Moser Research Builds Its Own AI Phone System

The Challenge

Moser Research is two people: Doug handles architecture and technical work, Courtney handles operations and scheduling. For the first few months, every business call went to a personal cell phone.

That arrangement has predictable problems.

When you’re on a client call and a prospect rings your cell, it goes to voicemail. If they leave a message at all. In our experience, most first-time callers who hit voicemail don’t call back — they move on to the next name on their list. We were losing opportunities we didn’t even know about.

Beyond missed calls, there was the professionalism issue. Our caller ID showed personal numbers. Clients and prospects had no way to distinguish a business call from a personal one, and neither did we. Work calls came in at dinner, on weekends, during family time. There was no separation between “the business is open” and “Doug’s phone is on.”

We looked at the alternatives. Traditional answering services run $200–$500 per month for a small business (as of early 2026), and most of them are just people reading from a script. They can take a message, but they can’t answer questions about your services or intelligently route a call based on what the caller actually needs. After-hours coverage usually means an extra fee for the same limited script.

Virtual phone systems with auto-attendants — the “press 1 for sales, press 2 for support” variety — are cheaper, but they create the exact impersonal experience we were trying to avoid. Nobody wants to navigate a phone tree to reach a two-person company.

There’s an irony in being an AI consultancy that helps businesses automate their operations while running your own phone system off personal cell phones. We decided to fix that the same way we’d fix it for a client: start by understanding the problem, then build the right solution.

What We Did

We approached our own phone system the way we approach any client engagement — understand first, then design, build, and maintain.

Mapped the Problem

Before writing a line of code, we reflected on our actual call patterns — who was calling, when, and what they needed.

The picture was clear. The majority of inbound calls fell into a handful of categories: people asking who to talk to about a specific service, requests to schedule a conversation, and general questions about what we do. A smaller but persistent category was spam — robocalls, SEO pitches, and the usual noise that every business number attracts.

We also knew that a meaningful number of legitimate calls arrived when both of us were already on other calls or away from our phones. Those were the ones turning into missed opportunities.

This gave us what we needed: a clear picture of the call types, the routing logic, and the failure modes we needed to solve.

Designed the Automation

With the call patterns mapped, we designed a call flow that could handle the common cases without human intervention while still connecting callers to us when it mattered.

The flow works like this: a caller dials our toll-free number (888-600-9330) and is greeted by an AI receptionist that sounds natural and conversational. The AI identifies what the caller needs — are they asking about a specific service, looking to schedule a conversation, or do they have a general question? It can answer basic questions about our services and team on its own. When a call needs to reach a team member, the system transfers with whisper screening: you hear who’s calling and why before deciding whether to pick up. If we’re unavailable, the AI takes a voicemail and delivers a transcription via SMS.

We also designed spam filtering with rate limiting, so the persistent robocallers get caught before they ever ring our phones.

Built the System

The AI receptionist runs on a serverless stack designed for reliability and low cost. Twilio handles the telephony layer with a toll-free number. Cloudflare Workers provides the compute — serverless, so we pay nothing for idle time and the system stays available without managing servers. The AI itself is powered by Claude for conversation, ElevenLabs for natural-sounding speech, and Deepgram for speech recognition.

The key feature that makes this work for a small team is whisper screening. When the AI transfers a call, the team member’s phone rings and they hear a brief whisper: the caller’s name and what they’re calling about. Press 1 to accept the call. Press 2 and the AI seamlessly picks back up, apologizes that the person is unavailable, and takes a voicemail. The caller never knows the difference — they just experience a professional interaction either way.

Voicemail messages are transcribed and delivered as SMS, so we can read and prioritize them without listening to audio.

The AI receptionist knows about our services, our team, and how we work. It can answer the kinds of questions that used to require one of us to pick up the phone — or that went unanswered when we couldn’t.

Throughout all of this, our personal phone numbers are never exposed to callers. Every interaction goes through the business number.

Keeping It Running

A system like this isn’t a one-time build. We’re iterating on the AI’s conversation handling based on real calls — adjusting how it introduces itself, how it handles edge cases, and how it responds to unusual requests. Spam thresholds get tuned as new patterns emerge. We’re adding features as our needs evolve, like email notifications and outbound call support.

This is the maintenance side of any custom system: it works on day one, and it works better on day thirty because you’re learning from real usage.

The Results

Immediate impact: Since the system went live, inbound calls are answered professionally around the clock. No more calls going to personal voicemail. No more prospects who called once and moved on.

Privacy: Our personal cell numbers are completely shielded from callers. Every interaction — inbound and outbound — goes through the business number. Clients and prospects see a professional toll-free number, and our personal phones stay personal.

Cost: At our current call volume, the entire system runs for under $10 per month. That covers Twilio’s telephony fees, AI API costs for conversation and voice, and Cloudflare Workers compute on the free tier. Compare that to $200–$500 per month (as of early 2026) for a traditional answering service that can’t answer questions about your business or screen calls intelligently.

Team experience: Whisper screening changed how we think about incoming calls. Instead of answering every ring and hoping it’s not another SEO pitch, we hear who’s calling and why before we pick up. Press 2 and the AI handles it gracefully — the caller gets a professional voicemail experience, and we get a transcription via text. We answer the calls that matter and skip the ones that don’t, without anyone on the other end knowing the difference.

Caller experience: People who call our number have a natural conversation with an AI that understands our business. No phone tree, no hold music, no “your call is important to us.” The AI can answer common questions, route to the right person, and handle voicemail — all in a way that feels professional and responsive.

Why this matters: This is the same Understand, Automate, Build, Maintain methodology we use with clients, applied to our own operations. We mapped the problem before building anything, designed the automation around our actual call patterns, built a custom solution on the right architecture, and continue to maintain and improve it based on real usage.

Key Takeaway

We didn’t build this for a demo or a case study. We built it because we had a real operational problem: two people, one phone system that wasn’t working. Prospects were calling and getting voicemail. Personal numbers were showing up on caller ID. There was no separation between business and personal calls.

The fact that it runs for under $10 per month on serverless infrastructure is a side effect of choosing the right architecture for the job — not overbuilding, not subscribing to an expensive service that does less than what we needed.

That’s what we mean by Build — not technology for its own sake, but custom solutions designed around how your business actually works.

How We Help

This project followed the same methodology we use with every client:

• Operations Audit — We mapped the problem before building anything
• Business Automation — We designed the call flow and spam filtering
• Custom Applications — We built a purpose-built AI phone system
• Reliability Retainer — We keep it running and improving

Facing a similar challenge? Get in touch and let’s talk about it.

This case study describes our actual experience building and deploying an AI phone system for our own business. While we believe this approach can produce similar results for other small businesses, specific outcomes depend on your call volume, team structure, and requirements. It does not constitute technical or professional advice.