Stop Chasing the Next Model — Fix Your Data
5 April, 2026
Guest Column: Shai Friedman, CTO at CodeValue, explains why organizations rushing into AI are overlooking the most critical factor — the quality of the data powering their systems
Imagine hiring the best chef in the world. They have a state-of-the-art kitchen, premium knives, a skilled team — everything needed to create a perfect dish. And then you hand them wilted vegetables, expired meat, and a sauce that’s been left out for two days.
You might still get a plate that looks impressive — but the first bite will be a disaster.
That’s more or less what’s happening today in thousands of organizations rushing to adopt the next AI model, without stopping to examine what they’re feeding into it. The arms race between Anthropic, OpenAI, and Google generates headlines almost weekly — but it also creates a smokescreen. As executives scramble to check the “we have AI” box, they often skip the least glamorous — and most important — question: is the underlying data even fit for use?
Take, for example, an airline launching an AI-powered customer service assistant. The company invests heavily, connects it to internal databases, and presents a polished demo to the board. On paper, everything looks great. In reality, the system relies on poorly maintained data: baggage policies updated in one system but not another, destination names appearing in multiple formats, and refund information spread across three systems that don’t always align.
The result? One customer is told they’re allowed a suitcase, another is told they’re not, and a third is issued a refund they weren’t entitled to. This is no longer just a bug — it’s operational, service, and reputational damage.
This is where much of today’s confusion begins.
Many people talk about models as if they’re the magic itself — as if choosing the right model automatically delivers intelligence, accuracy, and value. But a model isn’t magic. It’s an engine. And if you feed it duplicate, incomplete, inconsistent, or outdated data, it won’t fix reality. At best, it will recycle the problem. At worst, it will do so fluently, quickly, and with complete confidence.
We used to call this “Garbage In, Garbage Out.” Today, it’s far more dangerous — because the garbage out no longer looks like a mistake. It looks like a professional answer.
And that’s what makes this issue so critical. Once AI systems start influencing decisions — in customer service, pricing, marketing, healthcare, banking, or operations — data quality stops being just a “data team” problem. It becomes everyone’s problem.
Data preparation is the Cinderella of the tech world: unglamorous, painstaking, invisible work — but without it, the entire promise of AI collapses.
What does that actually mean in practice? Not just “cleaning data,” but ensuring it is complete, consistent, reliable, and up to date. It sounds basic, yet in many organizations this is exactly where things fall apart: the same customer appears in multiple versions, the same product is described differently across systems, one policy is updated while another isn’t, prices differ between the website and the CRM. All this chaos is fed into a model — and people expect a miracle.
There won’t be one. At best, you’ll get a decent-looking demo that barely works.
Of course, models do matter. There are real differences between them, and in some cases those differences are significant. But in many organizations, model selection isn’t the primary problem. The real issue is that leaders want to feel like they’re “in the game” — without doing the slow, messy, unsexy work of building solid data foundations.
And that’s exactly the work that separates organizations experimenting with AI from those actually generating value from it.
So before asking “which model should we choose?”, it’s worth asking three far less exciting questions: what data are we relying on, who is responsible for its quality, and what price are we already paying because it isn’t good enough?
That’s the real story: data doesn’t just determine whether an AI system works — it determines whether it saves money, wastes money, improves service, or creates damage at a scale and speed we’ve never seen before.
The race for the next model will keep making headlines. But the real question isn’t who won this week’s benchmark — it’s who is finally willing to fix their raw materials.
Posted in: AI
