Feb 13, 2025 · 1h 11m · mad
Farewell, Chatbots: AI Agents Are Taking Over Customer Service | Mike Murchison, CEO, Ada
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Mike Murchison, CEO of Ada, joins host Matt Turck on The MAD Podcast to discuss how autonomous AI agents are revolutionizing enterprise customer service by shifting focus from simple deflection to true issue resolution, advanced multi-model orchestration, and cross-channel agentic workflows.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 20.6% of the talking time here. How this is scored →
Matt as informed peer 2.7 Guest teaching 2.9 Guest disagreement 0.1 Matt pushing back 0.3
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 35:31 Rejecting the idea that human handoff is a failure
Mike explicitly challenges the standard industry belief that delegating to a human is a system failure, reframing handoffs as AI quarterbacking inside the company.
Hardest push from Matt ▶ 46:41 Challenging the premise on department specializationMatt pushes back on Mike's thesis that department silos are artificial, arguing instead that human cognitive limits inherently require specialized sales and support roles.
Biggest teaching moment ▶ 1:01:30 Counterintuitive psychological delay in AI emailsMike educates the host on human user psychology, revealing that instantaneous AI email replies actually harmed customer trust until an artificial eight-minute delay was added.
Matt holds his own ▶ 1:09:58 Framing cross-agent collaboration as the automated enterpriseMatt demonstrates macro tech expertise by synthesizing Mike's agent workflow into the broader venture capital thesis of the fully automated enterprise.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Episode Highlights and Key Interview Quotes | 1 | 0 | 0 | 0 | Matt introduces the episode, framing customer service as a key proving ground for deployed AI agents. Mike notes that global expenditure on customer service remains high despite poor customer experiences. | |
| Value Accrual at the AI Application Layer | 3 | 2 | 0 | 1 | Matt brings up the narrative that foundation models might displace application-layer AI companies, citing Klarna's headlines. Mike explains that value accrues at the application layer due to the complexity of control, observability, and continuous improvement. | |
| Ada's Journey from Pre-ChatGPT Classifiers to LLMs | 2 | 3 | 0 | 0 | Matt asks about Ada's transition from pre-ChatGPT intent classifiers to modern LLMs. Mike details how scaling customer service historically forced a trade-off between quality and cost, which LLMs dismantle. | |
| Multi-Model Orchestration and Instruction Adherence | 3 | 3 | 0 | 1 | Matt makes a timely joke about DeepSeek and asks about Ada's underlying model stack. Mike explains their reasoning engine's dynamic orchestration across 7 to 9 models and the challenge of testing instruction adherence. | |
| Dynamic Routing, Cost Efficiency, and Specialization | 4 | 3 | 0 | 1 | Matt asks technical questions regarding dynamic routing factors, fine-tuning, and prompt engineering. Mike clarifies that Ada custom-assembles dynamic prompts per customer rather than retraining models for each client. | |
| RAG Architecture, Security Guardrails, and AI Expectations | 3 | 4 | 1 | 0 | Matt inquires about RAG, accuracy, and guardrails against hallucinations. Mike points out a double standard where buyers expect absolute perfection from AI agents while accepting high human error rates. | |
| Redefining Conversation Resolution and Transcript Analysis | 3 | 4 | 0 | 1 | Matt asks how Ada defines 'resolution' versus a frustrated customer hanging up. Mike explains that 100% automated transcript evaluation allows LLMs to grade conversation quality better than human annotators. | |
| Actions, Web Navigation, and Cross-Channel Continuity | 3 | 3 | 0 | 0 | Matt probes the shift from static FAQ bots to action-taking AI agents, referencing OpenAI's Operator. Mike describes web actions that navigate UI back-offices directly when APIs are unavailable. | |
| Seamless Authentication and Unified Omnichannel Profiles | 3 | 2 | 0 | 0 | Matt asks about authentication challenges during automated agent interactions. Mike details authenticated SDKs and cross-modal capabilities like voice agents sending text links mid-call. | |
| Rethinking Handoffs: Human Delegation and the AI Quarterback | 2 | 4 | 1 | 0 | Matt asks how human handoffs are triggered and whether they will vanish. Mike reframes handoffs away from being viewed as system failures, describing the AI as an internal quarterback delegating tasks. | |
| The Rise of AI Customer Experience (ACX) Management | 2 | 3 | 0 | 0 | Matt asks how customer service jobs will change over the next few years. Mike describes the rise of ACX (AI Customer Experience) managers who oversee AI agents rather than handling individual tickets. | |
| Analytics and Customer Insights in Ada | 2 | 2 | 0 | 0 | Matt asks what dashboards and analytics ACX managers use. Mike explains how aggregated customer conversations allow clients to discover missing product offerings, like launching a new sunglasses SKU. | |
| High-Empathy "Serve to Sell" Strategies | 3 | 3 | 0 | 1 | Matt notes that specialization exists because humans have cognitive limits. Mike responds that AI removes departmental boundaries, re-unifying customer service and sales into a single interface. | |
| Overcoming Bad Chatbots through Resolution | 3 | 3 | 0 | 0 | Matt brings up widespread consumer hatred for frustrating chatbot loops. Mike explains that traditional chatbots failed because they optimized for deflection rather than true issue resolution. | |
| Empathy Customization, Coaching, and AI Pride | 3 | 3 | 0 | 0 | Matt asks about tone and empathy customization. Mike highlights natural language coaching features and notes that clients literally post photos of their top-performing AI agents on office walls. | |
| Accelerating AI Onboarding with "Agent in a Box" | 2 | 3 | 0 | 0 | Matt asks how companies accelerate AI onboarding. Mike shares that Ada shifted from blank-slate customer setup to providing an out-of-the-box 'Agent in a Box' on day one. | |
| Expanding Ada across Voice, Email, and Modalities | 3 | 4 | 0 | 0 | Matt asks about Ada's expansion into voice and email modalities. Mike shares a counterintuitive finding: Ada had to artificially delay email responses by eight minutes because instant replies made users distrust them. | |
| Real-World Voice AI Complexity and Multilingual Capabilities | 3 | 3 | 0 | 0 | Matt asks about real-world voice AI constraints. Mike emphasizes the stark gap between smooth WebRTC web demos and low-bandwidth, noisy highway phone calls in multiple languages. | |
| Agentic Collaboration and the Automated Enterprise | 4 | 3 | 0 | 0 | Matt connects Mike's vision of customer agents talking to engineering agents to the macro thesis of the automated enterprise. Mike agrees, citing internal examples where AI feedback automatically triggers code fixes. |
Statements from this episode (29)
Murchison: Average American spends 40 days on hold in their lifetime
“Like the average American will spend 40 days of their life waiting on hold.”
Murchison: $500B spent annually on human-led customer service globally
“Globally, every year, five hundred billion dollars is invested in human-led customer service operations around the world.”
Monday.com automates 50% of customer support contact using AI
“Monday.com CEO a couple months ago in their earnings report shared that they're automating 50% of their customer contact, but also that they're talking to their customers more than they ever have before.”
Ada orchestrates seven to nine AI models per customer inquiry
“We do, we run on multiple models. The way to think about Ada is we call it our reasoning engine. Is we orchestrate on your behalf anywhere right now between, it seems like seven to nine models, depending on the inquiry, and those models work together to unders…”
Murchison: Top AI reasoning models may soon be overkill for customer service
“I think we're very quickly about to reach a tipping point whereby the most performant reasoning model available in the world may be overkill for the average customer service use case.”
Mike Murchison: LLM model costs plummeted 90% in 12 months
“And as you know, well, you know, we're seeing model costs plummet. They've plummeted 90% in the last 12 months.”
Ada does not train models from scratch on individual customer data
“But we don't train models from the ground up based on individual customer data. And because we haven't seen it, that, that yields a performance improvement yet.”
Ada's AI agents resolve up to 85% of customer service conversations autonomously
“Our agents are now autonomously resolving. We have a new high watermark as of last week, week before, 85% of all conversations that, up to 85% that they're seeing. That's without human involvement”
Building effective AI agents requires custom prompts for every individual customer
“Teams who start to build their own agents inside their companies quickly realize this, that really what you need to do is you need to assemble a custom prompt for every individual customer.”
Murchison: Ada uses LLM judges to verify AI responses against knowledge bases
“We use language models as judges to ensure that your generations are grounded in the knowledge and policies that you've that you've connected to ADA.”
Ada uses runtime guardrails to prevent inaccurate AI responses entirely
“Increasingly, we make those assurances in runtime, so it's actually not possible for ADA to deliver an inaccurate response.”
Ada measures customer conversation quality as accurately as humans across 100% of interactions
“And we can say with confidence right now that yes, that is exactly true. We can measure the quality of a conversation transcript as well as a human annotator. And we can do it with a hundred percent coverage.”
Murchison: AI agents will soon face zero technical action limitations
“I think we're very soon going to be in a position where there's really no action that a, that, that ADA can't perform on behalf of a customer. And the limiting factor will be whether or not the business wants that action to be automated.”
Murchison: Ada's voice agents send callers follow-up text instructions
“A lot of our voice AI agents now are offering to they'll often send you text instructions over text. So you'll be speaking with them. It's so annoying to talk to a voice agent or someone on the phone and ask, you know, have, you know, and like have them, you k…”
Ada has no human agents built in, routing handoffs to legacy systems
“Yeah, Ada is a, is an AI native application that has no human customer service agents in it. So whenever a human customer service agent is needed, we will route, we'll hand off, to a human customer service agent who lives in an incumbent system, like a service…”
Ada is shifting from human handoffs to AI delegating tasks to humans
“And so increasingly we're starting to think less in terms of handoff and more in terms of human delegation. And we're very excited about that because what it means is that instead of just simply handing off a conversation to a human agent and having them figur…”
Murchison: Customer service roles are evolving into managing AI agents
“I think that broadly the big shift that's happening is, is customer experience, customer service professionals are evolving from, You know, reacting to incoming customer service inquiries to proactively managing them by really serving as the manager of an AI a…”
Murchison: Deploying customer-facing AI effectively increases customer interactions
“When you deploy customer-facing AI effectively, you end up talking to your customers more.”
E-commerce client used Ada's AI reporting to launch new sunglass SKU
“We had a customer recently who they're a large e-commerce business. They discovered through our reporting that they should probably create a new product. Like, in other words, that they, in this case, it was a skew of sunglasses that they didn't have. It was p…”
Unified customer-facing AI will soon outcompete humans in sales and support
“There will just be a single customer-facing AI that most businesses deploy, and that AI will not only serve as well or better than a human, but it will sell as well or better than a human too.”
Murchison: Resolving customer issues with Ada yields five-star CSAT equivalent to humans
“What we see inside ADA is when an issue is resolved, when we resolve an issue, it is almost the same thing as a five stars, five star human reported CSAT experience.”
Murchison: Ada's empathy will eventually personalize to individual user preferences
“Over time we'll, we'll see like this dimension of ADA Like, really become personalized on an individual user basis.”
Murchison: AI software buyers should expect continuous improvement, not static consistency
“Businesses should not be purchasing, you know, customer service software in twenty-twenty-five and expecting the results to be consistent. You know, they're not buying a piece of software to solve a problem today. What they should be purchasing is an expectati…”
Murchison: Many Ada customers hang AI agent performance metrics on office walls
“Many of our customers, they literally have pictures of their AI agent's performance on the wall. And their monthly, the monthly performance of them.”
Murchison: Pre-configured AI agents drive faster customer deployment than customizable platforms
“Now we've discovered that we can just accelerate results far faster by Giving you an agent in a box that is great and then just ensuring that you know how to coach it to improve over time. That's been a big unlock for our customers.”
Murchison: Ada added artificial email delays because instant AI replies had lower open rates
“So, like, you know, the, in other words, like, customers weren't opening these emails because they didn't believe that the email could be helpful if it came so instantly, right? So it was like an eight-minute delay that a lot of our customers use.”
Murchison: Voice AI latency over 3.5 seconds creates a frustrating user experience
“You know, if you're above three and a half seconds or four seconds latency, it is, it's a frustrating experience. Doesn't matter how capable you are. If you're not fast enough in over voice, it's painful.”