Data Strategy

One Power BI Developer Is Not a Data Strategy
(And Neither Is Your New AI Tool)

Businesses are doing the same thing with data that shoppers do in the Costco TV aisle. They walk in knowing what they need, get completely overwhelmed, and either buy nothing or buy the wrong thing. Here's why that keeps happening and how to fix the order.

6 min read Data strategy Point of view

The Costco problem

Walk into Costco to buy a TV. You've got a rough idea - bigger screen, good picture, smart features. An hour later you walk out with nothing. Not because they didn't have the right TV. Because somewhere between the OLED demo, the QLED vs. Mini-LED debate, and learning that you apparently need to care about refresh rates you've never thought about, you went from confident buyer to totally lost. So you made no decision at all.

That's exactly what's happening to businesses trying to figure out their data strategy right now. They walk in knowing they need better analytics. They leave overwhelmed. Data lakehouses. AI copilots. Fabric architectures. A dozen vendors each saying theirs is the answer. The result is almost always one of two things: they do nothing, or they buy something that doesn't work because the groundwork was never laid.

The problem isn't that the tools are bad. The problem is buying the TV before you have electricity in the house.

Your data person is great. That's not the same as having a strategy.

Most manufacturers I work with have basically the same setup. There's one person, sometimes two, who handles anything data-related. They write SQL queries. They built the Power BI dashboard the ops team looks at Monday mornings. They're genuinely good at it. They're also completely maxed out.

And here's the thing - that's not their fault. The failure is treating one capable person as a substitute for an actual data strategy. What they're really doing, however skillfully, is moving numbers from one place to another. Producing outputs on top of whatever data already exists. They're not solving the master data mess that has your biggest supplier showing up as four different vendor records across your three ERPs. They're not integrating your ops data with your financial data to give you a reliable view across the business. They're not building a foundation. They're decorating a building that was never fully finished.

It works until it doesn't. And it tends to break exactly when the business is growing and needs reliable data the most.

Then the shiny object shows up

Something triggers it. A board meeting where someone mentions AI. A conference where every vendor has Copilot in their slides. A competitor who announced they're "transforming operations with AI." Whatever it is, leadership starts asking if you're falling behind.

What happens next is pretty predictable. The business starts evaluating AI tools. The analyst gets pulled into experiments they weren't hired for. Leadership hears words like data lakehouse, agentic AI, and vector databases and starts feeling exactly like they felt in Costco. The instinct is to do something. Buy something. Move. That usually ends in a pilot that underperforms, a conclusion that AI is overhyped, and a retreat back to the Monday dashboard with the numbers nobody quite trusts in column G.

A recent example, not a hypothetical

"We tried Microsoft Copilot against our Planful GL data. It was a bust."

The work that keeps getting skipped

There are three things most businesses bypass because they're unglamorous, they don't make for great conference slides, and nobody posts a LinkedIn announcement when they finish them. They're also the reason AI works at one company and completely falls apart at another - and why some dashboards get trusted while others get quietly ignored.

Data integration

Actually connecting your systems in a way that produces reliable, current, consistent data. Not a one-time export to a spreadsheet. Not a nightly batch that sometimes fails quietly. A maintained connection that reflects what is actually happening in your operations. Without this, every downstream analysis is working from data that is either stale, incomplete, or both.

Data quality and master data

Ensuring the same customer, supplier, or product is recognized as the same entity everywhere it appears. The supplier that shows up as four different vendor records across your three ERPs is not a curiosity. It is a working capital problem, a negotiating leverage problem, and a compliance problem simultaneously. Cleaning it up is not glamorous. Leaving it alone is expensive in ways that compound quietly for years.

Data enrichment and augmentation

Adding the context that makes raw records analytically useful. Not every useful attribute lives inside your ERP. Contract terms, industry classifications, external reference data, supplier performance history, customer segment data. The analytical layer and the AI sitting on top of it need that context to produce reliable outputs. Without it, you are doing math on incomplete inputs and wondering why the answers feel off.

None of this is new. These have been the prerequisites for good analytics for decades. What's changed is that skipping them used to mean your dashboards were slightly off. Now it means your AI pilot fails in front of leadership. That's a more expensive lesson.

The wrong order and the right one

The sequence most businesses follow isn't stupid. Each step on its own makes sense. Hire someone to build reports. Build dashboards on the data that's already there. Notice competitors talking about AI. Evaluate an AI tool. Discover the AI doesn't produce reliable results. Conclude AI is overhyped.

The problem isn't any individual decision. It's the order. Every tool you add on top inherits the same problems from underneath. They just show up differently.

The sequence problem
What most companies do
1
Hire an analyst or Power BI developer to produce reports on existing data
2
Build dashboards; discover the numbers sometimes disagree with each other
3
Evaluate AI tools after competitors start talking about them at conferences
4
Connect AI to existing data; outputs are unreliable or confidently wrong
5
Conclude AI is overhyped; return to the dashboard that nobody quite trusts
What actually works
1
Connect your systems with maintained integrations that produce current, reliable data
2
Normalize master data so one customer, supplier, and item exists across every system
3
Enrich the data with the context needed to make it analytically useful and AI-ready
4
Build the analytical layer on a foundation that produces consistent, trustworthy outputs
5
Add AI and new tools to a foundation that is ready for them; they actually work

Why hiring a report writer doesn't fix it

When leadership sees the data gap, the instinct is to hire. Analyst is stretched thin, so get more analysts. Dashboards aren't quite right, so hire someone better at dashboards. Makes sense on the surface.

Here's the problem. The person you'll almost certainly hire is a report writer. Someone good at building dashboards on top of whatever data already exists. That's a real skill. It doesn't fix what's underneath. Fixing what's underneath means data architecture, integration engineering, master data management, and industry-specific domain knowledge. You're not getting all of that in one hire at a reasonable salary.

The real cost isn't the salary. It's the two or three years you spend building something slightly wrong, on a foundation that wasn't designed to hold it, before you figure out the foundation needs to be rebuilt. By then the original hire has moved on and you're starting over with more debt than you started with.

A business that buys an industry-aligned data platform will almost always outperform one that tries to build it from scratch. The economics aren't close and the time to value isn't close.

I genuinely believe that. Not because it helps us at Marquis to say it - I've just watched the build path play out enough times to know what it actually costs and how rarely it produces what was intended. The engineering talent you'd need to build a real integration and mastering layer is expensive, competitive, and not particularly interested in working at a mid-market manufacturer. Buying from someone who's already built it, for your industry, on the connectors your ERP actually uses - that's not settling. It's just the smarter call.

The right division of labor

Here's the split that actually works.

The platform, the integrations, the master data cleanup, the connector maintenance as your ERPs update - buy that from people who've built it before, for companies that look like yours. An industry-aligned data platform isn't the same as a generic analytics platform. The connectors, data models, and quality rules are built for manufacturing. You start faster and cleaner than you would building it yourself, and you're not trying to hire and retain a data engineering team to maintain it.

The analysis, the business judgment, knowing what a margin trend means for a specific customer or product line - that stays with your team. Your people know your business in ways that can't be outsourced. Give them clean, reliable data and they'll do good work with it. That's the point.

Partner for the foundation. Own the analysis. That's the setup that lets you actually compete instead of spending years trying to get the data ready.

Back to Costco

What you actually need in that TV aisle isn't another sales associate. It's the friend who's been in your living room, knows what you watch, knows your budget, and has already filtered out the features you'd pay for and never use. They point at one TV and say: get this one. You buy it. You go home. It works.

That's what a data partner who knows your industry does. They've been inside the data of companies that look like yours. They know which capabilities matter for your situation and which ones look great in a demo and disappear in practice. You walk out with the right setup and you actually use it.

The businesses that'll get the most out of AI aren't the ones who started with AI. They're the ones who started with the foundation. Clean integrations, solid master data, enriched records that are actually ready to be analyzed. When AI lands on that, it works.

You don't need to be behind. You just need to be in the right order.

Marquis IQ · Industry-aligned data platform
The foundation, built for manufacturing. Already done.

Marquis IQ is an industry-aligned data platform built specifically for manufacturing companies: the ERP connectors, the master data enrichment, the integration layer, and the analytical modules. You bring the business context and the analysis. We bring the foundation that makes the analysis reliable. No report writers required.

Certified ERP connectors already built for Epicor, Dynamics, Sage, Infor, SAP, and more
Master data enrichment that creates one golden record per customer, supplier, and item
IQ modules that sit on a clean data foundation and produce reliable outputs from day one
AI-ready data structure so the tools you evaluate in 12 months actually work when you get there

FAQ

Questions we hear from manufacturing teams

The honest answers to the questions that come up most often when businesses are figuring out their data situation.

Why did our AI tool produce bad results with our ERP data?

AI tools do not fix bad data. They amplify whatever quality problems already exist. If your ERP returns inconsistent naming conventions, jagged hierarchies, or data organized by transaction type rather than by analytical question, the AI will reflect that structure in its outputs. The fix is not a better AI tool. It is data integration, master data normalization, and enrichment done before the AI is introduced. Once the foundation is correct, the same AI tools work significantly better against the same underlying questions.

Is one data analyst or Power BI developer enough for a company our size?

A single analyst can be excellent at producing reports and dashboards on top of existing data. What they cannot do alone is solve the foundational problems that make those reports unreliable: data integration across multiple systems, master data deduplication and enrichment, connector maintenance as ERPs update, and industry-specific data modeling. These require different skills and more capacity than one person can carry. The right answer is not necessarily to hire more analysts. It is to separate what belongs in-house (data analysis and business interpretation) from what should be sourced from a specialized partner (the platform, integration layer, and data quality work). That division of labor produces better outcomes at lower total cost than simply growing the analyst team.

What does "data foundation" actually mean in practical terms?

A data foundation has three components. First, reliable data integration: live connections to your source systems that pull current, trustworthy data without manual intervention. Second, master data quality: a process that ensures the same customer, supplier, or item is recognized as the same entity across every system, every time. Without this, your analytics will disagree with themselves depending on which system they pull from. Third, data enrichment: adding the context that makes raw ERP records analytically useful, including industry classifications, contract terms, external reference data, and whatever additional attributes your specific business questions require. These three things together are what allow every analytical tool on top of them, whether dashboards, reports, or AI, to produce reliable outputs.

Should we build our data platform in-house or buy one?

For most mid-market manufacturing companies, buying an industry-aligned platform produces better outcomes than building one. Building requires data engineering resources that are expensive to hire and hard to retain, a multi-year runway to get to production quality, and ongoing maintenance capacity as source systems change. Buying from a vendor who has already built the platform for companies that look like yours means you start with industry-specific connectors, data models, and quality checks already in place. The trade-off is flexibility versus time to value. If your requirements are standard for your industry, buying wins. If you have genuinely unique data requirements that no existing platform addresses, building may be warranted, but that is rarer than most companies assume when they start down that path.

How do we know if our data is ready for AI?

A simple test: ask your AI tool a question you already know the answer to from your own analysis and compare the result. If the AI returns a number that differs materially from what you know, the data context is incomplete or the underlying data has quality problems the AI cannot compensate for. The more rigorous check is to evaluate the three foundation elements before introducing AI: are your systems integrated with reliable, current data? Are your master records deduplicated so the same entity is recognized consistently? Is the data enriched with the context the AI needs to understand what it is analyzing? If the answer to any of those is no, the AI will struggle regardless of how capable the model is. Fix the foundation first. The AI can wait.

See what the right foundation looks like for your business

Bring your ERP stack and your biggest data frustration. We will show you what an industry-aligned data platform changes, and what it makes possible that was not possible before.