The Capstone of the Product Data Refinery Series

We have walked the refinery end to end. Eight issues. Six stations, in order, each producing a higher grade of fuel for the one downstream. If you have read every installment, you now know what the model looks like.

Knowing what it looks like is not the same as running it.

You don’t build a refinery to look at it. You build a refinery to refine, continuously, forever. The construction was preamble. The refining is the work.

This is the issue where that distinction has to land. Refining is a process, not a project. It is not something the business completes, files, and moves on from. It is a way the business operates, every day, every quarter, every reorg, every leadership change. And the call I want to make to the community is this: stop trying to finish data. Start refining.

Most organizations are still treating product data the way they treated their last ERP implementation. Get the team. Get the budget. Get it done. Move on. That instinct is the single most expensive mistake in this entire discipline, and it is the reason the capability everyone wants is the capability almost nobody actually has.

Refining is the work, not the refinery

Think about how your business runs everything else. Your sales coverage model is not a project. Nobody assembles a team to “finish sales coverage,” files the documentation, and pivots to the next thing. The model is owned, measured, refined, and run, continuously, by people whose job it is to operate it. Pricing discipline works the same way. So does supply chain strategy. So does category management. None of these are things you complete. They are how the business runs.

Product data is no different. The Refinery is the operating model. The stations are not deliverables. They are the work, sequenced and named, that the business performs on its product data the way it performs its sales motion or its supply planning.

That framing has been threaded through every issue, but it is worth restating in one place. SKU Definitions is not a cleanup. It is the intake valve, operated continuously, deciding what gets into the refinery and on what terms. Regulatory Data is not a one-time profile build. It is the gate, operated continuously, deciding what is cleared to move. Operational Data is the measurement station, captured and recaptured as products and packaging change. Fulfillment Data is the commitment standard, set against the customer and reconciled against the supplier on an ongoing basis. Marketing and Merchandising Data is the visible refinery, governed category by category as the catalog grows. And Product Relationships is the rule set on top of all of it, tuned by leadership and read live by the systems downstream.

Six stations. One operation. None of them are done.

What breaks when you treat this as a project

Watch what happens when an organization treats this as a project.

Funding gets approved for a data initiative. A team gets pulled together, often a mix of IT, operations, and whichever business owner felt the pain loudest. A cleanup plan executes. Categories get touched. Standards get drafted. Something visible improves. Six months later, the project is declared a success. The team disbands or pivots to the next priority. The cleanup gets a slide in the quarterly deck.

Eighteen months after that, the same pain shows up again, only worse, because everyone now believes the data work was done.

I have seen this pattern more times than I can count, and the version that is most painful to watch is the migration project. A new ERP, a new PIM, a new commerce platform. The organization treats the data work as part of the implementation. Cleanup happens because cleanup has to happen for the system to go live. The system goes live. Everyone celebrates. The data work is, in everyone’s mind, complete. Then the catalog grows, the team turns over, the standards drift, and within two years the new system is hosting the same chaos the old system did. The platform changed. The operating model did not.

The pilot category is another version of the same mistake. A team picks one product line, cleans it up beautifully, writes a deck about how the model worked, and then the next category never gets touched, because the people who cleaned the first one moved on, and nobody was operating the discipline that would carry it forward. The pilot was real. The operating capability was not.

The most expensive version, in my experience, is the PIM purchase that becomes a project rather than an operation. The platform is real. The investment was significant. The implementation took a year. And after go-live, nobody is responsible for operating it. The PIM sits there, full of partial data, while the organization waits for the next data quality initiative to clean it up again. The platform was bought. The operation was never staffed.

None of these are technology failures. They are operating-model failures. The refinery was treated as a project to finish, when it was always a way to work.

What good looks like when the refinery is operated

When the refinery is operated, refining stops feeling like a project and starts feeling like a process.

Every station has a named owner who would be embarrassed to be asked who is accountable for it, because the answer is sitting in their job description. The cadence is set. Standards are documented and revisited on a schedule, not when something breaks. Drift gets caught, because somebody whose job it is is watching for it.

Maturity compounds instead of resetting after every reorg. When a leader changes, the next person inherits a running operation, not an archaeology project. When a new category is launched, the model picks it up the same way it picked up the last one, because the playbook is the playbook. The question the business asks shifts entirely. The conversation is no longer “is the data done?” which has no satisfying answer. The conversation becomes “is the refinery running?” which has answers people can act on.

And the moat we named at the end of Stage 6 starts to show up in practice. Competitors can buy your PIM. They cannot buy your operating model. The discipline of running this work, year over year, with the same conviction the business runs its sales engine or its supply chain, is the part that does not replicate. It accumulates. Quietly, station by station, until one day your catalog answers questions theirs cannot, your AI agents pick your products instead of theirs, and you have an advantage that took years to build and would take them years to catch.

That is the prize. Not a clean catalog. An operating model that compounds.

The leadership mindset: operate, do not complete

The leadership decision at this point is simpler than it looks, and harder than most leaders are willing to make.

The decision is to commit to operating the refinery as a permanent capability of the business. Not as a project to be completed. Not as a budget line that ends. Not as something the IT team owns because nobody else wants it. The refinery is run the way the sales engine is run, the way the supply chain is run, the way the customer experience is run. By a business owner. With a budget. On a cadence. With metrics. Forever. (And this becomes so much more obvious within a digital environment!)

That commitment changes the questions leaders ask. The right ones are direct. Who owns each station? What does ownership mean in practice for them? What is the cadence of review? How is performance measured, in operating terms rather than completion terms? When urgency shows up, what gets protected? The wrong question, the one I hear most often, is “when will the data project be done?” That question has been quietly costing this industry money for years. It is the wrong question because it has no right answer.

I want to be very clear about one specific failure mode, because I see it everywhere. Leaders who push this work back to IT lose. Not because IT is incapable. Because product data is a business operating model, not a technical implementation. The category strategy decisions, the standard-setting, the relationship logic, the prioritization rules, these are business decisions. IT operates the infrastructure. The business operates the refinery. They work as a team, but with very different functions. Confuse the two and the refinery does not run.

Where to start: diagnose your refining

If this issue resonates and you are wondering where to start, the answer is the most honest assessment you can do of your own refinery, today, as it actually exists.

That is what the Product Data Refinery Diagnostic is for. Take it. Score each station honestly, the way you would score a category manager’s performance or a region’s quota attainment. Be specific. Identify the weakest station, not the loudest one and not the prettiest one. The weakest, because the refinery is only as strong as its slowest station, and your downstream stages are already paying the cost of whatever upstream station is producing dirty fuel.

That is your starting point. Not the cleanup. The operating model for that station. Who owns it. What the cadence is. What good looks like. How you will know it is running.

Start with one station. Ensure it is operating. Then the next. That is the work.

What comes next

The refinery is whole. Six stations, in sequence, each producing fuel the next can use. The model is named, the work is mapped, and the operating posture is on the table. What happens from here depends entirely on whether you treat this as something to file away or something to run.

I will keep writing alongside you as you do. The next stretch of The Digital Roadmap is going to push into the territory the refinery makes possible. Prioritizing Data processes. AI enablement using the refinery. Process automation, where the rules from Stage 6 stop being a diagram and start running the business. Governance, the discipline that holds the operating model together when nobody is watching. And a few other threads I am not ready to name yet.

But all of it depends on the refinery operating. This is the foundation. This is the model for the process to follow.

The refinery is yours to refine.