You say you want to be data driven. Perhaps you’ve invested in making it possible - new dashboards, new tools, new warehouses. Maybe you even layered some LLM-based interface on top of it.

But you’re not seeing the upside yet. Your team still exports the dashboards into Excel. They still have to hunt around for what they need. They have a bunch of data, but find it difficult to do anything actionable with it. And the fancy chat-based LLM interface seems to be giving back answers that feel stale, or just plain wrong.

We have good news. First, you’re not alone. This is a common issue. Second, we know how to fix it.

It’s not a technology problem. It’s actually an operational design problem. All the data in the world can’t help you unless it’s exposed to your team in ways that actually let them use it, in the flow of their work.

This is what we call Ops-Led Analytics. And in this article we’ll describe what it is, and how you can move from analytics as a reporting function to analytics as a tool for constant, confident, adaptive decision making.

Why Top-Down Analytics Fails

The main problem: across industries, only a fraction of employees use the data the organization invests in.

According to MIT, only 28% of employees regularly use reusable data assets. This is true even in companies that have mature data programs.

A few reasons. It’s often hard to find. It’s often mandated from the top down, but without sufficient understanding of the day-to-day workflow of front line teams. Sometimes the teams aren’t trained on what to look for, what to infer from the data that is shown to them, how it should impact their work.

We ran into this on a recent project for a healthcare provider. The operators had dashboards and access to reports. But they still couldn’t answer the questions that were most important to them. Things like:

• What is happening in my facility right now?
• Where am I below expectations relative to my peers?
• Where are there anomalies in the data?

They also had access to a modern GenAI assistant. But like many healthcare organizations, the complexity of their data environment made it hard for the model to reliably connect concepts, metrics, and definitions.

This is a widespread challenge across the industry. When data is spread across multiple systems with inconsistent permissions and schemas, LLMs struggle to provide answers operators can trust.

Why Ops-Led Analytics Works

The goal with Ops-Led Analytics is to solve for these issues. It assumes that the data exists somewhere (or can be wrangled with a proper data foundation initiative.)

Instead of focusing on the dashboards, on assuming that if you build it they will come, it starts with deeply understanding the operators and their actual workflows. Only then does it layer in analytics, to help augment real-world decisions.

This make intuitive sense. And there’s research to back it up.

HBR found that organizations who expose frontline workers with digital tools designed around they way they actually work report material gains in decision-making quality and employee engagement.

McKinsey also found that companies who are seeing actual business upside from AI are much more likely to have done the work of deeply understanding daily workflows and integrating AI in ways that make sense for those workflows.

Four Ops-Led Analytics Best Practices

From the work we’ve done in this area, we have found 4 key practices to make this approach work:

1. Start by fixing the data foundation.

You can’t do much without a proper data foundation. And without it, any GenAI implementation will fail.

It’s not sexy work, but it’s critical. And organizations are glad they do it once they’re done - what McKinsey calls “no-regrets moves”.

2. Design solutions around the operator.

A nuance we’ve found - many companies think they’ve done this work. They did the interviews with frontline workers.

The challenge: especially in hybrid environments, is that it’s often hard for your team to accurately describe their workflow via a Zoom meeting. You have to sit with them - literally see how they do their job.

This is how Palantir built their reputation. They embedded teams on the ground, mapped out their real-world behaviors, and explicitly designed experiences around what they learned. Successful Ops-Led Analytics initiatives require this same fieldwork.

3. Focus on living, adaptive dashboards

It’s easier to create static dashboards. But this inevitably leads to a dashboard proliferation problem. As your team identifies edge cases or follow-up questions, new dashboards get built to answer those new questions.

A better approach:

• Focus on the consistent 20% data that drives 80% of decision-making.
• Derive that 20% from the team’s actual KPIs.
• As KPIs change, adjust the 20% accordingly.
• Expose your team to a smart queryable interface (ideally one that’s LLM-based) to answer those edge case questions.

That last point is a critical one, and a new opportunity. Most of the questions they want answered can (and should) be able to be answered by ML models over time.

4. Push data to the operator, don’t wait for them to pull it.

If you want to change behavior, you don’t want to rely on them having to remember to log back in constantly.

Deliver the most important info via APIs to where your team actually works (Slack, Teams, email, etc.)

This is a design pattern that consumer SaaS companies have long understood. The intelligent use of triggers, alerts, and digests, even though they might not feel core to the product, are actually the tools that drive retention and long-term adoption.

Getting Started With Ops-Led Analytics

You now have the ability to finally give operators the data they’ve always wanted. Modern data architectures, unified semantic layers, and conversational AI make it possible to see metrics that matter. To ask deeper questions. To surface issues early. To facilitate smarter, faster decisions.

It requires a clean enough data foundation. A real understanding of how your team works. And adaptive systems that show them what they need when they need it.

If you’d like to learn more about how Manifold can help kick-start your Ops-Led Analytics journey, don’t hesitate to reach out.

‍