. Those are different jobs and only one of them is a software subscription you can buy off the shelf.

An AI intake router for service businesses is three things stacked: one address, one classifier, one queue. Every channel points to the same place. A model reads each message, pulls out who it is from, what they want, how fast it needs an answer, and which of your existing processes it belongs to. Then it drafts a reply, sends a booking link, or flags you. The plumbing is boring on purpose. Twilio or 360dialog for the WhatsApp Business API, one Google Workspace alias for email, Make.com or n8n to move payloads, Airtable or Notion as the queue, and a small cheap model like GPT-4o mini or Claude Haiku for the classification. No custom development. I have built versions of this in an afternoon.

Four buckets, and the eight fields that matter

Before you touch a tool, write down the only four categories any request can fall into in a service business: new business, existing client request, admin and vendor, and noise. Almost every founder I work with is unknowingly treating all four as the same priority level, which is why the vendor asking about a W-9 interrupts a $40,000 proposal.

Now build the queue. An Airtable base with eight fields: Name, Channel, Type, Urgency, Summary, Suggested next action, Owner, Status. That is your client request management system, and it will do more for you than a $200 a month platform with a dashboard you will never open. Each bucket gets its own view, and each view gets a rule. New business gets a summary, a source channel, and a Calendly link. Existing client requests get routed to a project thread with a date attached. Vendors get batched into a Tuesday and Thursday digest, because they genuinely do not need same-day answers and you know it. Noise gets archived with a polite holding reply that costs you nothing.

The weekend build order

Start with the doors. Move everything onto one WhatsApp Business number and one hello@ address, then forward the stragglers into it. Instagram and Facebook messages can route through the Meta API with Make or ManyChat. Now you have stopped creating new inboxes faster than you can close them.

Then write the classifier prompt. Three rules and a schema, not twenty rules and a personality. Paste in the eight field names, ask for JSON, keep the temperature low, and give it the bucket definitions verbatim. Vague prompts produce a router that guesses, and a router that guesses is worse than no router.

Then wire the outputs. A Slack channel for new business, a drafts folder for client replies, an automatic booking link on qualified inquiries, and a Thursday digest for the rest. If you want to watch this get built rather than read about it, there is a walkthrough in the workshop archive at https://workshop.mastermindshq.business.

Finally, shadow it for a week. Let the router classify everything and draft everything but send nothing. Compare its drafts to what you would have written. You will find the prompt is wrong in two specific ways, you will fix those two things, and then it will not be wrong anymore.

Where the router should stay out of the way

The fastest way to lose a good client is to let software handle a moment that needed a person. Do not let it quote prices. Do not let it apologize on your behalf or promise a timeline you have not approved. Do not let it answer anything that starts with "are you okay with." Those are the three places trust gets built or broken, and trust is not a classification problem.

Set one hard escalation rule: if the model is uncertain, or the message contains money words like refund, contract, invoice, or cancel, it goes straight to you with a summary and no reply. Then spend ten minutes every Friday reviewing the mismatches. That review is the entire maintenance cost of business systemization for founders, and skipping it is how people end up telling me AI does not work.

None of this is about replacing a human being. It is about making sure that WhatsApp and email client requests in one place, and that the place is not your head at 11pm. A front desk does not need to be a person anymore. It does need to exist. That is the difference between a business that grows and a business that just gets louder.

If this resonates and you want to build this kind of infrastructure in your own business, the Mastermind is where we do the work live. You can learn more at mastermindshq.business, and if you want to see how it stacks up against other programs first, that comparison lives at https://www.mastermindshq.business/mastermindvsl.

Ready to put this into practice?

Join Joe Che's AI Business Mastermind, a small cohort for founder-led service businesses that want to systematize their operations with AI.