What I Learned Building an AI Receptionist for Small Business

Why I Built AlphaAssist After Watching Small Businesses Lose Customers

Last month I called seven local contractors for quotes on some home repairs. Four never answered. Two went straight to voicemail with boxes that were full. One actually picked up.

That's not unusual. Small service businesses miss roughly 30% of inbound calls, according to my own informal tracking of demo calls we've made to test competitors. When a plumber is on a roof at 2pm and his phone rings, he's got maybe three seconds to decide: climb down and risk falling, or let it ring and hope the caller leaves a voicemail they probably won't.

Most choose safety. The caller moves to the next Google result.

I built AlphaAssist to solve this specific problem — not to replace human receptionists (most small businesses don't have those anyway), but to catch the calls that would otherwise go unanswered. After six months of operation, I've learned what actually works for an AI receptionist and what doesn't.

The Technical Stack That Actually Answers Calls

AlphaAssist runs on Twilio for call handling, OpenAI's Realtime API for voice conversation, and Cartesia Sonic 3 for text-to-speech. I chose this combination after testing Retell, Vapi, and building our own solution from scratch.

OpenAI Realtime gives us sub-200ms response times, which matters more than I initially thought. When someone calls a business, they expect human-like conversation flow. A one-second delay between "Hello" and "Hi, thanks for calling Smith Plumbing" feels broken. Retell was consistently hitting 800ms-1.2s delays in my testing, which made conversations feel robotic.

For SMS follow-up, we use Claude Haiku instead of OpenAI because latency matters less than structured response quality. Haiku consistently formats appointment confirmations and basic information requests better than GPT-4o in our testing.

The hardest part wasn't the AI — it was figuring out call routing. Small businesses have weird phone setups. Some forward their Google listing number to a cell phone. Others use VoIP systems they set up five years ago and barely understand. We had to build compatibility with seventeen different forwarding scenarios just to handle our first fifty customers.

What It Actually Does (And Doesn't)

AlphaAssist handles three things well: answering basic questions, scheduling appointments, and taking detailed messages. It knows business hours, services offered, and pricing for common requests. When a customer calls asking "Do you do emergency plumbing?" it can say "Yes, we offer 24/7 emergency service. Emergency calls start at $150 for the first hour."

For appointment scheduling, we integrate with Google Calendar and Calendly. The AI can check availability and book standard service appointments. A typical conversation goes: "I need my gutters cleaned." "I can schedule that for you. We have availability Tuesday at 2pm or Wednesday at 10am. Which works better?"

But there are hard limits. AlphaAssist can't handle complex technical troubleshooting, give medical advice, or negotiate custom pricing. When someone calls with a complicated HVAC problem, it takes a message and promises a callback. When someone wants a quote for a unique construction project, same thing.

The AI also struggles with extremely noisy environments. Construction sites, busy restaurants, cars on highways — anywhere the background noise consistently hits 70+ decibels. We've improved this with noise filtering, but it's still not perfect.

Where Most Competitors Get It Wrong

Companies like Nextiva and Rosie focus on enterprise features: complex call trees, CRM integrations, advanced analytics. That's exactly backwards for small businesses. A three-person landscaping company doesn't need a sophisticated phone system — they need something that answers the phone when they can't.

Goodcall tries to be a complete business management system with inventory tracking and invoicing. Most small business owners I talk to already have systems they like for that stuff. They just want their phones answered.

Bland focuses on outbound calling and sales automation. That's useful for some businesses, but the primary pain point for service businesses is inbound calls going unanswered, not making more outbound calls.

The sweet spot is simple: answer the phone, sound professional, handle the basics, and escalate when necessary. Everything else is feature bloat that increases cost and complexity without solving the core problem.

Real Implementation Challenges Nobody Talks About

Training the AI on your specific business takes longer than vendors admit. We tell customers to budget 2-3 weeks for setup, not the "instant deployment" that most competitors promise. The AI needs to learn your pricing structure, your service area, your scheduling quirks, and your brand voice.

Phone number porting is a nightmare. If you want to keep your existing business number, you're looking at 7-14 days minimum, assuming no complications. Half our customers hit complications: the number isn't portable, there's a billing issue with their current provider, or the number is tied to a business loan covenant they forgot about.

Integration with existing tools rarely works smoothly on day one. We've connected with hundreds of different scheduling systems, CRMs, and business management tools. Each one has quirks. QuickBooks integration might work perfectly for one customer and completely fail for another depending on how they set up their chart of accounts.

When You Shouldn't Use an AI Receptionist

If your business takes fewer than twenty calls per month, paying $29.99-$99.99 monthly for AlphaAssist doesn't make financial sense. Just use a good voicemail system and return calls promptly.

If your calls require immediate human judgment — medical practices, legal advice, crisis counseling — don't use AI. The technology isn't there yet, and the liability risks aren't worth it.

Professional services that sell primarily on personal relationships (financial advisors, consultants, high-end contractors) should be cautious. Some customers specifically want to talk to the owner, not an AI. We recommend using AI for basic screening and scheduling, but making it clear that the owner will call back for substantive discussions.

Complex B2B businesses with long sales cycles probably need human receptionists who can qualify leads and handle nuanced conversations. AlphaAssist works best for transactional service businesses where customers are calling to book standard services or get basic information.

The Math on Missed Calls

Here's the calculation that makes AI receptionists obvious for most small service businesses: if you miss thirty calls per month and convert even 10% of answered calls to customers with an average job value of $200, you're losing $600 monthly in revenue. AlphaAssist starts at $29.99.

The payback period is usually under two weeks. I've seen customers track significant revenue increases within the first month, not because the AI is amazing, but because they're finally answering their phones consistently.

The less obvious benefit is sleep. Business owners who gave me their personal cell numbers as their business line often mention that the AI letting them silence their phones at night was worth the subscription cost alone.

Try It Before You Commit

We run a live demo line at (413) 331-7776. Call it right now if you want to hear how AlphaAssist actually sounds. It's set up as a sample HVAC business, so you can ask about furnace repairs, air conditioning installation, or emergency service.

The demo shows both the capabilities and limitations. The AI will answer questions, try to schedule an appointment, and take a message if you ask for something complex. You'll get a sense of the conversation flow and response quality within thirty seconds.

If it sounds like something your customers would accept, we can set up your actual business system in 1-2 weeks. If it doesn't sound professional enough for your brand, you'll know immediately without wasting anyone's time.