Two-thirds of businesses say legacy systems block AI adoption. See why systems integration, not a better model, is what actually unstalls an AI rollout.
Legacy systems and AI adoption are on a collision course inside most established businesses, and it rarely looks like a collision from the inside. It looks like a pilot that worked beautifully in the demo, then quietly stopped scaling once it had to touch the CRM, the finance system, and the spreadsheet three people still maintain by hand. Leadership reads that as an AI problem. It almost never is.
TL;DR: Pega’s 2025 global research found 68% of IT decision makers say legacy systems are stopping their organisation from fully embracing AI, and businesses running fragmented systems are 30% more likely to hit AI implementation delays. Fix the plumbing first: data readiness and systems integration, not a better model, are what actually unblock a stalled AI rollout.
Where the Rollout Actually Stalls
Ask most operations leaders why their AI initiative is behind schedule and you’ll hear about the model: it hallucinates, it’s not accurate enough, the vendor needs another quarter. Dig one layer deeper and a different pattern shows up. The model can’t see clean, current data because that data lives in four systems that don’t talk to each other, one of which is a decade-old ERP nobody wants to touch. The AI tool isn’t underperforming. It’s being fed a fraction of the picture, late, and often wrong.
This is not a fringe complaint. Pega surveyed more than 500 IT decision makers globally in 2025, working with research firm Savanta, and found that 68% say legacy systems and applications are preventing their organisation from fully embracing modern technology like AI. The same research found that companies running fragmented or legacy-heavy environments are 30% more likely to experience delays getting AI projects live. Nearly half of respondents said they can’t retire their legacy applications even though they want to, because the business still depends on them.
The Blocker Is Data, Not the Model
Gartner has been more specific about the mechanism. In a February 2025 press release, the firm predicted that through 2026, organisations will abandon 60% of AI projects that aren’t backed by AI-ready data. A separate Gartner survey of data management leaders found 63% either lack the right data practices for AI or aren’t sure whether they have them. That uncertainty is the tell: most businesses have never actually audited whether their data is fit for what they’re asking AI to do with it.
An AI model is only as useful as what it’s allowed to see. If customer records live in one platform, order history in another, and support tickets in a third with no shared identifier between them, no amount of prompt engineering fixes that. The model will guess, and guessing at scale is worse than not automating at all.
Systems Integration Is the Unglamorous Fix
This is the part nobody wants to hear, because it’s not a new tool and it doesn’t make for an exciting pitch deck. Before AI can do useful work, the systems underneath it need to be connected, the data needs a single source of truth, and someone needs to map where information actually flows versus where the org chart assumes it flows. That’s systems integration work, not AI work, and it has to happen first.
In practice this looks like building proper API connections between core platforms instead of manual exports, standardising how customer and product records are identified across systems, and cleaning up the duplicate or stale records that accumulate over years of ad hoc fixes. None of it is glamorous. All of it is what makes an AI tool trustworthy enough to actually run in production rather than stay in pilot purgatory.
What an Audit Actually Reveals
A systems audit isn’t a compliance exercise, it’s a map. It shows exactly where data breaks between systems, which integrations are held together with manual workarounds, and which legacy platforms are quietly absorbing hours of staff time every week just to keep functioning. Businesses that run this exercise before their next AI initiative tend to scope smaller, faster projects with a real chance of shipping, because they know upfront what the AI can and can’t reliably see.
Skipping that step is how a six-week AI pilot turns into a six-month integration project nobody budgeted for.
The Fix Comes Before the Rollout
The uncomfortable truth in the Pega and Gartner numbers is the same: most AI rollouts don’t fail because the technology is immature, they fail because the business tried to bolt AI onto systems that were never built to share data cleanly. Legacy systems and AI adoption will keep working against each other until the integration problem gets solved on its own timeline, ahead of the AI project rather than discovered halfway through it. Businesses that treat the audit as step one, not an afterthought, are the ones whose AI initiatives actually reach production.
If your AI rollout has stalled and you suspect the real issue is buried in your stack rather than your AI vendor, a systems integration audit is the place to start. It tells you exactly what’s blocking you before you spend another quarter guessing.
Frequently Asked Questions
Why do AI projects stall even after a successful pilot?
Pilots usually run on clean, hand-picked sample data. Production requires pulling live data from multiple real systems, and if those systems are fragmented or poorly integrated, the AI tool can’t get consistent, accurate input at scale, so the project stalls.
What percentage of businesses say legacy systems are blocking AI adoption?
Pega’s 2025 global research, conducted with Savanta across more than 500 IT decision makers, found that 68% say legacy systems and applications are preventing their organisation from fully embracing AI and other modern technology.
Is bad data or old software the bigger blocker to AI adoption?
They’re connected. Old, siloed systems are usually the reason the data is inconsistent in the first place. Gartner predicts that through 2026, 60% of AI projects will be abandoned because they weren’t backed by AI-ready data, and that data problem is typically rooted in disconnected legacy systems.
What does a systems integration audit actually involve?
It maps how data currently flows (and where it breaks) between your core platforms, identifies duplicate or inconsistent records, and flags which integrations rely on manual work-arounds. The output is a prioritised list of what needs fixing before an AI project can run reliably.
Should we replace our legacy systems before starting an AI project?
Not necessarily. Full replacement is expensive and often unnecessary. Most businesses get further, faster by integrating and cleaning up what they already have so data flows properly, then layering AI on top once that foundation is solid.