The $500K AI Readiness Question Your ERP Vendor Isn’t Answering

Sponsored by TeamCentral

I talked to Marc Johnson and Andy Park last week… Andy told me about his friend who runs a manufacturing company and is seeking some answers about why he should modernize his ERP.  

The guy’s been running Epicor from a server closet at his plant for years. Products ship on time. Processes work. The system does exactly what he needs it to do.

Then the vendor starts hounding him to migrate to their cloud version. The cost? Three times his current annual spend.

Here’s the kicker though… There’s no direct migration path. It’s not an upgrade. It’s a full reimplementation. New system, new risks, and you know the stat: 75% of ERP projects are subject to failure.

So Andy’s friend is sitting there thinking, “Why would I spend half a million dollars and risk my entire operation when what I have works fine?” 

This isn’t really a story about cloud migration. There are very valid reasons to move to modern ERP, like better security patches, improved interoperability between systems, inability for vendors to support multiple platforms long-term, and yes, new recurring SaaS revenue streams.

Most mid-market companies are about to get blindsided by AI, and they don’t even know it yet

This is a story about why most mid-market companies are about to get blindsided by AI, and they don’t even know it yet terms of AI Readiness.. The problem isn’t that vendors want customers to move to the cloud for AI Readiness. The problem for AI Readiness is the timeline.

These migrations generally aren’t aligned with customer AI readiness. Companies are being forced to move on the vendor’s schedule for AI Readiness, not their own schedule, and without adequate consideration for the people, process, and budget impacts of that change.

Vendors aren’t meeting their customers where their needs are, and this creates two critical AI readiness gaps:

1. Customers who want to remain on legacy systems won’t be able to take advantage of AI in their current state, not without implementing a proper data infrastructure strategy first.
2. If they do move to the cloud without thinking strategically about data architecture, they’ll completely miss the window to position themselves for a future where AI plays a significant role in operations.

That’s the $500K question Andy’s friend is really facing. It’s not just about cloud migration costs and AI Readiness. It’s about whether he’s building the AI Readiness foundation that makes AI possible, or just delaying the inevitable while his competitors get ready.

I spent time with both Andy Park and Marc Johnson, Co-Founder) of TeamCentral. They spent almost 20 years together at a Global Consulting Firm seeing the same integration problems at every single customer before they decided to build a simpler solution. That pattern recognition matters for AI Readiness, and what they’re building could change how mid-market companies think about AI readiness. 

When I asked what they wish more companies understood about AI readiness, Andy didn’t hesitate:

Why 94% of Companies Can’t See Their Own Supply Chain

Let me give you a number that should make you uncomfortable.

Only 6% of companies have complete end-to-end visibility into their supply chain, while 62% report having only limited visibility (and not full transparency) across their operations. That’s according to the GEODIS Supply Chain Worldwide Survey.

Think about that for a second. We’re talking about the backbone of how products move from raw materials to customer delivery. And 94% of companies are flying blind when it comes to making intelligent data driven decisions.

Andy explained it to me this way, “Imagine you’re a salesperson looking at inventory in your system. You see 20 units available. A customer needs 15.

But here’s the problem. You don’t have visibility into when the next shipment arrives. You don’t know if manufacturing needs some of those units. You have no idea if another salesperson already promised them to their customer.

So what do you do? You put a hold on all 20 units.

It’s not malicious. You’re just trying to keep your customer happy. That’s literally your job.

But multiply that behavior across your entire sales team, and suddenly you’ve got inventory paralysis. Units sitting in “reserved” status that may never ship while other salespeople scramble to find stock.”

Poor data connectivity creates AI Readiness inefficiency, yeah – but it also drives otherwise rational people to make decisions that hurt the business in ways they don’t anticipate.

COVID exposed this in brutal detail. When supply chains broke down, companies without AI Readiness that couldn’t see their full value stream couldn’t respond. The ones with real visibility could reroute, adjust, and keep moving. This is exactly the kind of pain that validates a market, and it’s how TeamCentral knew they were onto something:

The Band-Aid Tax: How Quick Fixes Become Technical Debt

Here’s how most mid-market companies have built their integration architecture over the last 20 years:

System A needs to talk to System B. IT finds the cheapest, fastest way to connect them. Maybe it’s a custom script. Maybe it’s a basic API call. Maybe someone literally exports a CSV every night and imports it somewhere else.

It works. It’s not elegant, but it works.

Then System C comes along. Same process. Then System D. Then your e-commerce platform. Then your EDI feeds. Then your call center system.

Andy calls these “band-aid integrations.“ And when you have 15, 20, or 50 of them, you end up with massive technical debt wrapped up in “Spaghetti Architecture” that blocks AI Readiness.

Here’s what band-aid integrations don’t include:

• Logging when things break
• Forensic analysis to find missing data
• Redundancy for critical workflows
• Any way to test changes without breaking production

So you end up with a team of people whose job is to monitor integrations and scramble when they break. And they break constantly because they’re brittle point-to-point connections that weren’t designed with any overall architecture in mind.

The hard part is that every individual decision made sense at the time. Connect the CRM to the ERP in the most cost-effective way possible. Don’t overthink it.

But architectural debt compounds just like financial debt in any AI Readiness effort. And the interest comes due when you try to do anything sophisticated, like prepare your data for AI.

When “Easy” Integration Tools Make Things Worse

Think traditional enterprise iPaaS platforms are too complex? You’re not alone.

Companies like MuleSoft and Informatica force you into architectural rigor. They make you think about decoupling systems, testability, and proper data flows. It’s heavy and complex, but it forces you to build things the right way.

The problem is that those point-to-point integration platforms, whether they’re custom scripts or basic APIs, create massive technical debt. They work in the moment but fall apart at scale.

Then the “citizen developer” platforms showed up promising to democratize integration. “Low-code.” “No-code.” Anyone can build connections between systems. 

Platforms like Power Automate and Zapier are excellent at what they do… solving basic automation and repetitive tasks. Moving email attachments to files. Simple data syncs. They’re perfect for that.

But citizen developer solutions can’t solve complex data governance and data automation use cases.

Marc gave me an example: A mid-market manufacturer has customer data spread across seven different systems – CRM, ERP, e-commerce platform, EDI feeds, call center software, warehouse management, and their legacy AS/400 for financials.

Good luck building a clean, governed integration for that with a point-and-click tool.

The point is, you need to use the right tool for the right AI Readiness problem.

The Middleware Approach: Architecture Without the Complexity

This is exactly why TeamCentral built their platform differently.

Instead of just making integration “easy,” they built a middleware approach to no-code integration that’s far more scalable and cost-effective from a TCO perspective.

Think of it this way:

• Point-to-point tools → Fast to build, brittle, create spaghetti architecture
• Citizen developer platforms → Great for simple tasks, fail at enterprise scale
• Middleware no-code platforms → Enforce data architecture while maintaining development speed

TeamCentral forces you to think about your data model first. It’s the architectural rigor of enterprise iPaaS with the speed and accessibility of no-code development, all without the brittleness of point-to-point connections. This infrastructure-first approach is harder to build and harder to fund, but Marc frames it differently:

The Four Pillars Every AI Project Dies Without

I asked Andy and Marc what “AI ready” actually means to them from a data standpoint.

Andy laid out the AI Readiness framework: Connected, high quality, accessible, and secure.

Marc added the strategic context, “Each one of these has its own level of pain. You’re gonna get overrun by this wave if you don’t get in front of it.”

Let me break down what each of these means in practice for AI Readiness.

Pillar 1: Connected

All your systems need to speak the same language for true AI Readiness. Not JSON here and XML there and EDI somewhere else. A common business (semantic) language that translates technical data structures into something humans (and AI) can actually work with.

After 20+ years of working with ERPs, Marc and Andy knew about everything there is to know about how Oracle, Salesforce, Microsoft, and SAP model their data. Based on that knowledge, they built a common data model based on what actually works for real businesses.

Customers, vendors, orders, invoices, items, inventory – the true foundation of AI Readiness. The foundational stuff that every for-profit business needs.

In other words, instead of making you architect your perfect data model from scratch (which takes so long most companies never finish), they give you a proven starting point and let you extend it.

The platform includes over 80 pre-built, SOC 2 compliant connectors that handle thousands of data automation scenarios right out of the box. This means you’re not starting from zero, you’re starting from proven patterns that already work for companies in manufacturing, supply chain, construction, and logistics.

Pillar 2: Quality

Here’s where most AI Readiness data governance projects die: They try to define perfect data structures upfront, then spend years implementing controls that never actually get enforced.

TeamCentral’s approach is different. They deploy automated governance during synchronization.

Every time data moves between systems, business rules get applied. Deduplication happens. Validation happens. Data gets cleaned incrementally as it flows, not in one massive cleanup project that never completes.

Think of it like the medallion architecture in data warehousing. Raw data gets refined through stages until it reaches production quality. But this happens in real-time across your operational systems, not in a warehouse you query later.

TeamCentral calls this their “normalized data model with automated governance.” As data synchronizes through their embedded enterprise data model, the quality automatically increases. You’re not just moving data, you’re improving it with every transaction.

Pillar 3: Accessible

This is where natural language and AI actually matter for AI Readiness. Once your systems are connected and your data quality is solid, you need to interact with it without being a SQL expert.

Andy’s building MCP architecture that lets you use one agent experience to query data across all your connected systems.

Imagine asking Microsoft Copilot about your supply chain, and it pulls data from SAP, Oracle, Salesforce, and your warehouse management system to answer. That’s accessible.

TeamCentral’s platform is built for this hybrid infrastructure reality. It integrates and migrates data between cloud systems and legacy on-premise ERP, CRM, and WMS. Whether your data lives in a server closet or in Azure, the platform treats it all the same.

Pillar 4: Secure

None of this works without proper security and privacy controls for AI Readiness. Especially when you’re connecting multiple systems and letting AI access sensitive data.

Reality check: If you don’t have all four pillars, you’re not AI ready. And trying to build AI solutions on top of broken foundations just means you’ll automate bad processes faster.

Marc emphasized this point, “The complexity of building AI systems is already daunting. But if you don’t have the security model designed out, if you don’t have the connectivity pieces, if you don’t have frameworks in place for data governance and clean quality data, the agentic pieces will never work. That’s the AI Readiness blocking and tackling that needs to be done before you can put an LLM on top of it.”

Or as Andy put it, “We don’t want to use AI to automate bad processes and bad data. You’re just going to produce more bad data and bad processes faster.”

Start Small or Fail Big: The Incremental Governance Playbook

Andy told me about sitting with a CIO recently who was working on “AI readiness.”

The CIO’s first step toward AI Readiness? A massive data definition project. Get the entire organization to agree on what a “customer” means. Define every field. Document every standard. Build the perfect governance framework.

Sound familiar? It should, because this same project has been failing at companies for the last 20 years.

The projects take so long that by the time you’re done defining standards, business requirements have changed. So you never actually implement anything.

Here’s Team Central’s approach instead:

First, model your data holistically. Don’t think about connecting your CRM to your ERP. Think about what customer data should look like across every system.

Second, start with the smallest possible scope. Pick one specific workflow. Strip away all complexity. Make the rules as simple as you possibly can. Get that one thing working.

Third, iterate. Add the next workflow. Refine your data model. Add governance rules incrementally as you learn what actually matters.

This is particularly challenging in any AI Readiness initiative because it requires discipline to start small when everyone wants to solve everything at once. But it’s the only approach that actually ships.

Marc’s advice to customers, “Don’t worry about the end systems to start. Just model your data. Create the definition of what your data should look like. Then we’ll move into designing how to move it from one place to the next.”

This is especially critical for AI Readiness if you’re facing a legacy ERP migration. Most vendors will tell you to expect 12-18 months for a full AI Readiness reimplementation. TeamCentral’s platform delivers time-to-value in weeks (not quarters), because their customers can’t afford to have critical business processes offline for a year and a half.

Companies running GP, NAV, Sage, Epicor, SAP ECC, or JD Edwards are staring down inevitable end-of-life scenarios for AI Readiness.. TeamCentral’s no-code automation platform can streamline that data migration while keeping business-critical systems connected every step of the way.

Why “AI Ready” Still Means Different Things at Different Layers

I need to level with you about something Andy said that caught me off guard.

When I asked him about enterprise AI adoption, he told me, “Nobody really has this figured out yet. If you’re hearing people who sound like experts, there really are very few experts. Everybody wants to sound like an expert.”

He’s right, and it’s important to understand why.

Surface-level AI works great today, but it doesn’t equal AI Readiness. Using ChatGPT to draft content, scan business cards into your CRM, analyze simple datasets. These are real productivity gains using curated data that’s already in good shape.

Deep Enterprise AI is a completely different problem.

But deep Enterprise AI is a completely different problem. We’re talking about connecting legacy on-premise systems (like AS/400 financials in a server closet or datacenter) with modern cloud platforms (like Dynamics 365 and Salesforce) and manufacturing execution systems and IoT sensors on the shop floor.

Andy’s take, “We’re still like chapter 2 of 10 in AI. We’re early on.”

However, that doesn’t mean you wait. It means you focus on the foundational work that will enable AI when the technology matures.

That’s where MCP (Model Context Protocol) comes in. It’s a framework developed by Anthropic for enabling AI agents to communicate with external systems.

You can think about it this way: SAP has its own AI agent. Oracle has its own. Salesforce has Einstein. Microsoft has Copilot. Each one is built for its own tech stack, and extending them to other systems is way harder than vendors make it sound.

Most mid-market companies don’t live in a single-vendor world, which complicates AI Readiness. You’ve got SAP for financials, Salesforce for CRM, a legacy WMS in your warehouse, and manufacturing execution systems on the shop floor. Getting one vendor’s AI agent to work across all of those? That’s the problem.

TeamCentral is leveraging MCP to create a common framework for agentic AI in the enterprise. Instead of forcing you to pick one vendor’s agent and then struggle to extend it, they’re building an MCP server layer that connects all those vendor-specific agents to a single common data model and common security framework.

The result is that you can use whichever agent experience you prefer, Copilot, Einstein, whatever, to query data from any connected system.

Is building this level of AI Readiness easy? Absolutely not. As Andy said, “It’s a lot easier said and drawn on paper than it really is to build.”

But it’s the architecture that could actually make the single-pane-of-glass vision real.

Case in Point – What Happens When Your Copilot Can Talk to Every System You Have

Let me paint you a picture of what this looks like in practice.

You’re the VP of Operations at a mid-market manufacturer. It’s Monday morning. You open Microsoft Copilot and ask:

“Which orders are at risk of late delivery this week?”

Copilot pulls data from your ERP (order status), your manufacturing execution system (production delays), your supplier EDI feeds (inbound shipment delays), and your warehouse management system (inventory shortages).

It tells you: 12 orders are at risk. Three because raw materials are delayed. Five because of production bottlenecks. Four because of carrier issues.

You ask: “What’s the financial impact if we expedite shipping on the carrier delay orders?”

Copilot calculates the expedite fees, compares them to potential late delivery penalties, and tells you it’s worth it for two of the four orders.

You say: “Do it for those two. And draft emails to the other customers explaining the delay with a discount offer.”

This is what Andy means by 15-20% AI Readiness efficiency gains from integration and automation. You just made four decisions in 90 seconds that would have taken half a day of pulling reports, calling people, and doing spreadsheet math.

Team Central’s building this through their Corbi agent (stands for “Cortex of Your Business”). It includes enterprise search, task automation, and something they call Pulse, which is basically a role-specific data feed.

Think of Pulse like a social media feed for your business data. If you’re the CFO, you see progress against month-end close, profitability by line of business, aged AR compared to last quarter. You can act on it, share it, comment on it.

It’s available on mobile and desktop. Because if we’ve learned anything from consumer tech, it’s that people want to work from wherever they are.

This isn’t vaporware. It’s operational AI Readiness. TeamCentral expects, and early projects suggest, 15–20% efficiency gains from integration and automation, by eliminating manual work and by improving visibility across systems. The platform delivers rapid time-to-value because you’re not building from scratch.

Why TeamCentral Exists: Building from Midwest Infrastructure Reality

You can build anywhere, but the deepest capital pools still sit on the coasts. Marc and Andy built TeamCentral from the Midwest because that’s where their customers are. Manufacturing, distribution, construction… These aren’t coastal problems, but building outside Silicon Valley means navigating a fundamental tension:

That’s why TeamCentral’s approach is different. They’re not building for tech executives at Series C SaaS companies. They’re building for the CFO at a 50-year-old manufacturer running Epicor in a server closet who needs to modernize without betting the company.

The AI Readiness infrastructure play is harder to fund than verticalized point solutions. As Marc puts it, “We’re not selling an AI widget. We’re rebuilding the plumbing.” But plumbing is what makes everything else possible.

The Jobs AI Won’t Take (And the Ones It Will Elevate)

Here’s the question everyone’s dancing around… What happens to jobs when AI can do this much?

Andy’s take is the most grounded I’ve heard: AI eliminates low-level repetitive tasks and creates more opportunity for strategic thinking.

You’ll always need a person in the middle. AI shouldn’t make critical decisions without human review. We’ve already seen examples of what happens when companies let algorithms run unchecked, and it’s not pretty.

But here’s what changes… all the non-value-added work goes away. 

The manual data entry that slows AI Readiness. The constant monitoring of integrations. The spreadsheet reconciliations. The repetitive status emails.

That creates space for people with advanced skills to do actual strategic work. The kind of work that differentiates your business and creates competitive advantages.

The companies that get AI Readiness right will invest in hiring and training strategic thinkers. The companies that don’t will try to keep doing things the old way.

And unlike previous technology waves (e.g., big data, cloud migration), this one can’t be ignored. Andy’s words: “The people that ignore it are gonna have real problems.” Where to Actually Start

If you’ve made it this far, you’re probably thinking, “Okay, this all makes sense, but where do I actually begin?”

Start with the Four Pillars AI Readiness assessment:

• Connected: Can you easily get data from one system into another? Or is it custom development every time?
• Quality: Do you trust your data? Or are you constantly finding duplicates, missing fields, and inconsistencies?
• Accessible: Can non-technical people find the information they need? Or does everything require IT?
• Secure: Do you have proper access controls and privacy protections in place?

If you’re weak on any of those four, that’s your starting point. Not the flashy AI stuff. The boring infrastructure work that makes everything else possible.

TeamCentral has built their platform specifically to address these four pillars with minimal custom code. Their approach is to connect systems with no-code integration, normalize data through an embedded enterprise model, and layer AI-powered search and task automation on top of that foundation.

They offer an AI Readiness Guide that walks through this assessment in detail, plus resources on legacy ERP migration if you’re facing that challenge. Whether you’re in manufacturing, supply chain, construction, property management, or logistics, they’ve built industry-specific solutions with pre-built templates.

But here’s the real takeaway: the companies that win aren’t the ones with the fanciest AI. They’re the ones that did the foundational data work that everyone else skipped because it wasn’t exciting.

Marc and Andy spent 20 years seeing the same integration problems at every customer before they built Team Central. They’ve worked with Oracle, SAP, Microsoft, Salesforce, and dozens of other platforms across hundreds of implementations.

That pattern recognition matters. Because while everyone else is chasing the next AI breakthrough, they’re solving the data problems that make AI actually work.

The uncomfortable truth here though is that this requires patience and discipline. You have to be willing to start small, build incrementally, and focus on fundamentals when everyone around you is talking about agents and copilots.

But that’s exactly what separates the companies with real AI Readiness that will still be here in five years from the ones that won’t.

Lillian

P.S. I keep thinking about Andy’s friend with that Epicor system in his server closet.

He’s not wrong to resist. His system works. His products ship. Why risk a half-million-dollar implementation when you’re already profitable?

But here’s what keeps me up at night: five years from now, his competitors will have AI agents managing their entire supply chain. They’ll know about problems before they happen. They’ll optimize in real-time.

And he’ll still be manually checking if inventory is available.

The window isn’t closing because the technology is ready. It’s closing because the competitive dynamics are shifting underneath us.

Start with the boring AI Readiness infrastructure work. Your future self will thank you.

This post was sponsored by TeamCentral. Want their complete AI Readiness Guide and see how their no-code platform can help you connect legacy and cloud systems? Learn more here.