Introducing The Intelligence Factory
Decisions Are The Output. Data & AI Are The Means.
If you ask a business what its data platform delivered this year and you will probably get a list of technical artifacts that were built. Pipelines, a Lakehouse, a semantic model, a data agent etc. Then ask which business decisions are being made better by these artifacts, and the room will go quiet. That gap is not a technology failure. In our experience at Cloud Formations, it is a framing failure, and the words we (as an industry) advertise are part of the issue.
Two Factories Named After Raw Materials
For more than a decade the go-to-market label for data engineering has been the data factory. It was a useful metaphor for industrialising work that had/has been delivered by a technical role, and it sold well because a buyer could picture it. To be clear, this is nothing to do with Azure Data Factory the cloud resource. More recently the label evolved into the AI factory. Same metaphor, newer noun, aligning to the hype for all things AI.
To quote NVIDIA, who have done more than anyone to popularise that term, they define an AI factory as infrastructure whose "primary product is intelligence, measured by token throughput". They correctly identified the product intelligence and then chosen to measure it by the volume of raw material consumed. That said, GPU abstracted tokens for consuming models are not intelligence, any more than tables were. They are what intelligence gets made from.
There is a second problem. An artificial intelligence factory now belongs to the infrastructure vendors. It describes a facility. Very few of the customers I work with need a facility. They need answers to support the business. In short, a factory nobody can name the output of just becomes a cost centre. Degrading data driven thinking and having an impact on the overall data culture.
What the metaphor is good for though, the factory part, I do not want to throw that away, because one part of it is genuinely valuable, and it is not the part usually invoked. Factories are not impressive because they are big or fast. They are impressive because the hundredth unit is as good as the first. Repeatability, not throughput.
Applied honestly to our work, which means the hundredth decision should be as well-informed as the first. That is a considerably harder promise to make than a pipeline count or Lakehouse entities created and a far more useful one.
Delivery Value Today
We have various collateral we have been using in front of customers for a long time. One sets out client priorities on the left, time, cost, risk, trust, value, capability. Then our products and services in the middle, and the outcomes we deliver on the right.
Our other approach sets out the method in three lanes:
Value Discovery works from business context, the business model and user engagement through to use case ideation and a conceptual data model. It is the lane most programmes skip, and skipping it is why so many platforms arrive looking for a purpose.
Value Enablement takes the technical context source systems, architecture, constraints and produces a design and roadmap aligned to the outputs Discovery identified. Technical alignment with the required data outputs, rather than the other way round.
Value Delivery turns that into a source-agnostic reporting model, reconciles the attributes required against the attributes available, and produces a sized, prioritised backlog. You enter build knowing the business impact, the effort and the order.
Whether the answer eventually arrives from a dimensional model, a forecast or an agent is an implementation detail settled here. Choosing the technique before understanding the question is how businesses end up with impressive solutions to problems nobody had. A process, an output, and a measure of progress.
I’ll admit it took me an embarrassingly long time to notice that we had drawn and delivered a factory and never called it one. So, this is less a new idea than an overdue name for something we already do, which is the only kind of terminology I have any patience for.
The Intelligence Factory
An intelligence factory is not a platform, and it is not a project. It is the assembled capability that turns a business question into a repeatable answer, treating data, analytics and AI as ‘means rather than ends’. We could also describe the intelligence factory as the people, process and technology that turns everyday operational data into trusted insights.
I said above a factory nobody can name the output of is a cost centre. So let me name ours. The output is decision capability, and it moves in one direction.
Reactive we see what already happened.
Trusted we trust the data and speak the same language.
Insight led we understand why things happen and where to act.
Predictive we anticipate and prioritise with confidence.
Autonomous systems act and optimise outcomes with minimal human input.
Most organisations know roughly where they sit on that line. Almost all of them can tell you where they would like to be in two years. The distance between those two points is the work, and everything else the platform, the pipelines, the models are the capabilities supporting how we can close it.
There are things we can be held to on each engagement along the way. Business value from data. Faster time to insight. Lower cost platforms to build and to run. Reliable, low risk delivery. Trusted and compliant data, embedded from day one rather than retrofitted. Uplifted internal capability, because we would rather deliver with your team than to them. It’s part of our core values.
Powered By Products. Delivered With Expertise
That is our tagline, and the intelligence factory is what those two halves produce when you point them at something a business needs to know.
The products supply the repeatability. CF.Cumulus, our Lakehouse accelerator. CF.Nimbus, which interprets legacy estates. CF.Stratus, for platform assurance. These are the parts we refuse to rebuild by hand for every customer, and they are the reason the hundredth build can be as good as the first.
Then the expertise supplies the judgement about what is worth building at all. No product has ever supplied that, and I do not expect one to.
The Test
I am aware of the irony in coining a term while complaining about the industry's appetite for them. So let me set the test now. If the intelligence factory turns out to mean the same engagement with a different cover sheet, it deserves every bit of the cynicism it will attract. Personally, I give enough of it out.
It should mean this instead, we stop describing our work by what we build and start describing it by what you can now decide.
That is the standard we are holding ourselves to. In the next post I will take Value Discovery apart properly, because it is the lane that decides whether any of the rest is worth building.
Comments