The public and private sector know that accurate AI runs on clean contextualised data. Now they're asking who actually controls the stack they're running on, and the answer is making CIOs nervous for more reasons than one.
The pitch for AI-ready data is, by now, well-rehearsed. Clean it, model it, connect it, put it in context, and your models and agents might finally understand your organisation. But most enterprises are still nowhere near finished thanks to legacy systems, siloed stores, and poor data hygiene.
For years, a host of – predominantly American – vendors have been ready and willing to solve these problems.
But today, European tech leaders are tasked with procuring and building best-in-class technology while navigating pressure to ensure public sector tech stacks are sovereign, and while working out what ‘sovereignty’ even means...
Europe shifts to sovereign
Earlier this year, France’s domestic intelligence service ended its contract with Palantir. Germany’s intelligence agency and military have both ruled out working with the company on future projects.
The British government announced on Monday it will extend its contract with Palantir for the NHS’s Federated Data Platform (FDP), while setting the wheels in motion to procure a new system in 2029.
Prior to the announcement, the UK’s national data guardian Dr. Nicola Byrne wrote a letter to Parliament about her concerns around Palantir's contract with the NHS. Byrne reminded MPs, “The trust that underpins all our working relationships matters, and the past tells us that without sufficient trust and engagement, digital programmes struggle to achieve adoption."
On its surface, the push for alternative sovereign tech stacks looks simply like European governments swapping American technology for European equivalents. For example, in May the UK’s tax authority HMRC signed a 10-year £175 million deal with Quantexa, a British company.
But this framing misses a key point, says Quantexa’s VP for Data and AI, Dan Onions.
“Too often, sovereignty is reduced to a question of nationality. In reality, what matters is control. Can a government agency understand how the system operates, verify its decisions, intervene when required and avoid being locked into a single provider? If the answer is no, the vendor's country of origin is largely irrelevant. A domestic black box is still a black box, while a foreign-built system that is fully auditable and transferable may provide far greater sovereign control.”
The spotlight is firmly on the growing market for “decision intelligence” and contextual data analysis software that comes without reputational baggage or sovereignty concerns.
What's the alternative?
Quantexa is one such company. The UK-headquartered enterprise data analytics and AI software provider serves major banks like HSBC, BNY and ING, public sector organisations like the HMRC, the Cabinet Office, The Insolvency Service and a Five Eyes intelligence agency, where its software runs in an air-gapped environment. Sovereignty is a big part of its play.
Yet Quantexa argues that the sovereignty debate has become overly fixated on infrastructure. Where data is hosted and who owns the compute matter, but for many organisations those questions are already relatively well understood and manageable through sovereign clouds, air-gapped environments and customer-controlled hosting.
The harder problem is sovereignty over the intelligence layer itself: who controls the context, relationships, workflows and institutional knowledge that sit above the infrastructure to support mission critical decision making.
If that layer becomes locked into a single vendor's proprietary ecosystem, an organisation can find itself dependent even when the underlying infrastructure remains under its control, Onions explained.
Quantexa's more open and modular approach builds an “entity-centric” contextual layer across customers’ wealth of third-party software and data-hosting systems. Using open standards, open-source technologies and standard APIs, it lets organisations retain sovereignty over their data, technology choices and operating environment while achieving a unified, contextual view of people, places, assets and events.
“What [customers] are worried about is that if you choose the wrong option, then you're potentially giving away all of your knowledge and intelligence to other companies, or you're removing the control that you would want by going with vendors that are not sovereign,” Onions explained.
The customers most worried about this tend to be already conservative on infrastructure. “Typically, they will have their own places where they host those applications. But the challenge that they have is that they've got lots of different applications that are each holding separately fragmented data.”
Orgs need a data layer or context graph, often from a third party. And that is where procurement conversations are shifting to more open solutions.
“The risk has just kind of tipped over to something that they can't accept anymore,” Onions said. Models being switched off after a government deemed them a risk concentrated minds, he added. “I think it's really put into focus that they're not necessarily as in control of their infrastructure as they would like to be.”
So “build it” rather than buy it?
CTOs and their peers typically don’t want to move to a “build it” approach when looking for sovereign alternatives, according to Onions.
Complex interdependencies across disparate structured and unstructured data sources when running deep contextual analysis, plus the wealth of secure integrations needed with third-party systems are not things to take lightly.
Entity resolution (ER), the process of figuring out which records in a dataset refer to the same thing, and tuning that to improve data quality in a graph which shows the data relationship insights that help drive accurate decision making, is not something you can casually knock together with an AI coding assistant and then deploy into a security agency or a national health service. Organisations are looking for trusted partners to do this kind of work.
They are also scrambling to get their data AI ready, Onions said. Quantexa has done this work for some of the most heavily scrutinised and regulated sectors, including banking, healthcare and defence. These partners need to be able to tune their matching so that the data served to AI models is accurate and appropriate to the context, he adds.
The right tuning depends on the job. A Master Data Management (MDM) system, which maintains an organisation’s authoritative records, needs a low tolerance for false positives. A fraud detection system has the opposite priority. Fraudsters deliberately vary names, addresses and other details, so fuzzier matching that casts a wider net is preferable — even if it produces more false positives — provided it catches more fraud overall. It’s hard to build that kind of expertise.
Most organisations want something close to plug-and-play, deployed quickly, even if some learning and integration work comes first. But the questions a buyer should really be asking, per Onions, are about future leverage: “If you make a decision to say, ‘I’m just going to outsource the whole thing,’ then what leverage do you have to be able to change that relationship in the future? How can you easily build resilient applications on what’s been done?”
"If you have a single application that is essentially where all the data lives, and actually that's owned by the vendor, that's doing all the joining up of data, that's doing all the processing, that's also running the applications on top. You've basically got an impenetrable stack, you can’t then leverage other technologies within it."
Onions' prescription is neither to build everything nor to hand the whole estate to one supplier. "There's a sweet spot in the middle, that says where do you have a trusted vendor that solves a problem, and you buy that capability in, and you mix it with some other capabilities, and then you've got that kind of right balance of control, whereby you know in the future that if things change in your organization that you need to be able to pivot, then you have the leverage to do that."
The context layer that supports high-stakes decision making is where Quantexa plays, and Onions makes the pitch plainly. “You've got a challenge where data is very different in different sources. It was never designed to be matched. Quantexa can normalize that data to make it more consumable to support faster, more accurate and better business outcomes.”
Why trust a mid-layer vendor with that job? Track record, he argues. “Most of the largest banks in the world use Quantexa. They've not done that lightly.”
Delivered in partnership with Quantexa. For more information, read their whitepaper on AI sovereignty.