AI Has Made Process Maturity A Non-Negotiable For CX Leaders

Most contact center transformation conversations in 2026 sound the same: plug generative AI into the operation, watch the metrics improve. The reality on the ground is showing up differently, with the hype outpacing operational readiness and AI often acting as a magnifying glass on existing operational debt. Advanced models struggle to navigate broken workflows when layered over bad data and fragmented communication pathways. In today's world, success in CX transformation usually depends on an organization's willingness to clean up foundational processes before the technology gets implemented.

CX consultant Greg Sullivan has spent more than 20 years in the trenches of exactly this kind of operational friction. He has worked both sides of the contact center table, running a 150-seat omnichannel operation at LifeCare/Care.com and most recently serving as a CCaaS/CPaaS presales consultant at EvolveIP, where he led technical discovery and architecture for enterprise cloud migrations. His range covers AI-enabled IVR and agent assist, omnichannel deployments, integrations with Salesforce and ServiceNow, and one of the industry's earliest omnichannel rollouts. The combined experience gives him a working view of why so many CX transformation projects stall before they deliver business outcomes.

"Whether it’s a customer calling in with a service issue, or humans doing human work and using their brains, we’re dealing with people on both sides of the interaction. Contact center work is never a light-switch solution," says Sullivan. His framing cuts against the way most technical leaders are taught to think about infrastructure, where the conversation usually starts with compute, network throughput, and system architecture. The operational friction shows up almost immediately when that mindset hits a customer service initiative, since the contact center runs on the human judgment of frontline agents who actually have to use whatever tools land on their desk. A top-down deployment built around last-minute cheat sheets often leaves frontline staff to figure things out on their own, which is how even the most sophisticated architectures stumble at the agent desk.

Inside the data junk drawer

Data architecture is where the operational debt becomes hardest to ignore. Before turning on advanced models, most organizations have to find out what they actually know and audit how their baseline records are formatted. Sullivan often invokes Conway's Law to explain why these silos persist, since a tech stack tends to mirror a company's broken communication habits more than anyone wants to admit. With massive AI funding hype flowing into the category, market pressures push vendors to overstate the speed of implementation, and buyers who map their workflows and audit their data before the RFP process tend to walk into vendor conversations with realistic expectations on timeline. "My peers in the SaaS and CCaaS world do not want to tell you the reality," Sullivan says of vendor timelines. "I used to get in a lot of trouble for saying, 'That is not a three-month project.'"

The risk of leaving frontline staff out of the design phase compounds quickly, since the people closest to the daily work see the gaps the executives writing the RFP cannot. Sullivan points to an implementation where his team thought they had the workflow buttoned up, only to spend a day with the agents and discover otherwise. The resulting cascade of questions from a single agent rewrote the project timeline on the spot. "We basically did a day-in-the-life on the system with a bunch of the agents, and one agent ruined our month because she kept asking, 'What about this? What about this? What about this?'" Sullivan recalls. "We got schooled, and we actually pushed our go-live date out a month."

Operational debt also shows up at the data layer, where enterprise-grade software gets stacked on top of records that were never formatted for it. Sullivan recalls an implementation at a liberal arts college that stalled because the CRM was built on open free-text fields, with phone numbers entered in every format imaginable. The team eventually realized that without standardized data, the customer matching the AI was supposed to do simply could not function. "They had never formatted their data in their customer records. We were literally churning through a database. One record had 2035551234, another had (203) 555-1234 so they weren't getting matches," Sullivan notes. "We all have that junk drawer in our house or the closet that stopped being truly functional years ago. Daily life just does not let you empty the thing out, reorganize it, rebuild it and make it better, even though you know that's what you need to do."

A parallel path to success

The AI wave is just the latest in a long line of technological upgrades, following similar adoption curves as natural language processing and the cloud migration before it. The fundamental rules of change management, data hygiene, and cross-functional communication still apply, and leaders who spot these historical parallels tend to navigate the migration without burning a year of project budget. At a recent industry conference, Sullivan noted a clear consensus among practitioners on how the successful AI implementations were actually built: every team that landed a working deployment had spent meaningful time removing back-end friction before the model went live. "I think a lot of people are crashing on the rocks of AI," Sullivan says. "They're thinking that it is going to be somehow different this time and they didn't learn from their prior experiences, or maybe they just weren't aware of that element to it."

For CX leaders looking for a practical path forward, the work splits cleanly into two parallel tracks that can run at the same time. The first track involves restructuring internal silos to bring end-users into the design process and executing on quick, actionable wins like password resets that build momentum without requiring the whole business to pause. The second track is the longer cleanup of macro infrastructure and data hygiene, with measuring specific business outcomes on a micro scale providing the proof points leadership needs to keep investing in the bigger work. Sullivan's read is that running both tracks at once is what separates the implementations that survive contact with reality from the ones that stall, with defining workflows before buying software the non-negotiable starting point. "If there are small wins you can pursue, great," he concludes. "And in a parallel path, you have got to work on cleaning out the closet, but also find some really quick, executable wins like password resets. Get that going because you are going to have to trial and error that too."