Ask anyone working inside a large telecoms operator where information lives and the answer is usually the same. It exists. It is just not where you need it, in the format you need it, at the moment you need it. Dashboards, runbooks, engineering notes, vendor guidance, customer history, all of it is there, somewhere, distributed across systems that were never designed to talk to each other.

Gavin Guinane, head of EMEA solution engineering at Glean, has a phrase for it. “Data-rich but knowledge-poor.” Not a lack of information, he says, but a lack of usable context. “A frontline engineer may be dealing with a network issue and still have to hunt across half a dozen tools to understand what changed, who approved it, what the last workaround was, and whether there is already a known fix.” In an industry where uptime and response times are commercial commitments, that friction carries a real cost.

It is a framing that helps explain why employee-facing AI has lagged so far behind network automation in telecoms investment, and why that gap may be closing.

Network automation has always had an easier business case to make. The starting point is well-defined: structured data, measurable outcomes, clear before-and-after metrics around provisioning speed or fault rates. Employee-facing AI is messier. It has to operate across unstructured information, inconsistent documentation, varying permission levels and the deeply human reality that every team uses language and tools differently. “If network automation is about optimising machines,” Guinane says, “employee-facing AI is about augmenting judgment. That is a bigger cultural and operational leap.”

There is a trust dimension too, and in European markets a regulatory one. Workers’ councils in several countries treat AI as a disruptive force in how employees work — which means involving those groups early is not optional if operators want deployment to proceed without delays. Guinane is direct about this: “Telcos face sovereignty challenges, so ensuring that AI respects those boundaries is also critical.” Legal sign-off on the data layer alone, he notes, can stall projects that looked straightforward on paper.

Glean works with Ericsson, among others, and Guinane’s account of what goes wrong in enterprise AI deployments is instructive. The most common failure mode is not technical. It is organisational. “AI cuts across security, data governance, IT, legal, and individual business units, so even getting started requires more coordination than many organisations first expect. If those groups are not aligned early, promising initiatives can get stuck in pilot mode.”

The other trap is scope. Large operators tend to arrive with ambitious lists of use cases. The ones that deliver fastest value, Guinane says, are almost never the headline-grabbing examples. They are narrower: a specific workflow where people are already losing time, where knowledge is fragmented, and where better access to context would clearly improve speed or quality. “The real unlock is helping experts get to the right context far more quickly so they can make better decisions.”

Governance, in Guinane’s view, is what separates enterprise AI from a proof of concept. Without permissions and access controls that employees can trust, adoption collapses regardless of the underlying capability. “If people are unsure whether the answers are based on approved sources, or whether access boundaries are being respected, adoption drops quickly. Governance is not there to slow AI down. It is what makes AI usable at scale.”

The biggest misconception operators carry into internal AI projects, he argues, is that deployment has to wait for clean, centralised, well-organised data. “If organisations wait for perfect conditions, they will wait forever.”

The more productive framing is to treat internal AI as a tool for navigating the reality that already exists, legacy systems, siloed teams, uneven documentation, rather than a reward for having solved it. In telecoms, where the pace of change in the network is constantly outrunning the pace of knowledge transfer inside the organisation, that distinction matters more than most sectors.

RELATED STORIES

Salute CEO: Investing in talent will support AI data centre growth

What is liquid cooling, and why does it matter for AI data centres?