$700 a Month, 80% of Tickets Gone: The New Support Math

Occasionally someone in tech does the thing you've been begging them to do: they skip the demo, skip the whitepaper, and just tell you what it costs and what happened. AssemblyAI's Matt Lawler did exactly that, publishing that his company replaced their support bot with an AI agent, pays $700 a month for it, and the thing resolves 80% of incoming support tickets. That's the whole story. Published October 4, 2026.

I want to be upfront: the source page wasn't available when I wrote this, so I'm working from what was reported in the summary. I'll tell you exactly where the edges of my knowledge are. But the three numbers Lawler put out there are worth taking seriously even at face value, because a named executive at a named company said them publicly. That's rare enough to be news.

What Lawler actually said

According to the published summary, AssemblyAI (an AI audio and speech intelligence company) was running a support bot. They swapped it out for an AI agent. The agent costs $700 a month and resolves 80% of their support tickets without human intervention. Matt Lawler is the person who said it, which matters because that's an attribution you can trace, not an anonymous case study.

That's what I have. Price, result, name, company. I don't have the vendor name, the ticket volume, the previous cost of the bot they replaced, or what happens to the other 20% of tickets. If those details matter to your decision, they have not been published in what I can see.

What I can do is tell you why those three numbers are important even without the footnotes.

The new baseline problem

Here's the framework I'd ask you to hold onto: call it the Baseline Shift Test. When a real company publishes a real number, that number stops being a benchmark and becomes a floor. The question is no longer "could AI handle my support volume?" The question is "why am I spending more than $700 a month on something that handles less than 80%?"

I've been doing something close to this math since we built an AI receptionist for a client whose front desk was missing roughly 40 inbound calls a month. Those weren't spam calls. Those were patients calling during lunch, after hours, on a Saturday, and getting voicemail. The fix cost less per month than a single no-show appointment cost the practice. TPC Scottsdale ran the same play on their guest services, and the logic holds across industries: a live human handling repetitive inbound questions is not a competitive advantage. It's just overhead dressed up as service.

The Baseline Shift Test runs like this. Take what you're currently spending on the function, whether that's a staffer's time, a vendor contract, or a tool subscription. Then ask: does it resolve 80% of requests without my team touching it? If the answer is no, you have a gap. If the answer is "we don't track that," you have a bigger gap.

What this means for your business

Most local businesses I talk to in Phoenix and Scottsdale are not running support bots. They're running one of three things: a human who fields questions all day, a FAQ page nobody reads, or a contact form that creates a 24-hour lag before anyone feels helped. All three of those options cost more than $700 a month when you account for the actual time involved, and none of them resolve 80% of anything on their own.

The category of business this applies to most directly: anyone with a high volume of repetitive inbound questions. Dental and ortho practices. Home services companies. Med spas. Law offices. Any business where the same 10 questions come in 50 times a week and someone on your team answers them manually each time. That someone could be doing something your AI agent cannot do.

The category that can probably ignore this for now: businesses with genuinely complex, bespoke customer interactions where every conversation is different and requires judgment, context, and relationship. If your clients are paying you for something irreplaceable, they're not calling your support line anyway.

On cost: $700 a month is in the range where the math works for businesses doing any real volume. That said, I don't know which vendor Lawler is using, and I'd encourage you not to assume every $700-a-month product delivers the same result. The price is a signal, not a spec sheet. You'd want to know ticket volume, ticket type, and how the 80% was measured before you signed anything.

For what it's worth, our AI receptionist builds sit in a similar cost range for clients, and the resolution rate depends almost entirely on how well the underlying knowledge base is built. That's where most implementations fail, not the AI itself.

Under the hood

For the technically curious: the jump from "support bot" to "AI agent" is not cosmetic. A traditional support bot is a decision tree. You ask it a question, it matches keywords, it routes you to a canned answer or a human. It doesn't understand context, it doesn't retain memory within a conversation, and it falls apart the moment a question doesn't match a pattern it was built for.

An AI agent running on a large language model reads the full conversation, pulls from a knowledge base, can take actions (look up an order, check a status, escalate with context), and generates a response that actually addresses what the person asked. The 80% resolution number is only possible because the agent is doing something closer to reasoning than pattern-matching. The 20% that still goes to humans is presumably the stuff that actually needs a human, which is how it should work.

The practical implication: the setup cost is higher than flipping on a chatbot widget. You need a knowledge base that's accurate, maintained, and scoped correctly. Garbage in, garbage out, and in support contexts garbage out means an angry customer who got a wrong answer from a confident AI.

What to do this week

A few concrete moves, none of which require a vendor call yet.

• Count your repetitive inbound questions. Pull the last 30 days of emails, DMs, or call notes and tally which questions came in more than three times. That list is your AI agent's starting knowledge base. If the list is short, you may not have the volume to justify the spend. If the list is long, you're already paying for someone to answer it manually.
• Run the Baseline Shift Test on your current setup. What does your current inbound support function cost per month, fully loaded? What percentage of questions does it resolve without a follow-up? Compare those two numbers to $700 and 80%. The gap will tell you whether this is worth investigating further.
• Don't buy anything until you can define "resolved." Lawler's 80% number is meaningful because resolved presumably means the customer got what they needed without escalating. Make sure any vendor you talk to defines resolution the same way you do. "Deflected" and "resolved" are not synonyms, and a lot of vendors use them interchangeably.
• Watch for the vendor disclosure question. As you look at AI agent options, some platforms are starting to require or recommend disclosing that customers are talking to AI. This is a real and growing issue, especially in voice and text-based customer service. Build your policy before your vendor builds it for you.

Skip it if

You can ignore this one if your inbound question volume is genuinely low (fewer than 20 to 30 repetitive contacts a week) or if your customer relationships are the kind where a canned AI response would cost you more in trust than it saves in time. Not every business needs a support agent. Some businesses need to pick up the phone.

The question I'd actually ask you, the one I'd ask a peer over coffee: what's your current cost per resolved support question, and do you actually know the number? Because Lawler knows his. And right now, that's the whole advantage.

Sources

• AssemblyAI's Matt Lawler: A $700-a-Month AI Agent Replaced Our Support Bot and Resolves 80% of Tickets, finance.biggo.com, October 4, 2026