When Alyssa Nolte compared today’s AI implementation to the failed promises of health scores, I felt that in my bones.
As customer success teams, we’ve been here before. A decade ago, we were sold a dream: health scores would magically predict churn, flag at-risk customers, and make customer loss “a thing of the past.” It turned out to be a bunch of empty promises that left teams disappointed and skeptical.
Now we’re standing at the same crossroads with AI. And if we’re not careful, we’re going to make the same mistakes, just faster and at greater scale.
Alyssa Nolte is a serial entrepreneur, CEO of truvue, and host of The Growth Signal podcast, where she explores what it means to build customer relationships when expectations change constantly and loyalty is at an all-time low. In our recent conversation, we dug into how customer success teams can adopt AI without sacrificing the human connection that customers are increasingly demanding.
The Risk of Efficiency Over Relationship
Here’s the uncomfortable truth: every time we implement technology for efficiency’s sake at the expense of customer relationships, we damage trust.
Remember phone trees? Those “press 1 for this, press 2 for that” systems that made everyone scream “talk to a person!” into their phones? There’s literally a website, gethuman.com, dedicated to helping people bypass those efficiency-optimized nightmares.
AI risks becoming the new phone tree.
Alyssa put it perfectly: “I worry that we’re at the same inflection point when it comes to bringing AI into our business and particularly bringing AI into any kind of touch point with our customers.”
Recent studies confirm this concern. Consumers are increasingly looking for alternatives to AI interactions. They want to speak with real humans. And yet, organizations keep implementing AI solutions that prioritize operational efficiency over customer experience.
The question is how to use AI without repeating history’s mistakes.
The Unglamorous Truth: AI Needs Better Data Than You Have
Before you can do anything meaningful with AI, you need to confront an uncomfortable reality about your data.
“How many of us trust anything that we see in our CRMs?” Alyssa asked. “I don’t trust any of it because I know that it’s outdated or it’s something that somebody dropped in simply to bypass a required field.”
This is the unsexy, unglamorous work that most organizations are ignoring in their rush to implement AI. Everyone wants to talk about the exciting possibilities of AI-powered customer success. Nobody wants to talk about data hygiene.
But here’s the reality: You’re connecting your AI to CRMs filled with data that you, as a human, don’t even believe. And you expect it to do something magical?
AI needs three things from your data:
- Clean data: Data that’s accurate and current, not artifacts of workarounds
- Good data: Data that’s relevant and meaningful, not just fields filled to satisfy requirements
- Useful data: Data that actually connects to the problems you’re trying to solve
As Alyssa emphasized, “We have to go back a little bit to the fundamentals because for AI to be successful, there’s a lot of fundamental ground level work that organizations haven’t done.”
This isn’t glamorous. It won’t make you look innovative at your next board meeting. But without this foundation, your AI initiatives will fail.
Start Small, Get Specific, Avoid Scope Creep
When I work with organizations on AI adoption, I see the same pattern repeatedly: teams want to solve everything at once.
Alyssa’s advice is refreshingly direct: “You first have to set a very clear and specific use case. My AI will do this. And whatever that is, don’t scope creep yourself.”
Treat your AI implementation like you would treat a client project:
- Define one specific outcome
- Reverse engineer what data points you need
- Get that quick win
- Then build the flywheel to add more capabilities
She shared a story about a client who said, “We’re gonna roll out AI and then we’ll figure out how to make it do the things we want it to do.” Her response? “You’re gonna fail. That’s not gonna work.”
This aligns with what I’ve been saying about technology adoption: Customer first, technology last. And maybe AI isn’t even the answer.
Before you implement any AI solution, ask:
- What specific customer problem are we solving?
- How will this impact our business outcomes?
- Could we solve this with simpler technology for less cost and faster deployment?
Don’t start with “how can we use AI?” Start with “what problem do we need to solve?” Then determine if AI is actually the right solution.
The 8% Rule: Test Before You Scale
When I mentioned Experian’s approach, rolling out their AI chatbot to just 8% of customers initially, Alyssa raised a critical point about selection bias.
How you choose that test group matters enormously:
- Your biggest fans will give you grace and inflate your success metrics
- Friendlies will forgive issues that general customers won’t tolerate
- Random selection gives you realistic feedback
- At-risk accounts test AI’s capability to handle difficult situations
Alyssa shared a cautionary tale about a major CRM that consulted only with their biggest fans during development. Those superfans understood the ecosystem, gave tons of grace, and provided positive feedback. But when they rolled it out to the general public? The reception was entirely different.
The general public doesn’t give you grace. They don’t have 10 years of relationship history to soften their judgment. They evaluate your AI based on their immediate experience.
This principle extends beyond AI testing. When you’re developing any customer-facing solution, whether it’s a brand strategy, a new product, or a messaging campaign, don’t just ask your mom if it’s a good idea. Go get unfiltered opinions from people who will actually challenge you.
The Overwhelm of Possibility
The biggest challenge facing customer success teams trying to adopt AI? Sheer overwhelm.
“They are being told by everyone with a C in their title or a board of directors you have to implement AI,” Alyssa explained. “It’s the great thing about everything that we’re doing with AI… it is a total blue ocean of opportunity.”
This reminds me of the early internet era. Everyone knew they needed a website, but nobody really knew what made a good one. We got a lot of crappy websites and terrible user experiences because organizations felt pressure to do something, anything, online.
We’re at that same inflection point with AI.
The solution is to do less, but do it better:
- Determine your Minimal Valuable Business Outcome
- Focus on one high-impact use case
- Set realistic expectations
- Deliver actual value before expanding scope
- Measure success against customer outcomes, not just operational efficiency
Under-Promise, Over-Deliver (Especially with AI)
Here’s where framing becomes critical.
If you walk in saying “this AI will solve all my problems” and it solves 50% of them, you’re disappointed. But if you say “this will solve 25% of my problems” and it solves 50%? You’re thrilled.
Same outcome. Different expectations. Vastly different perception of success.
“You have to make promises that you can keep,” Alyssa said. “And the smaller the better, because you wanna always under promise and over deliver. And I fear we are over promising right now.”
I’ll go even further: What if your AI only solved ONE of your problems? If it genuinely solved one real problem, isn’t that success?
Alyssa shared a perfect example. She uses an AI note-taking tool that connects pages in her notebook, surfacing related content she wrote months ago. “That’s so helpful,” she said. “And that immediately was a positive experience for me.”
It didn’t promise to eliminate note-taking. It didn’t claim to read her mind. It solved one specific problem: helping her find connections across her notes. That’s enough.
The AI industry is over-promising right now. As customer success leaders, we need to counter that by setting realistic expectations, internally with our executives, and externally with our customers.
Breaking Down Silos: AI’s Killer Application
For 15 years, organizations have been trying to break down data silos. We’ve failed, consistently.
AI might finally give us the tool to succeed.
Imagine you’re taking notes on a customer expressing frustration about a problem. On the right side of your screen, AI surfaces: “A customer in a different industry had this same problem. Here was their solution.”
Suddenly, knowledge that was locked in someone else’s notes, in a different team’s database, in a conversation you weren’t part of, becomes accessible exactly when you need it.
“It doesn’t require me to know everything about every customer that is at our company or has ever been at our company in order to be successful in the moment,” Alyssa explained.
This is where AI becomes genuinely transformative. Not by replacing human judgment, but by making collective knowledge accessible.
Think about how this changes customer success:
- Sales teams get instant access to support insights during pitches
- Product teams see patterns across customer feedback from multiple sources
- Support can leverage solutions discovered by customer success
- Everyone can tap into institutional knowledge without needing years of tenure
But, and this is critical, this only works if you’ve done that unglamorous foundational work. If your data is siloed, inconsistent, or unreliable, AI will just surface garbage faster.
The Personalization Challenge
Here’s something most people miss about AI adoption: the models get better the more you use them.
Alyssa has been using ChatGPT since before they had a premium version. Her ChatGPT knows her writing style, her priorities, her communication patterns. When she tests other AI platforms, they can’t compete, not because they’re technically inferior, but because they haven’t been trained on her specific needs.
“My ChatGPT knows me better,” she said. “Unfortunately, because I’m so enmeshed in ChatGPT, the others can’t compare simply because they haven’t been trained on me.”
This creates a fascinating challenge for organizations. How do you:
- Train multiple AI models without redundant effort?
- Ensure consistency across different AI tools?
- Allow employees to use best-of-breed solutions while maintaining personalization?
My prediction: we’ll eventually move toward shared AI profiles, think of it like a driver’s license or professional credential that AI systems can access to understand your preferences, style, and context. We’re probably a couple years away from that reality.
In the meantime, invest time in training whatever AI tools you adopt. The more context you give them, the more valuable they become.
Practical Steps for Customer Success Teams
Based on this conversation with Alyssa and my own experience working with organizations on AI adoption, here’s how customer success teams should approach AI:
1. Audit your data honestly
- Can you trust what’s in your CRM?
- Are there fields being filled just to bypass requirements?
- Do you have consistent data definitions across teams?
2. Choose one specific use case
- What single problem, if solved, would have the biggest impact?
- Don’t scope creep yourself
- Get a quick win before expanding
3. Reverse engineer data requirements
- What does the AI need to know to solve this problem?
- Where does that data currently live?
- How clean and reliable is it?
4. Set realistic expectations
- Under-promise on capabilities
- Frame success around solving one problem well
- Prepare leadership for gradual improvement, not magic
5. Test thoughtfully
- Start with a small percentage of customers
- Choose your test group carefully to avoid bias
- Gather unfiltered feedback
6. Measure customer impact, not just efficiency
- Are customers getting better outcomes?
- Are you strengthening relationships or just processing tickets faster?
- Would customers choose to interact with your AI, or are you forcing them to?
The Future: Humans + AI, Not AI Instead of Humans
As we wrapped up our conversation, Alyssa shared how she uses AI daily, from search to troubleshooting to her favorite prompt: “roast this.”
She posts her writing and asks AI to tell her everything that’s wrong with it, what people will make fun of, what lacks clarity. “Sometimes when you’re too close to your own stuff, you just don’t see it for what it really is,” she said.
This is the right framing for AI in customer success. It’s not about replacing human judgment or eliminating customer conversations. It’s about becoming “Iron Man or superhuman” in your capabilities.
AI should make you better at understanding customers, faster at solving problems, and more effective at building relationships.
The customers calling for real human interaction aren’t rejecting technology. They’re rejecting bad technology that prioritizes efficiency over their needs. Give them AI that genuinely helps them, that respects their time, that maintains the human connection they’re craving—and they’ll embrace it.
We’re at an inflection point. The choice is ours.
Connect with Alyssa Nolte on LinkedIn where she’s known as “the notorious plant killer,” or check out her podcast, The Growth Signal, for more conversations about building customer relationships in the age of AI.
This conversation reinforces what I’ve been saying for years: customer first, technology last. AI is a powerful tool, but it’s just a tool. The question isn’t “how do we implement AI?” It’s “how do we better serve our customers?” Sometimes AI is the answer. Sometimes it’s not. But it always starts with the customer.