How Embedded AI Co-Pilots Help Unlock Deel's CS/CX Teams To Own Relationships And Drive Retention

Customer success has a data problem disguised as a strategy problem. Agents juggling emails, CRM records, and call notes across a full book of business will miss things, not because they lack the instinct but because no human can hold that much context at once. The teams pulling ahead are embedding AI directly into their workflows as the infrastructure that makes human judgment possible: surfacing churn signals, flagging missed commitments, and handling the customer service transformation work that frees CSMs to focus on upstream prevention of risk rather than reacting after the damage is done.

Darren Byrne is Manager of Customer Success at Deel, a global HR and payroll platform operating across 150+ countries. Before Deel, he spent nearly six years leading CS teams at Sage across Ireland and the UK, and served as Head of SME Success at Birdie, a healthcare technology platform. With more than 12 years in customer success, he has a clear view of where AI creates leverage and where human judgment cannot be replaced.

As AI models grow more capable and more affordable, the pressure to embed them into customer-facing workflows is accelerating. For Byrne, that makes the question of where to draw the line more urgent, not easier. "AI gathers the context. CX owns the relationship and the engagement," says Byrne.

Going beyond prompt templates

The biggest obstacle to empathetic customer service is rarely indifference. It is information overload. When agents switch between emails, tickets, call notes, and CRM records throughout the day, past conversations get lost and commitments slip. AI changes that equation by pulling unstructured data from across fragmented systems into a single, actionable picture, so the human walking into the conversation already knows what matters.

But that only works if the boundary holds. Customers are increasingly attuned to automated outreach that reads like a template, and Byrne has seen the consequences firsthand. "I've engaged with businesses and I can see AI slop coming out," he explains. "It comes across as impersonal. If you're sending out communications to a client and it doesn't come across as sympathetic, that's a failure."

The end of the dropped ball

The division of labor Byrne describes has a second function beyond empathy. When CSMs carry large books of business, the sheer volume of daily context switching means promises get made and quietly forgotten. Byrne points out that AI can close that gap by tracking open commitments and surfacing them before the client has to follow up.

"Too often you have individuals dealing with multiple clients throughout the day, switching from context to context, from email to call," he notes. "We leverage AI to achieve context depth by using it not only to drive an engagement forward, but to understand where we've actually missed the ball.

Designing the AI co-pilot

The watercooler problem is real. Large language models process what lives in systems, including emails, CRM records, and call transcripts, but they cannot capture the conversation that happened in the hallway before the call, or the informal agreement made outside the platform. That gap is where human judgment remains non-negotiable, and it is why fully automated outbound engagement carries real risk. Leaders who blend human and AI strengths from the start design for this gap rather than discovering it after a customer complains.

Byrne is direct about where the line sits. "Anybody that is driving core outbound engagements, automated outbound engagements, is going to fail at some point because AI is absolutely going to make mistakes. It will conflate context, it'll hallucinate due to too much context," he says. The answer is not less AI but better-positioned AI. Byrne points to a co-pilot model built around a shared command layer where human and AI agents operate in tandem. "Having AI as a co-pilot or companion for any types of engagement is where you're going to allow for both of them to work in tandem, and still allow that customer success manager to own that judgment and ultimately own the relationship," he adds.

Benchmarks change, core principles don't

Governing AI in a CS workflow requires more than a deployment plan. Byrne notes that leaders who struggle to evaluate or course-correct their AI systems often lack a foundational grasp of how those systems actually work: what a token is, what a context window limits, and at what point a model starts to hallucinate. As expectations for AI in CX rise across the industry, that literacy gap becomes a governance gap. "For any CS leaders looking to adopt AI, I would encourage them just to understand how it works. What is a token? What is a context window? When do you get to the point where AI starts to hallucinate?" Byrne says. "You can go down the rabbit hole of understanding AI benchmarks, but you probably end up getting lost because benchmarks change on a regular basis."

Once the mechanics are understood, measuring impact becomes straightforward. Teams can move past vanity metrics and track whether headcount is plateauing as the business grows, run simple before-and-after tests on time savings, and listen to whether frontline teams are asking for more workflows. That last signal is often the most reliable. "You will hear firsthand whether a workflow is either working or whether it's not working," he notes. "If it's not working, they're going to tell you. If it is working, they're going to ask you for more workflows."

The math on repeatable work

The efficiency case eventually connects to a simpler one. When CSMs are not spending hours building QBR decks, preparing renewals, or writing call prep notes, they are talking to customers. That time is where retention happens. Byrne points to a concrete target: automate the repeatable work, and redirect the savings toward the relationships that drive revenue.

"CX and CSMs are far too expensive to be doing repeatable activities like creating a QBR deck, preparing a renewal, follow-up meetings, pre-meeting call preps," he concludes. "If you can automate 50 to 60% of that, you're naturally going to get a higher return on investment from your workforce and a workforce that's going to be spending more time with your clients, which is going to get you more revenue and more renewals."