Whether it is patient data trapped in siloed hospital systems or transaction logic locked in 20-year-old mainframe code, the pattern is the same: AI is finally making it practical to move critical workloads out of legacy constraints and into modern, adaptable architectures.
Every enterprise has at least one system that everyone is afraid to touch. We see it across industries — in healthcare and in financial services especially.
The pattern hiding in plain sight
In healthcare, clinical trial recruitment still runs on manual chart review. Research coordinators spend weeks combing through patient records to find candidates who match 30-50 eligibility criteria. Most trials miss their enrolment deadlines. The data exists to automate this, but it is scattered across EHR systems, locked in unstructured notes, and coded in incompatible formats.
In financial services, core transaction systems still run on mainframe codebases written decades ago. The systems work — reliably — but every change takes months. The developers who built them have retired. Licensing costs climb. And regulators increasingly expect technology estates that can adapt quickly.
Different industries. Same underlying problem: critical business logic trapped in formats that resist change.
What changed
Two things made the difference in the past 18 months.
First, AI models got good enough at reading messy, real-world data — clinical notes full of abbreviations, COBOL codebases with decades of patches — to be genuinely useful in production. Not perfect. Useful. Good enough that the output saves more time than it takes to review.
Second, organisations stopped waiting for perfect solutions. The economics shifted. The cost of not migrating — in delayed trials, in mainframe premiums, in regulatory risk — finally overtook the perceived risk of moving. We are past the tipping point where inaction is the safe choice.
Our teams are seeing this play out across engagements. The question is no longer "should we use AI to modernise?" It is "which workload do we start with?"
What the work actually looks like
AI-powered clinical trial matching means building a data pipeline that ingests patient records via FHIR APIs, parsing eligibility criteria into machine-readable logic, and running a matching engine that scores patients against those criteria. The hard part is not the model. It is the data engineering — normalising messy clinical data, handling edge cases in eligibility logic, building coordinator workflows that keep humans in the decision loop where they belong.
AI-assisted legacy migration means scanning millions of lines of old code to extract business rules, generating modern equivalents, and testing obsessively with automated differential comparisons. Again, the hard part is not the AI. It is the engineering discipline — incremental cutovers using the strangler fig pattern, equivalence testing at the field level, retraining teams who carry decades of institutional knowledge.
In both cases, AI is the accelerant. Engineering rigour is the foundation. The organisations getting real results are the ones that treat AI as a tool within a disciplined engineering programme, not as a substitute for one.
Why this matters for your roadmap
If you are a CIO or VP of Engineering at a mid-to-large enterprise, two questions are worth asking this quarter:
- Where is your organisation's most valuable data or logic locked in a format that resists change?
- What would it mean for your competitive position if you could move it — safely — in months rather than years?
These are not hypothetical questions. Organisations are answering them now. The ones moving fastest are not necessarily the ones with the biggest budgets. They are the ones that picked a specific, bounded workload and committed to production-grade engineering rather than endless proofs of concept.
The pattern is consistent: start small, prove the economics, then scale. A single therapeutic area for trial matching. A single module for mainframe migration. Get it into production. Measure the outcome. Then expand.
Where we come in
At Skillikz, this is the work we do — product engineering, data and AI pipelines, cloud modernisation. We help organisations move from "we should really do something about that system" to "it is live, it is working, and we can iterate on it."
If either of these patterns resonates with your current challenges, the detailed use cases below are worth a read.