“Almost all the CEOs have burned their hands in ten AI POCs, two pilots, one paid project,” says Abhijeet Singh, co-founder of iaKOR. “They know things can go wrong, but they have also tasted success.”
TelcoForge recently spoke to the founders of this old-new company to understand the state of play when it comes to AI implementation. After all, it has not seemed as though ‘play’ has been very fun for many companies.
McKinsey’s State of AI 2026 survey of 1,719 respondents, published in August, found 37% attributing any EBIT impact to AI, roughly the same share as a year earlier. Only around 6% qualified as “high performers”, attributing at least 5% of EBIT to AI. An IBM study of 2,000 CEOs last year found that 25% of AI initiatives had delivered their expected return, and 16% had scaled across the enterprise.
“Creation is not a problem,” Singh says. “The problem is, can you diffuse it in the company?”
iaKOR was founded in Spain a year ago; its CEO, Gonzalo Artiach, has run AI businesses continually since 2015, collaborating with Singh to create the new brand and company. Rather than build products, iaKOR brings AI products already deployed in the US, Europe and Asia to companies through a network of technology partners, and performs the consulting, process design and roadmapping – including to telecoms players, among a variety of industries.
That offers a good deal of insight on the nuts and bolts of making a successful, scaled initiative that IBM or McKinsey would applaud. Strikingly, there’s little differentiation in approach across industries.
Vision Quest
“Almost all the companies are looking for some version of an internal ChatGPT,” Singh observed. “It’s “Give me a context layer, build a brain for me, build a decision graph for me, and then give me a tool where I can create a sales agent, an HR agent, a commercial agent.” When we start working with them, we realise the problem with this dreamland.”
That “dreamland” is the familiar telco struggle with bringing together the fundamental elements that would enable a coherent context layer to build upon. Instead, many companies are taking a different approach.
MIT’s Project NANDA found employees at more than 90% of companies using personal AI tools for work, while only 40% of those companies had bought official subscriptions.
Brazil’s Algar Telecom has 142 large-scale AI agents in production, most created by employees to optimise their own workflow, TelcoForge reported earlier this month. Algar vetted each AI tool for enterprise-level security and required every agent to sit under a positive business case. “It was a hard endeavour, but otherwise people would make these agents anyway in a Shadow AI environment,” said Enock Cabral, Algar’s head of automation and AI.
What these examples have in common is that they work within existing silos. They don’t require the kind of “dream” context layer that would unite the understanding of what is happening within an organisation.
Dreamcatchers
There are many reasons for this.
When TelcoForge looked at telco data monetisation last November, one BSS vendor said that in an average six-week project integrating a service into a telco, five weeks would be spent finding the data. Gavyn Britton of Protegrity compared enterprise data to asbestos: “You know it’s in the building, you don’t want to touch it.”
It’s the same thing in many other large enterprises. “Maybe the packet core is totally on cloud, but the circuit core runs on a mix of cloud and on-prem,” says Singh.
“In that setup, implementing an AI agent is extremely difficult. You will never get an AI-driven system that is 100% right. And management wants 100% right!”
An Analysys Mason survey of 26 operators, published in March, described the sector as “in mid-transition” on data platforms, with fragmentation and siloed operating models holding back the value of early deployments. A Cloudera-commissioned survey of 1,270 IT leaders found that only 26% believed their enterprise data was fully governed, and 60% of telecoms leaders said they could not access the data they needed for critical operational decisions.
No Bottom
When one of the largest business owners in his region recently offered to pass Artiach on to a transformation officer, Artiach declined and proposed breakfast with the owner instead. “If you don’t say it needs to be done, it will never be done,” he says. He no longer attends meetings unless a company’s owner or CEO is present.
“It’s not bottom-up adoption. It is top-down adoption,” agrees Singh. “The CEO has to believe that he needs this.”
An HTEC survey of 255 telecoms executives, published in May, found 47.5% naming a lack of executive alignment as the biggest obstacle to scaling AI.
“It’s not huge budgets that you need to transform your company,” says Artiach. “You need to be determined. You need to be audacious, you need to be brave. And those things you can borrow from someone.”
That top-down sponsorship resolves one of the big problems that hinders these transformations.
A piece of research by data specialists Ocient describes a “Red Queen effect” in which business leaders push to move quickly while IT leans towards locking data down. Having leadership set a framework in place that resolves the tension with IT is crucial. Speaking at TM Forum’s Innovate Americas, Verizon’s Martin Kessler argued that “operationalized governance isn’t a barrier, it’s how we go from pilot to production with confidence.”
Grounded
Data foundation projects in telecoms have often run for years, consuming budget and sponsor patience before delivering anything a business owner could use. While constant acquisitions, divestments and layered legacy systems make a single “corporate data lake” unfeasible for operators, masking that complexity is becoming a credible way forward.
For example, Cloudera offers automated discovery of where data sits. Chief executive Charles Sansbury described one client whose consultants had spent six months mapping data in the business. “24 hours later we gave them a map of all their data. They sacked the consultants.”
This then can be used to create an understanding of the company’s proper context, thinking and behaviour.
In Bolivia, VIVA has selected Totogi’s ontology to connect its commercial and network systems, including vendor and homegrown applications running on-premise and in Microsoft Azure, into a single model of the business without replacing them. Its first use is to base network investment on the revenue each cell site generates. “It tells us where we actually make money and why, and gives us the foundation to make decisions that grow our business,” said VIVA CEO Ryan Alvarez.
It’s this kind of fundamental which iaKOR is also working on with clients across different industries, and which they see a need for pushing down even into SMEs.
“Most of the work, to begin with, starts on the knowledge layer or context layer,” says Singh. “Whoever builds the context layer will get to build the other agents on top of it.”
Stage Coach
Again, telecoms is working with close parallels to other industries when it comes to taking a progressive approach. The NGMN’s guidance on 6G migration, covered here in September, rests on the principle that “telcos shouldn’t have to deploy equipment without a business case being established”, favouring added capabilities over a new network built in one go.
Not a crazy new business idea. iaKOR sets up roadmaps for its clients to add capabilities in a number of stages, each expected to show value before the next. “You cannot expect everything to be built at once,” Singh says. “You have to build value so that everyone in the organisation sees what the value is.”
iaKOR pairs value mapping and return analysis with prototypes built by its partners at each stage – so the hard numbers can be assessed but also other elements of impact.
“Building a prototype is a one-day job these days,” says Singh. “Don’t tell them the theory. Help them face the product at MVP stage. The moment they see it, the confidence changes by 10x.”
This kind of specific-solution approach sounds like it could as well be the bottom-up adoption we saw earlier from across silos. However, the point is not that silo solutions are bad per se; but you can’t build from them to get something holistic. Start there and you can build solutions and redesign processes in silos, but also across them.
Or, in fact, buy solutions that work for you. A study from MIT found that AI deployments bought from specialist vendors or built with partners succeeded about 67% of the time, about double the rate of internal projects.
Hidden Knowledge
One current iaKOR project is with a manufacturer that has operated for around 100 years. “Their top 10 or top 20 employees are 55, 60 years old, which means all the tribal knowledge is with them,” says Singh. Telecoms operators face a similar risk with network engineering expertise as experienced staff retire.
“If I’m the top employee and I know things that aren’t written on paper, I will never reveal it,” Singh says. “That’s human behaviour. I’m very sure I would do the same in that situation.”
A study of 5,179 customer support agents by Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that an AI assistant raised productivity by 14% on average and by 34% for novice and low-skilled workers, with minimal impact on the most experienced and skilled. The authors suggest the tool spread the practices of the most able workers to the rest. That’s not an incentive for the most skilled workers, so a bit of creativity needs to come in as an inducement to help make that implicit knowledge explicit and recorded in processes an AI can adopt.
“If I’m 55 and I can get two more hours in my day, maybe I’ll spend more time on the beach or with my grandchildren,” says Singh. “So the approach is: add more value, then onboard. Add more value, then onboard.”
Safe and Sound
That may be a human in the pool, but what of keeping humans in the loop? To what extent should we expect to see automation as the end-goal?
It’s complex, says Singh, and varies by region. While Klarna’s chief executive led an AI-first customer service strategy, he also had a famous climbdown in May 2025 noting that the company was refocusing on the “quality of human support”. In the past year Singh says, “In Europe, I still don’t see any company owner fire people because of AI.” Instead, there’s a focus on augmenting what people can do.
There’s sense in that. Wharton’s Ethan Mollick argues that leaders should set the incentives for AI use, but that much of the useful experimentation comes from employees closest to the work… who will keep it hidden if they expect their reward is to be fired.
While there is more or less appetite for reducing headcount in Asia, Europe and the US, “Human in the loop will be there in all three markets,” Singh notes. “It’s just whether it’s the whole team, 20% of the team or 50%. That’s a cultural decision.”
For those leaders who are still worried about the various ramifications of AI adoption – the human element, the data boundaries, the operational changes it will bring, the embryonic regulatory landscape – CEO Gonzalo Artiach has a heartening message.
“Not a single company is going to disappear tomorrow because they don’t apply AI,” says Artiach.
“It will disappear in three years, five, eight.”
So that’s all right then…
