Despite rising demand for data-driven insights, a high share of mission-critical IT is either approaching or already at end of life.One 3,200 executive survey from infrastructure services provider Kyndryl found the proportion could be as high as 44%. The same polling suggested many organisations that have invested in data-infrastructure modernisation are not yet seeing a return on investment (ROI) from what can prove to be a costly exercise.Of course, for major projects of any kind it usually pays to think through priorities in advance, says Paul Henninger, partner and head of technology and data at KPMG UK.An obvious first step is to discuss and decide on key business outcomes to figure out what the organisation seeks to achieve in future. After all, that is what IT is for and if it does not assist that, any money spent is likely down the drain.Look at desired business outcomes and then decide how to proceed, says Henninger. What do you really need to fix, and how should you ensure you are ready for AI[artificial intelligence] in particular? Identify use cases and specific objectives.Data activities with value drive a specific business outcome; successful modernisation starts with business outcomes and works backwards from there. Thats true regardless of technologies and technicalities, he says.Petra Goude, global practice leader, core enterprise and cloud at Kyndryl, shares this view, while advising enterprises to triage their situation and prioritise critical changes.Dont make it all or nothing, she says. If we fail to meet an ROI, we fail. Doesnt matter if we achieve when its too expensive. Therefore, focus on business outcome.Goude notes that many organisations have regretted going all in on tech modernisations, such as when making a binary choice on cloud versus on-premise that ultimately blew budgets.Kyndryls survey also found technology outpacing training, with about 40% of surveyed leaders reporting skills gaps hindering modernisation.If youre not ready, you say modernisation can solve this, but if you dont have the future skills, it doesnt matter what you do, says Goude.Seth Ravin, CEO, president and chair at enterprise software and support services provider Rimini Street, adds that a lack of enterprise architects, data scientistsor integrators rather than programmers can also prove restrictive.Its tough to structure data in big data sets without understanding how that data is connected and structured, really understanding how to get the most out of data, Ravin says. We need people who can tie programs together using integration tool sets.When people see layoffs, only about half are typically about cost-cutting the other half is rotation for needed skills, moving people out and bringing people in with new skill sets, he adds.Once an enterprise has agreed, defined and described relevant business outcomes, then ask what data will be needed to achieve that, and how to collect, manage and control it.This way it is possible to minimise what would otherwise potentially result in an overwhelming or expensive volume of data to store, analyse and maintain.Data modernisation for data modernisations sake can have you in one of those hype cycles, Henninger adds.Often, its about acquiring a 360-degree view of the customer, yet organisations may fail to examine this data problem end-to-end. Instead, many simply add ERP, CRM or other IT solutions.For example, you might find you cannot answer a seemingly simple question about current employee numbers because when you talk to different functional stakeholders, the concept of employee varies.The number of employees for payroll purposes can be different from the number of employees for legal reasons, or the number when it comes to holiday pay, Henninger adds.Enterprises do not want to be in a position where they are trying to answer six different questions, and trying to fudge an average answer among them. That means ending up with a data set and a complicated, expensive data infrastructure built for everybody and useful for no one. That happens over and over again, Henninger says.Modernising data infrastructure is crucial partly because of the role that trust and security now play around data use in general.Partly, the artificial intelligence challenge makes it quite a lot easier to access and interrogate data sets, including potentially people you dont want, Henninger says. But on some level, the degree to which data was disorganised and trapped in documents was a natural form of security in the past.Previously, even if someone got into the network, they would still have to read the documents but this is much easier for everyone with AI, including malicious actors.The Kyndryl polling also reported that 65% of executives worry about cyber attacks, but just 30% say they feel ready to manage that risk.Organisations must be able to use their data confidently and measure the value of doing so, including identifying and setting appropriate metrics. Then when you can measure it properly, you can quantify progress or triage further intervention successfully, adds Henninger.Once an enterprise knows what data they need, who controls it and how it is maintained over time, they can start to work out the infrastructure needed for necessary analyses.Goude prescribes thinking about it as the right workload in the right platform. Revisit each application and decide what they want to do: speed up, reduce cost, or whatever. Some might not even need to be maintained.A heavily transactional system in a bank, for instance, might skyrocket costs without adding value. In that case, it can make sense to decouple the data from the transaction, perhaps moving the data elsewhere. That might in turn offer different capabilities for cloud analytics or AI.You might enhance applications without completely redoing them. Or you might reinvent business processes, Goude says. If you do one approach on everything, you likely wont optimise.Henninger says that beyond a vanishingly small number of compute-intense analytical problems, technology questions for the infrastructure side of data modernisation have largely become software questions.Its more about business intelligence (BI) than AI and advanced analytics management reporting and the resources needed and it is less about how data is stored or queried, streamed or in tables, but creating the right controls and incentives to actively manage the datasets.Problems arise despite getting the data right, because there is drift, or it degrades or the person managing it leaves. Then data is unreliable and the whole system breaks down, and the organisation goes back into silos, says Henninger.Modern data infrastructure resembles lots of other things: likely cloud-based, he says.98% of what compute is needed for decision-making is in the realm of reasonably available commodity hardware.Ravin says it is also important to retain some budget for innovation and not spend it all upgrading multiple software packages.On this point, it is important to consider all the software and its true useful life. Then start making decision on investment in automations and productivity versus upgrades or migrations.Software vendors may say the ERP has to be changed up every three to five years, but thats a work project for everybody, Ravin says. Individual usage analysis might reveal its good for much longer.A rule of thumb is to spend no more than 60-70% of annual budget on operations, and leave 30-40% for innovation. Otherwise, youre dead, Ravin adds. Costs are up. You cant sell for more because of competition, and the place that gets squeezed is profits.Gartner has estimated that perhaps 90% of budgets go on keeping the lights on, with just 10% of modernisation or innovation, he says, adding that thisis similar for resource-strapped SMEs: SMEs tend to have fewer software products, but their needs are still pretty extensive, especially if working outside their home country. The cost of admin is getting higher.He suggests reconsidering the need to be in the cloud at all, especially without bursty, elastic demand, and particularly with equipment costs for on-premise increasingly leveraged.Weve seen all these companies that were cloud-first, finding out that theyre saving millions of dollars bringing it back, he says. Cloud is not always the answer.In Kyndryls survey, 76% of businesses reported investing in AI and machine learning, but only 42% had so far seen positive ROI. Yet benefits are there to be had.Kyndryl sees potential for automated resolution of up to 30% of IT issues, up from 8%, for instance saving massively on maintenance and downtime.Data-infrastructure modernisation of an ageing estate requires carefully examining every investment choice through the lens of ROI-driving business outcomes.You could easily spend enough money to actually have just about the perfect data set, but that could be incredibly expensive, Henninger says. Sure, the data infrastructure might be worth it, but only if it solves the right problem.Read more about legacy infrastructure modernisation projectsAt AWS Re:Invent 2024, CEO Matt Garman shared details of how its GenAI technologies are helping enterprises accelerate the pace of Microsoft and VMware datacentre migrations.VMware is not going away anytime soon. While some IT leaders may be feeling the pain of Broadcoms changes, they still need to seek a long-term plan.