How your legacy tech stack undermines the service experience

During his 20-year tenure as a BPO leader, Andy Lee saw firsthand how the biggest barrier to CX excellence was the tech that organizations implemented to support their human agents.

When artificial intelligence (AI) arrived, he took the opportunity to develop solutions that could address the problem head on. 

Now CEO and co-founder of Crescendo, Lee says the ceiling on AI in a service experience isn't dictated by the model, but the architecture.

In this interview with CX Network, Lee explains exactly what that means for customer service leaders and also covers: 

• How to distinguish which CX vendors are truly agentic
• His outlook for the BPO industry
• What an AI-native contact center means for human agents 

Don't miss any news, updates or insider tips from CX Network by getting them delivered to your inbox. Sign up to our newsletter and join our community of experts. 

Learn More

CX Network: You spent more than 20 years running Alorica. What did you see there that convinced you to start Crescendo?

Andy Lee: I had devoted my entire career helping clients solve a specific problem: every organization wants happy customers, but often, their fragmented tech stack was preventing consistent delivery. 

As a BPO, Alorica helped clients improve how they delivered CX. We helped them select and implement technology, but our core business was labor: providing it, optimizing it, and improving its performance.

That was valuable work, but it didn't fix the underlying problem. So, I stepped back to rethink the entire technical architecture beneath CX, especially for large and mid-sized enterprises.

Then AI arrived. I'd been watching it and thinking about its impact on customer service for years before it officially hit the scene. Early bots could answer questions, but they couldn't grasp intent. LLMs changed that overnight, and suddenly the possibilities were wide open.

Now I had a problem I'd spent my career trying to solve, and a technology that could finally solve it – at its core. 

That combination, and my passion behind it, led me to build Crescendo's CXP.

CX Network: Would you say that building a BPO such as Alorica, made you more or less bullish about AI replacing human work in customer service?

Andy Lee: AI will make the teams in the customer service loop more valuable than ever. Today, agents spend much of their day on repetitive keyboard work: typing, categorizing, selecting, clicking through screens. AI will take that over, so people can focus on what they do best.

I see agents moving into a smarter role.

They'll oversee the AI as it works, step in when something goes wrong, and add a human touch where it matters most. Their work becomes faster and more effective because their time goes to judgment, empathy, and problem-solving, not data entry.

In the long run, AI clears away the mundane so the humans in the loop can think strategically, evaluate interactions as they happen, and make every customer conversation better.

CX Network: Crescendo's argument is that the ceiling on AI in CX isn't the model, it's in the architecture. Can you explain that for CX leaders who aren't as technical?

Andy Lee: Right now, many companies are buying AI solutions and bolting them onto their legacy tech stack, where the AI has to get its information from.

It might include a CCaaS platform, an IVR, a ticketing system, a QA tool, a knowledge base, workforce management, voice of the customer, and the list goes on. A customer calls in, goes through the IVR, and lands on the CCaaS platform.

Some of those calls are recorded in a QA system. While the agent is talking, they're categorizing the call with a mouse and keyboard in the ticketing system. Email and chat often run through yet another workflow. Every one of those workflows has to be remembered, managed, updated, and maintained. It's a lot.

Now bolt AI on top and ask it to pull together a customer's history. It's difficult, and important context gets missed.

 Our view at Crescendo is that, even if you have the best possible AI, if the technical architecture underneath is a mess, then the AI won't be able to perform. 

When we first deployed our concierge agent to handle front-end interactions on top of clients' CCaaS and ticketing systems, resolution rates went up. Over time, they slipped. The business had changed, but no one had updated the knowledge base or workflows, so the AI's performance declined with them. 

That's why we say the ceiling for AI in CX isn't the LLM. It's the legacy infrastructure underneath it, and until that's fixed, greatness stays out of reach.

CX Network: According to our own research, integration with existing tools is one of the top three obstacles to CX investment obstacles. Is replacing the whole stack the answer, or is that too big a leap for most organizations?

Andy Lee: Replacing the whole stack at once can feel daunting. I would advise CX leaders to start with a clear picture of the end-state architecture that you want to have at the end of your journey, rather than continually adding point solutions into an existing stack, which layers on more complexity.

Anything that requires more manpower for maintenance allows for more gaps when capturing interactions and context.  Look at what you want to accomplish and think about what architecture would be most efficient and effective delivering that. 

The way leaders think about their tech stack is rooted in how software is built and managed today. To begin to understand the architecture of the future, leaders need to know how agentic works and how it's different from software. 

They need to look at the benefits of one versus the other and ensure that they're making investments towards a firm end goal. 

CX Network: There's a lot of talk in the community about how to tell what is actually agentic. What would you say to CX leaders trying to decode the vendor market?

Andy Lee: It's so noisy, and everyone's saying the same things. You could take everybody's product descriptions and scratch one name off and put it on the other. They look the same, they sound the same, but the reality is that they're not the same.

When it comes to AI, there are two philosophies: The first is more prescriptive and often based on decision trees: an "if [x] then [y]" approach. This was the first generation of AI. As LLMs came out, vendors began trying to box in the AI with logic, which is still essentially an "if [x] then [y]" approach because it's still prescriptive.

Then there's the third wave, like Crescendo, where we build the models to think. We start with guardrails, but we give the models more space to be more effective and feel more conversational. 

CX Network: What's the hardest type of customer contact for an AI to get right today?

Andy Lee: Anything where AI is dependent on back-end systems that are either not integrated or are missing information. Let's look at two examples:

Anything which is client-specific where the information isn't readily available in the knowledge base, or an issue doesn't frequently happen. For example, if an issue is recurring on our platform, we will solve it. If there's no policy or knowledge associated with it, we'll create that with a human.

Another example is when AI requires information from other systems – let's say an order management system or a field service system – if there isn't integration between those systems, questions may be hard to answer. Even where there is integration, if the data is incorrect, that's going to be hard, too. 

CX Network: Crescendo talks about AI taking a bigger operational role without reducing human oversight. What does the human role look like in a fully AI-native contact center?

Andy Lee: There will be a lot less clicking, typing and copy and pasting because AI will do this. Human agents will be more focused on observation and finding opportunities to join the conversation. From a managerial level, it will be less about compliance and reporting, with more time for coaching. 

Analyst roles will change, too. As an analyst, you will spend less time trying to get to data, you will spend less time trying to clean data, and you will spend less time trying to analyze data.

Your time will be spent looking at the output of the analysis that AI is doing.

Across every role, people will be freed to do their best work, and the result is better outcomes for the business and its customers.

CX Network: At this point in the AI maturity curve, what's your outlook for the BPO industry? 

Andy Lee: I have a lot of heart for this industry and I hope it looks very different in the years ahead.

Today, BPOs provide experienced, capable labour at the right time, in the right places, and doing the right things. The unit of measure is the human hourly rate.

I'd like to see the BPO industry move away from treating people as units of labor and start measuring outcomes. Once that happens, clients will begin to think in terms of outcomes too. BPOs will then have room to look beyond labour to the entire process, including the technology behind it, and they'll deliver far more value than headcount alone.
 
 

Quick links

• What does the future of customer insights look like in an AI-driven world?
• Why post-sales service is now your best AI search strategy
• Capita's new CX spin-off has a plan to fix customer service