Tool of the Week: Retell AI
Here's a number most business owners never calculate: what a missed call costs.
Someone calls, nobody picks up, they don't leave a voicemail — they call the next business on the list. For a service business, that's not a minor leak. That's a booked job walking to a competitor.
I've spent the past few weeks building AI voice agents to attack exactly this problem, and the tool I keep coming back to is Retell AI.
What it is: a platform for building AI phone agents — the kind that answer a call, hold a real conversation, answer questions about your business, and book an appointment. Not the phone-tree robot you hate. These agents handle interruptions, follow tangents, and come back to the point. Callers regularly don't realize they're talking to software until told.
What surprised me building with it:
You assemble it, you don't code it. There's a visual conversation-flow builder. You pick the voice (it supports several voice providers, including ElevenLabs), pick the brain (GPT, Claude, and Gemini models are all options), write the instructions, and connect a phone number.
You can test before going live. Retell runs simulated conversations against your agent so you catch the awkward responses and edge cases before a real customer hits them. This is the feature that separates "demo" from "deployable."
The pricing is usage-based, and honest math matters here. The advertised rate is $0.07/minute, but that's just the voice infrastructure. Once you add the language model and telephony, a realistic all-in cost runs roughly $0.13–$0.31 per minute depending on which models you pick. There's no mandatory subscription — you pay from $0 and only for connected calls.
Run the comparison: a part-time human answering phones costs $1,500+/month. An AI agent answering every call, nights and weekends included, at even heavy usage of 500 minutes/month runs you $65–$155. It won't replace a great front-desk person. It will beat voicemail every single time.
The honest caveats: voice agents are only as good as the instructions you give them, the first version will mishandle calls you didn't anticipate, and you need to review transcripts weekly early on. Budget for iteration, not perfection on day one.
Who this is for: Any business that takes bookings or inquiries by phone — home services, dental, salons, restaurants, law offices. If calls go unanswered at your business after 5pm, this category is worth your attention now, not next year.
Try it: retellai.com has a free tier with test credits. Build something rough, call it yourself, and you'll understand the category in ten minutes.
Quick Hits
Apple rebuilt Siri — and it runs on Google's Gemini. At WWDC on Monday, Apple announced the new Siri is powered by a custom Gemini model Apple licenses for a reported ~$1B/year, running on Apple's own private cloud. The bigger story for the rest of us: iOS 27 (coming September) lets you choose which AI powers your iPhone — ChatGPT, Gemini, or Claude. Your AI model is becoming a setting you pick, not a vendor you're stuck with. Expect every platform to copy this.
ChatGPT's memory is getting smarter — and coming to free accounts. OpenAI announced a rebuilt memory system ("Dreaming V3") on June 4: a readable summary page showing what ChatGPT has learned about you, controls to correct it, and memories that update as your life changes. It's rolling out to Plus/Pro users now, free users in the coming weeks. Practical move: when you get it, open the memory page and read what it thinks it knows about you and your business. Then fix it. An assistant with wrong memory is worse than one with none.
Claude is coming to Excel.At Build 2026, Microsoft announced Claude as an option in Excel's Agent Mode — write and explain formulas, clean data, and build analysis without leaving the spreadsheet. Excel is where an enormous amount of real business work actually happens, and AI showing up inside it (instead of in a separate chat tab you copy-paste from) is the adoption pattern that matters.
The first real US AI law kicks in June 30. Colorado's AI Act takes effect at the end of the month — the first state law with teeth governing "high-risk" AI use in hiring, lending, housing, and healthcare decisions. Small businesses get some breathing room (sub-$25M revenue has a grace period), but the direction is clear: if you use AI to screen job applicants or make customer-affecting decisions, documentation requirements are coming. Worth a conversation with whoever handles your compliance.
Prompt of the Week: The Missed-Call Math
Before you buy any tool — voice AI included — figure out what the problem actually costs you. Paste this:
Help me calculate what missed phone calls cost my business.My inputs:Business type: [what you do]
Average value of a new customer: [your number, or say "help me estimate"]
Calls received per week (estimate): [number]
Hours the phone is actually answered: [e.g., M-F 8-5]
What % of callers leave a voicemail vs. hang up: [guess if unsure]
Walk me through:A realistic estimate of missed calls per month (include after-hours,
lunch, and "everyone was busy" misses)How many of those were likely new business vs. existing customers
A conservative monthly revenue-leak estimate, showing your math
The break-even cost for any solution (answering service, AI agent,
extra staff) — what could I pay and still come out ahead?
Be conservative. I'd rather underestimate the problem than justify a
purchase with inflated numbers.The last line is doing the work. AI defaults to making your idea look good — telling it to be conservative gets you a number you can actually trust.
Most owners who run this are surprised in one of two directions: either the leak is bigger than they assumed, or it's small enough that a $30/month answering service beats any AI build. Both answers are wins. Knowing the number is the point.
Like what you're reading? Forward it to someone who'd get value from it. And if you're curious what AI automation could look like for your business, book a free 15-minute audit — no pitch, just a look at where you're leaving time on the table.