Your Stibo Environment Was Built for Launch Day
The system did exactly what it was told
I know of a company where an employee, on their way out the door, pushed a pile of terrible content through the product data system. It went live on the website, and the team spent the next stretch of time scrambling to pull it down. Nobody hacked anything. The system had outgrown the controls over who could enter data, who reviewed it, or who signed off on publication, so it published.
That environment was probably fine the day it launched. Most of the STEP environments I’m asked to look at were well built for a business that has since moved on. The data model, the workflows, and the permissions were set up for the channels, the team, and the product lines that existed at go-live. Every change after that came in as a ticket.
I spent close to 20 years at Stibo before I joined Pivotree, and I’ve been on both ends of that ticket. Go-live is when a STEP environment starts drifting away from the business it serves, and the gap only closes if someone was designated to close it.
Where STEP drifts first
Stibo now ships a STEP update every quarter, in March, June, September and December. Since June 2026, every customer has been on its continuous updates policy. The documentation is clear about who does the testing: “It is essential that all customers test their own implementations against the new updates in advance.”
That’s a sensible policy for the platform. For a team that owns a heavily configured environment, it means a regression cycle four times a year on Stibo’s calendar, whether or not anyone has the hours booked. This tends to be a significant drain on internal resources, so most teams end up finding out whether anything breaks in production.
Business rules are the next place to look. Stibo’s own good-practices guide warns that rules “can have a significant impact on functionality and performance,” and that splitting an oversized library into smaller ones “does not resolve the issue.”
Integrations come last, and they cost the most. A reviewer on Gartner Peer Insights put it plainly: “Every downstream integration is a project rather than configuration.” The same review says changes “need specialists and take time,” and that total cost of ownership “climbs once you count the specialist resourcing around it.” That cost rarely makes the go-live budget, because the channels that cause it haven’t been announced yet.
One of our recent deployments had more than 40 integrations. The customer was overhauling every system they run, stem to stern, to get their product data ready for AI, so all of those connections were on the table at once. If every downstream integration is a project, as that Gartner reviewer says, that’s more than 40 projects inside a single program, and the budget reflects it. That customer could see the bill because they planned for it. In most STEP environments, integrations are built one channel at a time after go-live, each one handled as another ticket, and the total only shows up when a program like this one forces the count.
Your next customer talks like Spock
Selling to a person is like talking to Scotty. He’ll work with a vague spec, fill in the gaps and get the engine running. An AI agent shopping on someone’s behalf is Spock. It wants every attribute filled, every unit consistent, and the same answer in your STEP environment, on your website and in your distributor’s feed. When it finds a conflict, it moves on to a product it can trust.
The environment you configured at go-live wasn’t built for Spock. Most of them were built to push product data out to a website and a few channel partners. Over the next few years, I expect more of that traffic to run the other way: distributors and agents connecting to your product data and pulling the slice they need, with no aggregator reformatting it in between. That only works if the slice they can see is complete and correct the moment they ask for it.
Getting there is the first bar. Your competitors will clear it too, and then complete and accurate stops being enough. Your data has to make a better case to the agent than theirs does, and that bar keeps moving with every new channel and every quarterly update.
Twenty years ago, plenty of manufacturers decided e-commerce could wait. If you think that was a mistake, it’s a good time to take agentic commerce seriously.
AI checks the humans, too
A year ago, almost everything in a STEP support engagement was done by hand. A process failed, someone opened the logs, and the hours started. Now we run AI through every stage, including the support work that happens after go-live.
When something breaks, the failure goes into a platform that already holds the context of that customer’s STEP configuration. It reads the logs, points to the likely cause and suggests a fix before an expert opens the ticket. We’ve seen root-cause diagnosis run up to 87% faster that way. The expert still makes the call, they just start from an answer instead of a hard to decipher log file.
Testing works the same way. When a quarterly update lands, automated tests run against your configuration, and 40 to 60% of the failures they raise get fixed without anyone touching them. People review what’s left.
People usually picture humans checking AI’s work, and that’s part of it. I wouldn’t let an agent loose on product data without a person signing off, and your legal team wouldn’t either. It also runs the other direction. We use AI to review data that people entered and flag the records that need another look. Your workflow has gates, the way a racehorse goes through a series of gates on its way to the track. AI is good at checking each gate, and a person decides what happens at the ones that fail.
The trap is over-engineering it. I’ve seen teams try to solve every edge case inside the workflow until nobody can follow it. Keep the gates few and well defined, let the checks run automatically, and send the exceptions to someone who knows the system.
That last part still matters most. Every customer is unique, just like everybody else. A team with the scars and experience that comes with 100+ STEP implementations should have strong opinions about how yours ought to run, and then adjust to what you need.
Stop paying by the hour for a moving target
Hourly support made sense when every fix took a predictable amount of manual work. It makes less sense when AI does the log-reading and the regression testing, and when a quarterly update guarantees a certain amount of work whether you planned for it or not. You end up paying for effort, and the bill goes up in the months when something goes wrong.
On our implementation projects, we moved to fixed bids. Unless your scope changes, the price we give you at the start is the price you pay. Stibo Intelligent Managed Services applies the same idea to life after go-live. You pay a fixed monthly subscription at the appropriate tier. We monitor, fix, enhance, and advise you on your STEP environment, safeguarding your platform investment while driving continuous optimization.
Each tier caps how many requests are active at once. That sounds like a limit, and it is one on purpose. On projects, I’ve watched teams get so excited about what the platform can do that everybody wants everything, and nothing reaches the finish line. A short queue forces the question of what matters most this month. Those requests move through faster, with routine enhancements delivered up to 55% faster than traditional manual work.
The AI tooling we described above comes in every tier at no extra cost, along with a recurring Stibo Intelligent Health Check. The Health Check audits architecture, configuration, governance, localization, integrations, security and upgrade readiness, so drift gets caught in a review instead of in production.
Start with what’s in there now
What the company in my opening story needed was someone checking who could publish what, years after go-live, before a departing employee found the gap.
That’s where we start with every STEP customer. We offer a free, one-time environment assessment: a full configuration review, no cost and no commitment. If you decide to work with us, the findings help inform the next steps. If you don’t, you keep the findings anyway.
If you’d like one, or you just want to compare notes on the next quarterly update, email me at willem@pivotree.com. I’m easy to find on LinkedIn, too.
Learn how per-integration pricing works, and what it could look like for your stack.
