Jul 26, 2023 · 46m · mad
AI-First Customer Service Playbook: Ada CEO Mike Murchison on Building Scalable Support with Gen AI
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
In this episode of The MAD Podcast, Ada Co-Founder and CEO Mike Murchison joins host Matt Turck to discuss how Ada evolved into a leading AI-first customer experience platform. He details Ada's multi-model architecture, the shift from deflection to automated resolution pricing, and the organizational strategies needed to treat AI as a digital labor force.
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 14.6% of the talking time here. How this is scored →
Matt as informed peer 3.5 Guest teaching 3.5 Guest disagreement 1.3 Matt pushing back 1.0
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 24:30 Gentle correction on hot-swapping models
In an extraordinarily collaborative episode, the guest's strongest pushback is mildly reframing the host's summary of model switching from 'hot-swapping' to 'ensemble optimization'.
Hardest push from Matt ▶ 24:18 Host challenges infrastructure framingThe host actively plays back his understanding of the model layer by framing it as a hot-swapping abstraction to future-proof the stack, prompting a technical clarification.
Biggest teaching moment ▶ 24:30 Fast versus smart model architecture analogyThe guest educates the host on model orchestration by drawing parallels to cognitive science frameworks like Kahneman's Thinking Fast and Slow.
Matt holds his own ▶ 29:45 Synthesizing the demise of useless chatbotsThe host demonstrates sharp domain expertise by distilling complex product metrics into a concise thesis on how automated resolution replaces frustrating legacy chatbots.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| The Founding Story of Ada | 2 | 2 | 1 | 0 | The host provides a warm introduction, noting his status as an early investor, and asks an open-ended question about Ada's origins. The guest responds with a long, detailed monologue about working manually as a support agent to understand customer support pain points. | |
| Ada Platform Capabilities and Architecture | 3 | 3 | 1 | 1 | The host asks for an overview of the platform capabilities and briefly interrupts to ask for a concrete example of an automated action. The guest collaboratively details product features and backend integrations. | |
| Adapting to Generative AI and Team Mindset Shifts | 3 | 4 | 1 | 1 | The host contextualizes Ada's early OpenAI partnership and asks how the team adapted to recent generative AI developments. The guest shares insights on organizational mindset shifts and how non-technical staff excel at prompt empathy. | |
| Multi-Model Architecture and Fast vs. Smart Models | 5 | 4 | 2 | 2 | The host asks about multi-model infrastructure and tests a framing about 'hot-swapping' models to future-proof the business. The guest gently clarifies that it is about ensemble optimization using fast versus smart cognitive models. | |
| Shifting from Containment to Automated Resolution | 5 | 4 | 2 | 2 | The host prompts a discussion on AI-first customer service strategy and later summarizes the shift from containment to resolution as the end of ineffective chatbots. The guest agrees and expands on outcome-based pricing models across the software industry. | |
| Onboarding AI Models and Restructuring CX Teams | 3 | 4 | 1 | 1 | The host asks how customers should onboard models and restructure their support teams. The guest explains treating AI as labor and details the three-stage ACX framework using real-world enterprise examples like IKEA. | |
| Generative Replies, Actions, and Strategic Alignment | 4 | 4 | 1 | 1 | The host prompts differentiation between generative replies and actions across sophistication tiers. The guest outlines the cultural and API prerequisites needed to achieve high automated resolution rates. | |
| Expanding to Ada Voice and Omni-Channel Support | 3 | 3 | 1 | 0 | The host asks about the recent Ada Voice product launch and its differentiation from digital channels. The guest elaborates on bridging telephony with messaging channels for true omni-channel support before a warm wrap-up. |
Statements from this episode (17)
Matt Turck: Ada has over 300 customers including Meta, Verizon, and Shopify
“Today, the company has over 300 customers using the platform, including Meta, Verizon, and Shopify.”
Matt Turck: Ada raised over $190 million in VC funding
“And also ADA has raised over A hundred and ninety million in venture capital money, including a hundred and thirty million series C into them and 21 led by our friends at spark.”
Ada co-founders simultaneously worked as remote support agents for seven companies
“And of the 12, seven of them said, sure, we'll hire you. And from a dingy office on the east end of Toronto, David and I worked for seven different customer service teams remotely at the same time, living and breathing customer service for what would become th…”
Customer service agent attrition rates exceed 40% normally, 80% during pandemic
“I mean, in the enterprise, customer service attrition rates are north of 40%. In the pandemic, they got as high as 80%.”
Discovering new AI capabilities no longer requires a machine learning PhD
“I think it's actually no longer the case that you need a PhD in machine learning to understand and push the frontier of what an ML model is capable of, and I would actually say that That I actually believe that the folks who are going to discover the newest ca…”
Ada provides every employee with a custom GPT model for work tasks
“We enable everyone inside Ada with their own Ada GPT, essentially a language model that we encourage everyone in default to interfacing with any time you're trying to do any work at all, period.”
Ada's software architecture utilizes separate fast and smart LLM API calls
“At the code level in ADA, you know, we make two different large language model API calls. We make a fast call, and we make a smart call.”
Ada trained AI models to evaluate customer service quality better than humans
“Starting this year, we've now trained models, specifically an AI model, to understand the quality of a customer service conversation better than a human.”
Ada shifts pricing to charge only for automatically resolved support conversations
“We're so big on this measure that not only is it the North Star measure of our entire company, but we're increasingly pricing, enabling our customers to price according to it. So we're actually only going to charge you for a conversation that's automatically r…”
AI reasoning capabilities will force software pricing from seats to value-based usage
“I actually think this is, this will be true for all of software. I think that the, because of the quality of large language models, reasoning capabilities specifically, I think we're going to see a major shift Not only away from seat-based pricing towards usag…”
While the internet made distribution free, AI makes cognition essentially free
“If the web made the cost of distribution essentially free for businesses, what AI is doing now is AI is making the cost of cognition essentially free.”
IKEA used AI savings to retrain support agents into design consultants
“Ikea did this recently in the form of reinvesting their AI savings into, I believe, design consultants that they re they retrain from former customer service agents who are calling customers to help Consult them on how to better put together, combine different…”
Ada customers will reach 100% automated customer service resolution within two years
“Which I believe we'll, we'll have customers who get there in the next, within the next two years”
Cultural barriers, not technology, block 100% automated customer service resolution
“The biggest hurdles to overcome to actually get there are actually interestingly primarily cultural. You would think it would be technological, but it's actually interestingly cultural”
Generative AI replies increase customer service resolution rates by 30 percentage points
“This is, it's resulting in customer essentially at 25 to 29, 30% Improvement, percentage point improvement in resolution rate.”