From the diary of Marilyn Tan

The questions to ask before investing further in dashboards, analytics or AI, from a founder who has built the infrastructure underneath them.

Many organisations have bought dashboards, analytics, AI tools and automation. Marilyn Tan has seen how the technology can become “an expensive layer between the data and the decision” rather than something that improves how the business operates.

In February 2025, Gartner predicted that through 2026 organisations would abandon 60% of AI projects that lack AI-ready data. In the projects Marilyn describes, the trouble was in the data underneath too.

Marilyn is the Founder of Curate IQ Lab. Her work, as she describes it, is taking information that already exists within an organisation and making it more usable for a business decision. She usually starts with the decision rather than the data. She describes her work as connecting business strategy, data, AI and operational execution, and she doesn’t see AI as something that should replace human decision-making. She sees it as “infrastructure that can help people make better decisions, faster, with better context.”

What Happened When the Output Arrived

She has seen projects where significant effort went into the analytical or AI layer, only for the organisation to discover that the operational data underneath was inconsistent, fragmented or poorly governed. In one of them, reporting had been built around operational information that looked useful at first glance. When the people on the ground received the output, they couldn’t confidently act on it. The definitions, ownership and data quality were unclear.

“The result was predictable: people questioned the numbers, escalated the issue, or simply reverted to their existing way of working,” she says.

“An accurate model is not necessarily a useful system.”

Marilyn Tan, Founder, Curate IQ Lab

She is careful not to name a single cause. “I would not say they fail because of one single reason. But very often, the model is not the real problem,” she says. “If nobody trusts the data, understands what the insight means, or knows what action to take, the sophistication of the model does not matter.”

Her own work starts earlier than that.

Starting With the Decision

“I usually start with the decision rather than the data,” Marilyn says. Before asking what data an organisation has, she wants to know what decision needs to be made, who makes it, how often, what information they use now, what happens when it is wrong or delayed, and what action should follow when the data points to a particular condition. Only then does she work backwards into the data, governance, workflow and technology. She describes her interest as the flow from “business problem → data → intelligence → decision → action → measurable outcome.”

She wants leaders to ask the same questions before they invest further: “What decision are we trying to improve? Who needs to act on it? What data supports that decision? And what needs to change operationally for the insight to actually create value?”

RAND, in a 2024 study of why AI projects fail, listed business stakeholders misunderstanding or miscommunicating the problem to be solved as the first of five root causes.

What she finds first is rarely a shortage.

More Data Than Clarity

“One of the first things I often discover is that the organisation has more data than it has clarity,” she says. The problem isn’t always a lack of information. Sometimes there is too much of it, captured by different teams for different purposes, with different definitions and no clear ownership.

She made the same point in a recent LinkedIn post with a restaurant kitchen. Messy data, she wrote, is like “cooking in a chaotic kitchen,” with ingredients “scattered across different fridges.”

This is where she puts governance to work. “This is where governance becomes operational rather than bureaucratic,” she says. “Good governance should help the frontline know what they can act on, what requires escalation, and who owns the next decision.”

Her view of what sits underneath a request comes from having built it.

What Building It Taught Her

Marilyn’s background spans Google Cloud architecture work, machine learning engineering and freelance work. As she puts it on her profile, the pandemic “became a turning point,” when she saw the potential in cloud technology, architecture, IoT, AI, machine learning and Python.

“Building the infrastructure gives you a different view of what sits underneath a seemingly simple business question,” she says. A leader may say, “We just need a dashboard,” or “Can AI predict this?” Underneath the request she finds questions about data definitions, system architecture, integration, data lineage, ownership, access, quality and operational behaviour.

“I have learned not to take the stated problem at face value,” she says. Sometimes the real problem is upstream. “Sometimes the organisation is trying to automate a process that has never been clearly defined.” And sometimes, in her words, “the technology is working exactly as designed, but the organisation has not designed the workflow around it.”

She learned something else along the way that has nothing to do with architecture.

The Question She Keeps Asking

“One of the hardest lessons in my career was learning that having the right answer is not always enough,” Marilyn says. “You also need to understand the environment, the people and the timing around that answer.” It taught her to spend more time understanding the problem before trying to solve it.

She says she was shaped by people who challenged her to look beyond the immediate task and understand the bigger business context. One lesson stayed with her: “do not just solve the problem in front of you; understand why the problem exists in the first place.”

It has become a habit. She keeps asking “why?” until she gets past the obvious answer. If someone tells her, “The data is not working,” her next question is usually, “What do you mean by not working?” That question, she says, often opens up the real problem.

The same instinct shapes what she counts as proof.

What Counts as Evidence

“The strongest evidence is not that we produced a dashboard or an AI model,” she says. “It is when the output changes what someone actually does.” The work she has been involved in has included segmentation, pricing models, data pipelines and analytical workflows. The common thread, she says, is taking information that already exists within an organisation and making it more usable for a business decision.

“The business made a different or better-informed decision because we built it.”

Marilyn, Founder, Curate IQ Lab

She is careful with numbers. “I would prefer to specify exactly what they were measured against, the underlying dataset and the period involved rather than present an impressive number without its denominator,” she says of any accuracy or performance figure.

What she wants to build next follows from the same test.

Where Marilyn Is Now

“I want to build practical AI and data systems that help businesses make better decisions, not simply more technology,” Marilyn says.

The reader she most wants to hear from is the business leader or founder who has data but isn’t turning it into decisions. “They do not necessarily need to be technical. In fact, I think that is precisely the point.” She wants to see leaders asking better questions before they invest further.

“Outside work, I am interested in how people learn, communicate and adapt to technology,” she says. “Technology adoption is ultimately not just a technology problem. It is a people and behaviour problem too.”

The line she would use for all of it: “Data creates value only when it changes a decision.”

Five Reflections from Marilyn

What decision are we trying to improve? She starts here, before she asks what data an organisation has.

Who needs to act on it? Her list also asks who makes the decision and how often.

What data supports that decision? Only then does she work backwards into the data, governance, workflow and technology.

What happens when the decision is wrong or delayed? It is one of the first things she wants to understand.

What needs to change operationally for the insight to create value? A system can work exactly as designed and still sit outside the workflow around it.