A lesser-discussed challenge in the AI boom: ongoing data limitations

Walk into almost any major company today, and you'll hear the same thing: "We're implementing AI." From retailers managing inventory to manufacturers tracking shipments, AI has become the go-to answer for nearly every operational challenge. There's just one problem: it may not always perform as expected, particularly compared to investors' and employees' expectations that it will.

The reason goes deeper than bad technology or overeager vendors. It points to a fundamental reckoning with the way businesses have built their software infrastructure for the past two decades and why that foundation may no longer be fit for purpose.

Great demos, thin results

The typical enterprise AI pitch goes something like this: feed your data in, watch the system analyze it, receive a recommendation for the optimal action. Clean dashboards. Confident predictions. But there's a question nobody asks: whose data is it analyzing? And is it actually connected to the systems where real work gets done?

The uncomfortable reality is that most enterprise AI is built on fragmented data pipelines, exported spreadsheets, and workaround integrations between systems never designed to share information. No matter how advanced the algorithm, garbage data in means garbage results out, just with a more polished interface. Industry analysts have started calling this "AI theater," and businesses worldwide are spending billions on it.

The SaaS era built a data problem hiding in plain sight

For twenty years, the prevailing logic in enterprise technology was simple: buy the best software for each function and connect them with integrations. The result was an explosion of Software-as-a-Service products and a sprawling, fragmented mess of disconnected data. A best-in-class tool for finance. A different one for inventory. Another for orders. Another for shipping. None of them talking to each other in any meaningful way.

That model is now running headlong into the AI era, and the transition has presented some challenges. AI doesn't just need data. It needs unified, real-time, transactional data: the actual record of what happened, when, and what it cost. That data exists inside operating platforms, the software businesses use to actually run their operations. It does not exist in usable form, scattered across a dozen SaaS subscriptions stitched together with middleware. This is what analysts mean when they say traditional SaaS models are evolving: not that subscriptions are going away, but that isolated tools with no unified data layer underneath are fundamentally incompatible with the AI-powered future every company says it wants.

Data isn't enough. AI needs context.

Data and context are not the same thing. Data is the raw record of what happened. Context is the layer of meaning that makes it actionable: why it happened, what else was happening simultaneously, and whether it represents a problem or an expected outcome.

Without context, AI produces answers that are technically correct and operationally useless. A 12% spike in shipping costs looks like a crisis, unless the AI knows you launched a promotion yesterday that deliberately pulled in expedited orders. A carrier delay flags as an anomaly, unless the AI knows whether it cost you $4,000 or $40,000 and whether it cascaded into missed SLAs. Stripped of context, AI flags anomalies. Given context, it explains them.

This is the difference between AI that observes your business and AI that understands it. A system built outside your operations has to guess at context from whatever data you send it. A system embedded inside your operations already knows: it was present for the promotion launch, the carrier negotiation, the warehouse staffing decision. It knows what normal looks like on a Tuesday in peak season versus a slow January. That embedded intelligence is what converts raw data into decisions.

Why operating platforms are the bedrock of AI transformation

Operating platforms don't need to import or reconstruct operational data. They generate it. According to Deposco, the company has supported more than $100 billion in commerce across over 5,500 brands and third-party logistics providers, and it isn't just running warehouse and fulfillment operations. It's building a broad transactional data network in the supply chain, where operational activities such as picking, packing, and shipping can contribute to ongoing AI model refinement over time.

The value of that aggregated data goes beyond any single transaction. Patterns emerge at scale that are invisible in isolation: which SKU combinations consistently stress labor, which carrier routes quietly erode margin, which promotional structures create fulfillment costs that offset the revenue they generate. That kind of intelligence doesn't come from a point solution with a narrow data view. It comes from a platform that has seen it all play out across thousands of businesses.

The hidden cost of getting it wrong

The stakes are real. Deposco describes a scenario from one of its early customers of Deposco's Supply Chain Intelligence platform, noting that the company's best-selling SKU, responsible for 18% of total revenue, looked profitable by every conventional measure. Operations knew the product required special handling. Finance thought the margins looked fine on paper. But nobody had ever connected the labor cost, the carrying cost, the expedited shipping cost, and the actual order-level profitability into a single view.

The result: that best-selling SKU was losing money on every single order. Not occasionally. Every order. And the company had no idea, because the data that would have revealed it lived in three different systems that had never been connected.

That's the kind of insight AI is supposed to surface automatically, continuously, and early enough to change course. Not as a one-time analysis buried in a quarterly report, but as a living intelligence layer woven into daily operations. Deposco calls this Supply Chain Intelligence: not a dashboard layered on top of operations, but an embedded intelligence layer that knows the cost of every decision as it happens.

Most enterprise AI systems today can't do that, because the data they need lives in too many disconnected places.

From insight to action: the next frontier

The bigger shift coming in AI isn't just about better dashboards. It's about AI that actually does the work. For decades, enterprise software assumed a human sits between insight and action. That model is changing. A new generation of "agentic" AI systems is designed not just to inform decisions, but to execute them—adjusting replenishment orders automatically, recalculating margins during a flash sale, reallocating labor before overtime kicks in.

Deposco's multi-agent AI system, called Felix, is built around exactly this idea. When a flash sale drives order volume up 40%, Felix doesn't wait for a monthly business review to surface the margin impact. It calculates in real time that those orders required 2.3 times the normal pick time, pushed shifts into overtime, and destroyed margin on 67% of transactions. And it may provide insights early enough to inform decision-making. Not as a report. As an active operational participant.

This is the version of AI that could genuinely transform how businesses operate. But it only works when the AI has access to unified, trustworthy data, not when it's stitching together exports from a dozen disconnected systems.

Three questions every business should ask

Where does the data actually live? If the AI requires you to export or sync data into a separate environment before it can do anything useful, that's not a technical footnote. It's a fundamental limitation. The gap between data and action is where AI value disappears.

Does it get smarter across customers? AI that learns only from your own historical data has a ceiling. Systems operating across networks of businesses can tell you not just what happened, but what's achievable, benchmarked against peers in the same market conditions.

Can it explain its reasoning? A recommendation without an explanation is just a guess with confidence. The best AI systems show their work: here's the data that drove this insight, here's what changed, here's why it matters for your margins. If you can't audit the reasoning, you can't trust the output.

The real transformation is still ahead—for those who build on the right foundation

The technology is real, the potential is genuine, and businesses with strong data foundations may be better positioned to see long-term benefits. But the SaaS era left most companies with data infrastructure never designed for this moment: dozens of disconnected systems built to move records, not power intelligence. That legacy may present limitations in certain contexts.

The path forward isn't buying another AI tool and adding it to the stack. It's consolidating onto operating platforms that generate unified, transactional data as a byproduct of running the business. That's what platforms like Deposco are built to do: not sell AI as a feature, but make AI inseparable from operations.

The AI era doesn't reward the most sophisticated model. AI systems tend to perform best when supported by well-integrated operational data, which is often housed within platform-based environments. Organizations without this level of integration may see more limited outcomes.