Share
Craig Colangelo, July 27, 2026
Get the Best AI Solution
for Your Business Today!
Summary
Gartner’s April 2026 research found that organizations with successful AI initiatives invest up to four times more in foundational data and analytics infrastructure than those that struggle. For anyone who has spent real time in Cognos or TM1 environments, this confirms what we’ve always known: garbage in equals garbage out. Data has been the lifeblood of analytics since the decision support era, and that doesn’t change because the downstream consumer is now an agent instead of an analyst. In IBM analytics environments, a strong data foundation means semantic layer quality, real governance workflows with actual stewards, and how you expose context at the presentation layer. When AI projects stall in these environments, the root cause is almost always upstream.
Gartner published research this spring based on a survey of over 350 data and analytics leaders. They concluded that organizations with successful AI initiatives invest up to four times more in foundational data and analytics infrastructure than organizations that struggle. The areas driving the gap are data quality, governance frameworks, AI-ready talent, and change management programs.
My reaction when I read it is, “yes, obviously.” From the decision support days through machine learning workloads to generative AI and agentic workflows, it’s always been garbage in, garbage out. The quality of your solution correlates to the quality of your data. That’s the way it’s always been and very likely always will be. What has changed is that AI has turned a data quality problem into a very visible, very expensive failure that shows up quickly.
Table of Contents
What Does “Data Foundation” Mean in a Cognos or TM1 Environment?
When people in the IBM analytics world hear “data foundation,” they tend to think about source system connections and ETL pipelines. That’s part of it. But the foundation for AI goes deeper and further up the stack than most teams have invested in.
At the source end, it means using the right tool for the right data engineering job. Not every objective is the same. Log data, sensor data, and clickstream data are fundamentally different from financial data, HR data, or sales data, and each requires real thought about the best approach given your constraints. I’ve worked with clients who weren’t funded to move, transform, and enhance transactional data in a conventional data warehouse paradigm. For those environments, a third normal form ingestion approach made more sense. Something purpose-built to take a massive flow of transactional data across multiple business domains and make it analytic-ready. Using the right tool matters a lot because it enables every downstream benefit that comes from ingesting, storing, and exposing data properly.
Then there’s the presentation layer, and this is where IBM analytics teams consistently leave value on the table. Governance artifacts, smart metadata, descriptions, crowdsourced ratings, contextual tags used to be documentation overhead, but now they’re directly relevant to how AI systems find and understand your data. Context matters more in the generative AI and agentic world than it ever has. Without it, your agents are shooting in the dark.
When AI Stalls in a Cognos or TM1 Shop, What’s Usually Upstream?
This comes up constantly with clients. An organization adds AI capabilities to their IBM environment and it doesn’t go well. Is it the AI tools? The implementation? Almost always, upstream factors are driving it, and those factors center on data reliability caused by a lack of real governance practice.
A lot of organizations think they have governance in place. But there’s a difference between having some governance and having a full governance practice with automated workflows, real data stewards with authority, and buy-in throughout the business. Without all three, you get downstream exposure – a lack of trust in outputs, lack of explainability when something looks wrong, and lack of observability across the system.
Black box doesn’t work anymore. Trying to explain a neural network to a financial analyst isn’t going to happen and isn’t going to drive adoption. But there’s plenty that is explainable, and building that trust is non-negotiable for any AI workflow that actually gets used and stays in production. IBM’s stack does a reasonable job here. Watsonx.ai and the observability tooling in the platform exist for exactly this reason. But those tools can’t fix a governance gap that exists before the data ever reaches them.
I’d also say to be deliberate about adopting V1 of agentic features. Version one is exactly that. The starting point, often rushed a bit, building a foundation but not yet delivering on all the promises. If you’re going to be an early adopter, fully understand the limitations, and understand how those limitations present not just to you as a technologist, but to your actual data consumers. A bad first experience with AI in your analytics environment is hard to walk back.
Why Getting the Data Right Is the Faster Path, Not a Detour
PMsquare has been data-focused since day one, which puts us at an interesting position now that AI has become the priority conversation. We sit at the convergence of cloud, AI, analytics, and planning, and solid data work is how AI actually gets delivered in practice.
One thing worth saying plainly: IBM Cognos and TM1 customers have already invested heavily in rich semantic layers. These are real assets, not liabilities. We’re not starting from zero when we talk about AI-readiness. The work is to extend what’s already there, to supplement those existing semantic models with the tenets of real data governance, and to make what’s trusted and already working ready for a new class of consumer.
And that new class is worth pausing on. It’s not just analysts and executives in your audience anymore. Bots are in there now. Agentic workflows service data the same way users do, but they need context presented differently. In the current version of Cognos Analytics, Markdown is one of the output options for presentation objects. That’s exactly what LLMs use to ingest and understand. Lean into it. Help your robot overlords a little bit.
How Should a Team Start Without Overbuilding?
When you’re building a decision intelligence workload that integrates AI, make it purpose-built and start small. A sandbox built around a specific data domain, with well-defined consumers and clear use cases, is far more likely to succeed than a broad deployment trying to address everything at once. Start there, prove the value, then expand.
You can bolt on capabilities as the core proves out. SPSS add-ons for statistical process control, vector databases, integrated Jupyter notebooks, and watsonx.ai for explainability all extend your world without requiring a rebuild. But they depend on having the core data domain solid first.
The bigger shift is organizational. AI prototypes that stay as prototypes don’t drive business results. Moving from prototype to practice requires breaking down silos between data engineering, AI engineering, and business delivery. It’s as much socio-technological as it is technical. The engineering pieces need to integrate, and the team structures need to reflect that. Organizational data and AI teams can’t exist in silos anymore if the goal is genuine operationalization.
Treat your data like the valuable commodity it is. Enhance it, govern it, and document it to the point where you can extract maximum value from it using the agentic workflows that are available now and the ones that are coming. The Gartner 4x finding isn’t a new insight for anyone who has been doing this work. It’s just finally getting the attention it deserves.
Questions I’m Hearing From Clients
It means a high-quality semantic layer with real governance artifacts (descriptions, metadata, access controls) and a data engineering approach that has been matched to the types of data you’re working with rather than copied from a generic template. The presentation layer matters too. Context needs to be in the system so that both human analysts and AI agents can find and understand what’s there.
If the problem shows up as a trust or explainability issue, users or downstream systems can’t validate outputs, or the AI is confidently wrong in ways that track back to data quality, you almost certainly have an upstream data problem. If well-governed, trusted data is consistently producing bad AI outputs, then you’re looking at the tooling side. In my experience working with Cognos and TM1 clients, the former is far more common.
Because AI failure is visible and fast. A poorly governed analytics environment could quietly produce bad reports for years; it was easy to blame the reports themselves rather than trace the problem upstream. With AI agents producing outputs at scale and in more contexts, a governance gap shows up as trust failures across the whole system. The urgency is real now in a way it wasn’t before.
It’s an asset, provided the layer is well-maintained and trusted. A mature semantic model gives AI agents an already-governed, already-tested structure to work from. The risk is accumulated technical debt that nobody wanted to address. AI will find those gaps faster than your analysts ever did. So treat this as a forcing function to tighten it up, not a reason to abandon the investment.
Start with what IBM has already shipped. The recommendation, summarization, and sharing agents in Cognos 12.1.2 are production-ready and run inside your existing security model. Users can only see what they already have access to. From there, lean into Markdown output for presentation objects, add context artifacts to improve agent findability, and consider a RAG-style supplement for decision intelligence workloads that need it. Build purpose-built and narrow, then bolt on as the core proves out.
We hope you found this article both intriguing and informative. At PMsquare, we specialize in cutting through the hype to deliver impactful, outcome-driven AI and analytics solutions. We help you build the data foundation, implement the right tools, and establish the governance needed to turn AI’s promise into your competitive advantage. If this is something you are looking for, contact us today.
Be sure to subscribe to our newsletter for more PMsquare updates and insights delivered straight to your inbox.
