Everyone wants the sizzle of AI. Boards green-light pilots, vendors promise copilots, and headlines insist that machine learning will rewrite the enterprise. Yet after two and a half decades building and shipping systems inside manufacturers, retailers and public bodies across Europe and Asia, I’ve learned a simpler truth: AI cannot compensate for broken enterprise plumbing. If your ERP is fragmented, your data lineage is murky, and your processes are glued together with spreadsheets, the fanciest model will only amplify the mess.
I am an enterprise technology leader who has spent 19 years deep in the trenches of JD Edwards EnterpriseOne—designing, integrating and modernising ERP estates for Fortune-level organisations from Denmark and Sweden to Italy and Dubai. My teams have led upgrades to 64-bit platforms, reduced paper across finance, automated supplier invoices and order flows, and put real-time integrity reporting into the hands of Accounts Payable, AR and GL. We’ve replaced brittle, manual steps with orchestrations, REST APIs and event-driven integrations—because that is what allows AI to matter later.
The Mirage of “AI First”
Too many transformations start with a model and work backwards. It’s seductive—and wrong. The value shows up when you start with flow: how orders become shipments, how shipments become invoices, how exceptions are trapped, and how those exceptions feed learning loops.
Consider two examples from my recent work:
Voyage shipments, end-to-end. We connected SharePoint, JD Edwards and a specialist logistics system so that shipment planning begins where users actually work, and then moves through validated, automated steps via JDE APIs and orchestrations. Result: cleaner master data, fewer touches, real-time error handling. When we later added analytics on top, the insights were trusted because the pipeline already was.
Invoice consolidation, done properly. Instead of asking finance teams to reconcile by hand, we integrated customer reference data directly into the sales-order process, then generated consolidated invoices with minimal human intervention. Accuracy rose, cycle time fell, and the downstream analytics—cash forecasting, dispute prediction—finally meant something.
In both cases, the “AI moment” arrived after the ERP was simplified and the data contracts were made explicit.
What Real Modernisation Looks Like
When people hear “ERP modernisation,” they think “multi-year, multi-million.” Sometimes that’s true; often it isn’t. The pattern that works is thin-slice modernisation: ship improvements that punch above their weight, in a sequence that compounds.
Here are five rules we’ve applied successfully:
1. Make integrity visible. If AP, AR and GL can’t see integrity breaches in real time, they’ll create parallel systems. Build integrity reporting early; you’ll cut rework and regain trust in core data.
2. Automate the edges first. Supplier invoices, order capture, reference data sync—these look small but drive big returns. In one programme, UXOne dashboards, watchlists and CafeOne composites reduced manual hand-offs and lifted operational efficiency by roughly five percent without a single AI press release.
3. Kill paper, then kill swivel-chair. Delivery notes, statements and delinquency letters are automation fuel. Replace them with e-data, then connect systems through orchestrations and secure MFT so humans stop retyping what machines already know.
4. Standardise the contract, not the tool. We’ve retired costly platforms by unifying APIs and EDI on a single middleware while leaving fit-for-purpose ERPs alone. The contract—events, payloads, SLAs—is what gives you optionality later.
5. Measure outcomes, not activity. In a performance analytics project, an early-warning model cut maintenance activities by ~30%, reduced plant-upset-leading issues by 20%, and avoided a potential $1M loss from a control event—the kind of result CFOs notice because it speaks the language of risk and cost.
Lessons from the Field
I’ve led programmes in healthcare, heavy industry and logistics; the lesson is universal. Flow before flair. For NHS England, our remit was not to build something cool; it was to simplify data flows across directorates so leaders could see a holistic picture of COVID-19 progress. Only after that consolidation did analytics provide meaningful guidance.
At a global manufacturer, voice-enabled picking sounded futuristic. What moved the needle, however, was integrating that system cleanly with JD Edwards via BSSV and MQ XML, optimising picking logic, and introducing grouping rules that raised throughput and lowered fulfilment cost. The AI veneer mattered far less than reducing friction in the core pathway from order to ship.
And when organisations obsess about “data lakes” but ignore master data, the lake becomes a swamp. We’ve had more success by automating item creation and enrichment via orchestrations, adding guardrails and Groovy-based error handling to keep quality high from the first keystroke. Clean inputs are a strategy, not an accident.
The CIO’s Playbook for 2025
If you’re a CIO staring at an AI wish list while your ERP shows its age, here’s a pragmatic 12-month play:
Quarter 1: Ship integrity reporting across finance; publish a data contract for two critical domains (customers and items). Stand up secure MFT and event logging.
Quarter 2: Automate supplier invoices and sales-order references; expose orchestrations/REST endpoints; retire the worst spreadsheet workarounds.
Quarter 3: Thin-slice UX for high-friction roles (watchlists, CafeOne, composites). Put a lightweight product operations cadence in place—backlog, release gates, defect hygiene.
Quarter 4: Layer in predictive analytics where the process is now stable: exception forecasting, lead-time variance, maintenance risk. Measure outcomes in cost, days, and defects—not dashboards.
This is not an argument against AI. It’s a blueprint for making AI useful—grounded in the unglamorous but transformative work of enterprise integration.
The Human Side of Transformation
Technology only lands when people own it. I’ve led onsite implementations in Denmark and Sweden, run discovery workshops and training, and used agile release trains to move from “project” to “product.” The cultural shift is from “deliver and disappear” to continuous innovation and continuous delivery—a cadence where UX, security roles, orchestrations and APIs evolve together. That is how modern enterprises stay modern.
A Final Word to Boards and Ministers
Digital sovereignty and competitiveness won’t be won by press releases about AI pilots. They will be won by cleaning the pipes of the real economy: procurement, manufacturing, logistics, finance. When those flows are trustworthy and observable, AI becomes more than a demo. It becomes an advantage.
I’ve seen the cost of ignoring this—incidents that burn $10,000 at a time, outages that threaten seven figures, and teams that drown in reconciliation instead of creating value. I’ve also seen the upside: streamlined voyages, consolidated invoicing, predictive maintenance, and ERP estates that welcome AI instead of resisting it. That is the work worth funding.
About the author:Javed Yunus Mehsania is an Enterprise Technology Leader based in Mumbai with 24+ years in software and 19+ years delivering JD Edwards EnterpriseOne–centric ERP transformations and integrations for Fortune-level organisations across Europe and Asia.