"Most people don’t have data from today": Is business intelligence stuck in the past?
"If an answer shows up a week later in a PDF, you’ve probably missed the window where it actually mattered."
If you work anywhere near data, you’ve probably heard some variation of the same sales pitch for the last decade: dashboards, insights, “single pane of glass”, etcetera. The colours got nicer, the charts smoother, but the core workflow hasn’t changed much since the early 2000s – shove data into a warehouse, pre-aggregate it, wait for a report, build a slide deck.
Marc Stevens thinks that entire model is running out of road. Stevens is CEO and cofounder of Row64, a “real-time visual intelligence” platform that very deliberately avoids calling itself BI. Traditional business intelligence, he argues, is fundamentally about looking in the rear-view mirror. Row64 wants to be the live dashboard on the windscreen.
“We’re not just faster BI,” he tells me. “When we talk about operational intelligence, we’re talking about what’s going on now.”
That distinction – historical reporting versus live operational awareness – is the hill Row64 is choosing to fight on.
The 3D graphics mindset meets enterprise data
Row64’s origin story starts in an unusual place: gaming and visual effects.
Stevens spent years building tools for animation and games studios, living through the industry’s shift from CPU-only rendering to full GPU acceleration. His cofounder came from the Commodore 64 generation and later the world of large-scale data science.
Together, they watched an odd divergence unfold: consumer and creative software became blisteringly fast and interactive, while most enterprise analytics still behaved like it was 2008.
They also knew what happened when an industry finally learned how to use GPUs properly.
READ MORE: Sizing up agentic AI: The data challenge facing every enterprise
“Everyone in VFX and gaming already had GPUs, and we learned how to squeeze performance out of them,” Stevens says.
Meanwhile a typical business laptop – often with a surprisingly capable integrated GPU and plenty of CPU cores – barely had any software that took advantage of it.
Around 2020, they began looking at BI with fresh eyes. The mismatch between how fast decisions needed to be made and how slow tooling was at surfacing insight looked increasingly absurd. That was before you add streaming telemetry from billions of IoT sensors, or the looming wave of agentic AI firing off autonomous tasks, transactions and recommendations at machine speed.
“If the answer shows up a week later in a PDF, you’ve probably missed the window where it actually mattered,” Stevens says.
Row64 was built around a simple, blunt idea: treat analytics more like a game engine – something you interact with live, not something you wait for.
The secret sauce: one memory format, no copying, no waiting
Under the hood, Row64 is built on a set of performance sins that Stevens absolutely refuses to commit.
The biggest? Copying data.
In the classic enterprise stack, data ricochets between tools and intermediate formats: operational databases → ETL jobs → warehouses → cubes → BI tools → exports. Every hop introduces latency, cost and an opportunity to lose something important.
Row64’s answer is a shared real-time memory format that the platform never converts or duplicates. Storage, calculation and rendering are treated as a single system. Once data lands in that memory model, every component – GPU rendering, CPU queries, filters, sorts, cross-references – works directly against it.
The layout is optimised for cache locality, making operations on hundreds of millions of rows effectively instantaneous.
READ MORE: "We still need humans!": Data centre truckers gear up for a blue-collar AI boom
“It’s one big block,” Stevens says. “Nothing gets copied or reformatted. That’s why you can scroll through a billion records like it’s nothing.”
A Linux server does the heavy lifting, on-prem or in a private VPC; the user interface is entirely browser-based, which eliminates client deployment headaches. Geospatial layers, images, video, PDFs, CAD files – it all sits in the same environment.
It’s an architecture designed for people trying to keep up with the real world, not the reporting cycle.
Two hours to nine seconds: the motorsport example
Stevens offers one of his favourite illustrations: a racing motorbike company that rented track time and drivers to test new iterations of a bike. Each session produced a torrent of telemetry. In theory, that’s ideal data for fast iteration.
In practice, they were waiting almost two hours between “bike returns to the pit” and “here’s a graph that tells us something”. Most of that delay was the analytics stack grinding through raw data.
Row64’s first run brought that down to nine seconds, without heavy optimisation. That compresses what can be tested in a day, which directly affects performance gains, engineering throughput and cost.
Retail: from 60-day hindsight to same-day decisions
Row64 recently published another retail example that makes the same argument in a different key.
A national retailer needed to understand why some stores were lagging behind others. The traditional BI workflow meant pulling data into spreadsheets, building pivot tables, and assembling static dashboards. It was slow, brittle and almost always based on stale data.
With Row64, the retailer could instead build a live operational view that let them:
- compare store performance dynamically
- tie sales patterns to geography, inventory, staffing and promotions
- track heatmaps of product categories across regions
- overlay external data like weather or regional events
- spot anomalies or outliers in real time
One insight stood out: the system identified a regional cluster where sales were underperforming due to a localised supply chain delay. Instead of waiting for the monthly report cycle, the team were able to detect the pattern immediately and divert stock.
The result wasn’t just “faster dashboards”. It was the ability to catch problems on the day they emerged, not weeks later when the opportunity (or damage) window had closed.
This is the same pattern Stevens emphasised in our interview: shrinking the decision lag.
“Most people don’t have information from today,” he says. “They get it long after acting on it would be valuable.”
Whether you’re tuning a racing bike or managing a thousand stores, the question is the same: how many cycles do you lose to delay?
When BI becomes operational intelligence
Row64 can behave like a classic BI tool – if you want a bar chart, you can have one. But Stevens is clear that this is not the point.
BI tools, he argues, force users to think in predefined slices of time and predefined chunks of data. They can’t hold everything you might want to look at simultaneously. You have to decide what matters before you ask the question.
Operational intelligence works the other way: the system holds everything, and you follow your instincts as you explore.
“You should be thinking of your operations as a real-time living system,” Stevens says. “Not at the speed your teams can build reports.”
That’s why Row64 lets you go from a high-level overview to a single row out of hundreds of millions without kicking off a new pipeline or annoying your data team. If something smells wrong, you can dive straight into the guts.
Which brings us to AI.
AI moves faster than people can think – but you still need a window to see inside
Most vendor roadmaps treat AI as a feature bolt-on. Stevens is more interested in what happens once AI agents start quietly running large parts of a workflow.
Whether the work is done by a human or a model, the organisation still needs visibility. If ten autonomous agents are making thousands of micro-decisions per hour, how do you know what’s happening?
“Whether it’s a person doing the work or an AI agent, you still need to see what they’re doing,” he says. “Otherwise you’re just hoping the machine made the right decision.”
LLMs and AI services already produce telemetry – prompts, latencies, model behaviour, anomaly indicators – but most companies have no real-time window into that. Stevens believes Row64 can fill that gap, surfacing AI activity alongside all the operational signals that models consume or influence.
It’s the same human-in-the-loop story: faster systems require faster oversight.
The era of streaming everything
Stevens points to moves like Snowflake’s Snowpipe streaming service as early signs of an irreversible trend: warehouses becoming real-time systems, not historical archives.
Once your storage layer goes real time, the analytics layer has to follow. Competitive advantage becomes a function of how tightly you can compress the window between event → understanding → action.
“Real time isn’t a magic number,” he says. “It’s as fast as you need it to be for the decisions that matter.”
For some businesses, that’s minutes. For others, seconds. For a growing number, it’s sub-second.
What’s not viable, he argues, is pretending the world still runs on a 30-day reporting cadence.
Your data firehose is already real time. The question is whether your decisions are.
