We’ve mastered data. Now we’re making AI work.

19 May 2025 | 3 minuten leestijd

Everyone’s building something with AI right now. Some are plugging in chatbots, others are running automation across tools like Slack or HubSpot. It looks slick from the outside, but under the hood, things often get messy. AI workflows only work if the data behind them is right. And that’s where we come in.

At i-spark, we’ve been building Composable Data Hubs for years. These are modular data platforms that ingest, transform, and activate structured data. They power dashboards and deliver Enriched Data Feeds to systems like email tools, ad platforms, and now AI agents. So when we talk about AI, we’re not switching lanes. We’re building on what’s already there.

AI workflows × Composable data

Roughly speaking, there are three kinds of AI-related projects we encounter.

First, there’s lightweight automation: tools like Make or n8n hooked up to business apps. You drop in some logic, call an LLM, and get an email or Slack message as output. These workflows are widely used and if that’s your use case, we can help.

Second, there’s data infrastructure. This is what we’ve always done. A Composable Data Hub that delivers both Insights (dashboards, reports, analyses) and Enriched Data Feeds as structured outputs ready for activation. These feeds used to flow into marketing tools. Increasingly, they now power AI-driven decisions.

Where it gets interesting is when these two worlds meet: AI workflows built on top of strong, structured data ecosystems.  That’s our focus: Not only prompting or developing interfaces, but creating value by applying ecosystems that give AI the clarity and context it needs to be genuinely useful.

One example: when the model looks smart but gets it wrong

In one of our MVPs, we used n8n to generate daily updates on product metrics, dropped straight into Slack. The idea was to replace manual analysis with something faster, fed by both internal data and outside context like competitor promotions or market news.

The LLM-generated query tried to sum up daily user counts to compare this week with last week. But these counts were snapshots, not flows, so adding them together didn’t make sense. A junior analyst might have made the same mistake. A senior one probably wouldn’t. That’s the point.

It isn’t just about fixing syntax. It’s about understanding what the data means, and how to catch flawed logic that looks polished. That’s why we believe data professionals have a key role in building AI workflows that use real business data.

"We’ve spent years building structured, governed data systems. Now we’re using that foundation to power AI workflows that are not just impressive, but reliable."

Why data teams matter in AI workflows

Building AI tools is a team effort. Front-end developers and automation builders all play their own part especially in crafting great interfaces and flows. But when your automation touches structured, sensitive, business-critical data, then you need people who know how that data is modeled, calculated, and governed. That’s where we - seasoned data professionals - bring real value, by ensuring the AI has clean inputs and applies the right logic:

• We know which metrics can be summed, and which can’t.
  
• We think in terms of lineage, context, and consistency.
  
• We catch when AI output doesn’t make sense, even if it looks fine at first glance.
  
• We build with governance, access control, and sensitivity in mind.

It’s the difference between making a tool that runs, and a system that holds up over time. Making AI reliable, not just impressive.

Or as we say:

If you’re serious about AI, you need to get your data right. And if you’re serious about data, you can’t ignore AI anymore.

What makes a business ready for AI?

Not all AI runs on structured data. But when it does, when it's powering decisions, automation, or personalisation, then the foundations matter. In our view, there are five building blocks that make an organisation AI-ready:

• Clear metadata & semantics: context around what the data means, how it’s defined, and how it should be used.
  
• Modular architecture & scalable infrastructure: modular systems that scale, connect well, and expose clean interfaces.
  
• Trusted data quality & traceable lineage: trusted pipelines with validation, freshness, and clear traceability.
  
• Robust governance & controlled access: knowing who can use what, under what conditions.
  
• AI-ready integration layer: workflows, promptable layers, or vector access points that allow AI to use the data meaningfully.

These aren’t just technical checkboxes. They’re what allow both teams and AI systems to make decisions with confidence.

Where we stand

We didn’t dive in headfirst when AI first took off. Not because we weren’t interested, but because we were focused on what would prove useful, not just novel. We explored, experimented, built MVPs. Now that the initial dust is settling and the path to real opportunities is becoming clearer, we’re ready to move.

If you’re building a custom front-end around AI prompts, we’re probably not your team. But if the power of your AI depends on structured, reliable data then that’s exactly where we come in. So if you're working on serious AI, and want it to hold up over time, we’d love to help.

We’re not changing direction.

We’re building forward.