There is a quiet crisis happening in e-commerce right now, and it is one that most people do not notice until the delivery truck arrives.

It is the crisis of Data Drift.

With the rise of generative AI, it has never been easier to fill a website with product descriptions. Machines can write poetry about stainless steel. They can generate thousands of words about the “sleek design” of a refrigeration unit in seconds. They make websites look full, professional, and comprehensive.

But commercial kitchens do not run on poetry. They run on physics.

A poetic description does not tell an electrician which circuit protection is required. A flowery paragraph does not tell a shopfitter whether the unit will fit through the doorway. It does not account for the strict installation constraints that determine if a site is compliant.

At KW Commercial, we use AI. We respect the technology. It helps us organise information and respond faster. But we have drawn a hard line in the sand regarding how we use it.

We are writing this to explain our internal standard—Source Lock (v14.5), a rule set designed to prevent assumptions where facts matter—and why we believe that in 2026, the most valuable feature of an online catalogue is not its size, but its truth.

The Friction Between Digital Text and Physical Reality

To understand why we built Source Lock, you have to understand the nature of the tools the world is currently using.

Large language models (the engines behind many e-commerce summaries) are optimised to be plausible, not necessarily accurate. They predict what comes next. Therefore, when data is missing, they are naturally tempted to “fill in the blank.”

Why “statistically likely” is still wrong

If you ask an AI to describe a generic two-door commercial fridge, it may tell you it uses a “standard 10A plug.” It might say that because many fridges do.

But if your specific model happens to be a heavy-duty freezer requiring a higher-current connection, the result is not just a data error. It is a planning error.

The real-world cost of a digital hallucination

For a consumer buying a t-shirt, a wrong description is an annoyance. For a commercial project manager, a wrong specification is a liability.

• It is an electrician on site charging a call-out, while a circuit upgrade is organised that was never planned.
• It is joinery that must be reworked because the wrong dimension was used in the drawing.
• It is a compliance conversation that starts late because a material detail was assumed instead of verified.

We realised early on that we could not—and should not—compete in a race to the bottom of “who can generate the most content.” That is a race we are happy to lose.

Instead, we chose a harder path. We chose to treat product data not as marketing content, but as engineering constraints.

The Discipline of “Zero Inference”

In the software world, there is a famous philosophy: “Move fast and break things.” That mindset can work for social media apps. It does not work for heavy machinery.

If we “break things” in our industry, a restaurant does not open. A warranty can be voided. A safety risk can be created.

This is why we built Source Lock not as an AI accelerator, but as an AI restraint. Most importantly, we define it by what it refuses to do.

The rule of empty fields

Our protocol (v14.5) enforces a standard we call Zero Inference.

If an official manufacturer datasheet is silent on a specific detail—say, “gas connection pressure”—our system is forbidden from looking at similar models to guess the answer.

Instead, the system must leave the field blank, or flag it as [Verify on Site].

This can feel counter-intuitive. Isn’t the goal to have the most complete data? No. The goal is to have the most trustworthy data.

We would rather show you a blank space—prompting the right question—than show you a “likely” number that leads to the wrong regulator. An empty field is an invitation to verify. A guessed field is a trap.

The Source Lock Rulebook (What We Refuse to Do)

To keep this practical, here is the simplest way to understand Source Lock: it is a hierarchy of sources, plus a set of “no-go” behaviours. It is intentionally boring. That is the point.

Table 1: Common high-risk fields (and our default response)

Table 2: Source hierarchy (what we trust first)

The Future Is Automated, But the Foundation Is Human

We make no secret of our direction. We know where the world is going. We see a future where procurement becomes more automated, where tools assist designers with planning, and where efficiency scales. We are building towards that future deliberately, one constraint at a time.

But we also know the fatal flaw of “PPT entrepreneurship”—selling the dream of automation without doing the dirty work of infrastructure.

You cannot build a skyscraper on a swamp.

If the underlying data—the volts, the amps, the dimensions—is hallucinated, then any automation built on top of it will simply automate disaster at scale. Therefore, before we race toward the future, we do the unglamorous work of making the foundation real.

We respect that a commercial combi oven is a heavy, expensive, complex piece of engineering that needs to arrive, fit, and work. No amount of AI poetry can change the laws of physics or local compliance requirements.

Our promise

As we roll out tools to make your job easier, this is our Source Lock promise: we will leverage AI to organise, to speed up, and to serve—but we will never let it pretend to know what it doesn’t.

In a world of synthetic content, we remain committed to engineering truth.

What Happens When We Don’t Know

This is the part most companies skip, because it introduces friction. We include it because it is the reality of fit-outs.

When a spec is not certain, we do not “fill the gap” to keep a quote moving. Instead, we use a simple rule: if the answer is not 100% certain, we stop.

The practical workflow

1. We label the uncertainty (blank field or [Verify on Site]).
2. We check primary documents (datasheets, manuals).
3. We seek confirmation (manufacturer support).
4. We ask for site context (when outcome depends on build).

FAQ (For Busy Builders, Designers, and Operators)

Why are some specifications blank?

Because the source is silent, and we refuse to guess. Blank is not neglect; it is restraint.

What does [Verify on Site] mean?

It means the answer depends on site conditions or undocumented details. We flag it early so it gets verified before it becomes a problem.

Do you still use AI?

Yes, to organise and format. Never to invent technical specs.

Glossary

Data Drift: The loss of accuracy when info is copied and “completed” by AI.

Zero Inference: Our rule: if a primary source doesn’t state a fact, we do not guess it.

Source Lock (v14.5): Our internal protocol enforcing source hierarchy.

Changelog

• 30 Jan 2026: First publication. Defined Source Lock v14.5 and Zero Inference.