Legacy Systems & AI: How to Teach Old Dogs Very Impressive New Tricks and the hardest part OF AI INTEGRATION ISNT TECHNICAL – ITS CULTURAL! 

Here’s a conversation I have at least twice a week:

Leader: “We need to do AI. Everyone’s doing AI. But our systems were built when floppy disks were cutting-edge technology.”

Me: “Perfect. That’s actually easier than you think.”

Leader: visibly sceptical “Really?”

Me: “Really. You don’t need to blow up your legacy systems. You need to stop thinking of them as embarrassing relics and start thinking of them as reliable veterans who just need a really good personal assistant.”

Let me explain.

Your Legacy System Isn’t the Problem (Shocking, I Know)

I’ve worked with organisations running systems written in programming languages so old that the only people who understand them are either retired, semi-retired, or thinking about retirement whilst muttering about “kids these days.”

Here’s what I’ve learnt: those ancient systems are actually brilliant at what they do. They process billions of transactions without breaking a sweat. They’ve survived decades of patches, updates, and that one time Dave from IT “thought he’d try something.” They’re the cockroaches of the technology world—indestructible and oddly admirable.

The problem isn’t the system. It’s that they speak COBOL or such like and your customers speak smartphone.

The solution? Stop trying to replace them. Add an AI translator.

Think of AI as that impossibly multilingual friend who can order dinner in eight languages, charm border guards, and somehow always knows which train to catch. AI sits between your legacy system and the modern world, translating both ways without either side needing to change.

Cost of replacing the mainframe: Eye-watering, business-disrupting, career-limiting.

Cost of the AI overlay: Manageable, business-enhancing, promotion-worthy.

I know which one I’d choose.

Start Where You Won’t Break Anything (The “Try Before You Cry” Strategy)

When I ask leaders where their teams waste the most time, I get remarkably similar answers:

• “Reading and processing documents that all say basically the same thing”
• “Copy-pasting data between systems because they don’t talk to each other”
• “Generating reports nobody reads but compliance requires”
• “Answering the same customer questions 47 times a day”

Congratulations! You’ve just identified your AI starting point.

Document processing? AI can read loan applications, extract information, and flag issues faster than your best analyst after three espressos. And it doesn’t need bathroom breaks.

System integration? AI can watch what humans do on those legacy green screens, learn the patterns, and automate the tedious stuff. (Yes, AI that literally watches screens and clicks buttons. It’s like having the world’s most dedicated intern who never complains.)

Compliance reporting? AI can monitor for regulatory changes, map them to your data, and generate reports whilst your compliance team focuses on actual risk management instead of spreadsheet formatting.

Customer service? AI chatbots that actually understand context and can query your legacy systems for real information. Not the awful “I’m sorry, I didn’t understand that. Please try again” bots we all hate.

The beauty? None of these touch your core systems. They sit politely on top, making life better without making your CTO break out in stress hives.

The “Shadow Mode” Trick (Or: How to Make Sceptics Into Believers)

Here’s my favourite strategy for risk-averse organisations: run AI in parallel with humans for months, then compare the results.

I worked with an insurance company whose claims adjusters were—let’s be diplomatic—not exactly enthusiastic about AI. So we ran the AI system silently in the background for six months. Every claim that humans processed, AI also processed. We compared. We documented. We built a spreadsheet that would make statisticians weep with joy.

Turns out: AI caught fraud patterns humans missed, processed straightforward claims in seconds instead of days, and made fewer errors on data entry.

Also turns out: AI was rubbish at understanding when customers were genuinely in distress versus trying to game the system, couldn’t handle the truly weird edge cases, and would have made some spectacular mistakes without human oversight.

The result? Adjusters went from “over my dead body” to “can I have this for every claim?” once they realised AI eliminated the tedious bits and let them focus on the interesting, judgement-requiring parts.

Lesson: People don’t resist AI because they’re Luddites. They resist it because they don’t trust it yet. Shadow mode builds trust through evidence, not PowerPoint presentations about “transformation.”

The Bit Nobody Talks About: It’s Not Really About Technology

Plot twist: the hardest part of AI integration isn’t technical. It’s cultural.

Your 25-year veteran analyst who can spot dodgy transactions by instinct? She doesn’t care about machine learning algorithms. She cares about:

• Will this make her look incompetent?
• Will she lose her job?
• Will she have to learn programming? (Spoiler: No)
• Can she still use her expertise and judgement?

I coach leaders to stop talking about AI replacing people and start talking about AI eliminating the soul-crushing parts of jobs nobody likes anyway.

Nobody became a financial analyst because they dreamt of data entry and formatting reports. They did it for problem-solving, strategic thinking, and maybe impressing people at dinner parties.

AI that handles data preparation, report generation, and anomaly detection? That’s not threatening. That’s liberation.

Frame it right: “You’ll spend 70% less time on spreadsheets and 70% more time on strategy” is a very different message than “We’re implementing AI to improve efficiency” (which everyone correctly translates as “redundancies incoming”).

Also: create AI ambassadors from within your teams. Not the IT department or consultants. Your respected, slightly sceptical, analytical people who learn it properly and report back honestly. When Sarah from Risk Management says “this actually works and makes my job better,” people listen.

Making It Stick (Because Nobody Wants a Repeat of “That SharePoint Project”)

The organisations that succeed with AI don’t treat it as a project. They treat it as building a new organisational capability, like you’d build a finance function or marketing department.

What works:

• AI Centre of Excellence: Small team (5-15 people initially) who become experts, set standards, and help business units implement solutions
• Continuous learning: Training programmes for executives (strategic literacy), managers (opportunity identification), and technical staff (actual skills)
• Innovation pipeline: Business units constantly identifying opportunities, Centre of Excellence prioritising and supporting
• Governance that doesn’t kill momentum: Clear rules about ethics, risk, and approval, but not requiring 47 signatures for every experiment

What doesn’t work:

• Hiring expensive consultants to do everything then leaving when project ends
• Waiting for “perfect conditions” before starting
• Trying to do everything at once
• Technology-led initiatives with no business strategy

The investment is real but manageable: most organisations spend £5-50M on infrastructure (depending on size) and £3-20M annually on people. But the ROI is also real: 20-40% cost reduction in targeted processes, 5-15% revenue uplift from better customer experiences, and substantially reduced operational risk.

Payback period: 2-4 years typically.

Alternative (doing nothing): Watching more agile competitors eat your lunch whilst you’re still processing paperwork manually.

The Bottom Line

Your legacy systems aren’t embarrassing. They’re stable, reliable foundations that got you where you are. AI isn’t about abandoning them—it’s about making them accessible, useful, and relevant to modern needs.

My advice:

• Pick one annoying problem and solve it with AI in 6-9 months
• Run in shadow mode to build trust
• Celebrate quick wins
• Invest in people as much as technology
• Accept this is a 3-5 year journey and that’s fine
• Remember: evolution beats revolution every single time

The organisations winning with AI aren’t the ones with the newest systems. They’re the ones who figured out how to make old and new work together beautifully.

Rather like a successful marriage, really. Just with more Python and fewer arguments about whose turn it is to do the washing up.

#agile

#coaching

#AI

#WomeninTech

#LinkedInNewsUK