Strategy
Why AI Isn't Helping You Create Value — and How to Fix It
We're in 1776, and someone has given us a Ferrari. To use it, we have to build roads and dig for oil.
Abstract
- 01The bottleneck for AI value is rarely model capability. It's that the environment around the model — the roads for the Ferrari — hasn't been built yet.
- 02Most businesses aren't legible to AI. The information that describes them is scattered across systems, inboxes, documents, and people's heads.
- 03Collecting the data isn't enough. It has to be connected the way the business actually works — as one interconnected system, not a pile of disconnected records.
- 04A common operating picture — a continuously updated digital representation of the business — is what lets AI reason in context instead of through a narrow window.
- 05Uber's matching algorithm is valuable because it operates on a complete picture of its world. The same algorithm in 1997 would have been commercially useless.
Imagine going back to 1776, during the American Revolution. Cars had not been invented, roads were rudimentary, and most people moved through the world either on foot or with horse power. Now imagine introducing a Ferrari into that world and telling people, “Here you are — now you can get around much faster.”
They would understand the value immediately. Of course it would be useful to move faster than a horse, and of course a machine capable of doing that would seem extraordinary. But almost nobody could actually use it, because the Ferrari itself would only be one piece of a much larger system that did not yet exist. There would be no roads designed for it, no petroleum distribution network, no gas stations, no mechanics trained to repair it, and none of the surrounding infrastructure required to turn the invention from a curiosity into something economically useful.
That is, in many ways, where we are with AI today.
The models are already remarkable. We have seen them write software, analyze contracts, summarize enormous bodies of information, generate images, answer questions, and increasingly reason through complex problems. We know, at this point, that AI can be useful. What remains much less obvious to many companies is how to translate that underlying capability into durable business value.
The problem is not necessarily that the models are not good enough. Often, the problem is that the environment around them has not been built yet.
A Ferrari needs roads. AI needs a business it can actually understand.
Making the Business Legible to AI
Most businesses are not naturally legible to an AI system.
Take a manufacturing company, for example. The information required to understand what is happening is usually scattered across dozens of places. Customer conversations live in email, Slack, phone calls, and meeting recordings. Orders and invoices live in an ERP, opportunities live in a CRM, production data lives in manufacturing systems, and machine telemetry may sit somewhere else entirely. Important context remains buried in PDFs, spreadsheets, shared drives, text messages, and, often, in the heads of the people doing the work.
Take a law firm, for example. The information required to understand a matter is rarely contained in one place. Client conversations live in email, phone calls, and meeting notes; documents and evidence sit in document management systems; deadlines and matter details live in practice-management software; billing and time entries live somewhere else again. Around all of that are contracts, filings, research, prior work product, internal messages, and years of accumulated institutional knowledge spread across different attorneys and systems.
Take a hospital, for example. The information required to understand what is happening with a patient, a care team, or the hospital itself is spread across a wide range of systems. Clinical notes and medical history live in the EHR, appointments in scheduling software, lab results and imaging in separate clinical systems, billing and claims in financial platforms, and patient communication across calls, portals, emails, and messages. On top of that are staffing schedules, bed availability, medication records, discharge planning, and the accumulated knowledge of the doctors, nurses, and administrators doing the work.
A person working inside the organization can move between these systems, ask someone a question, remember what happened last week, and piece the story together. An AI system cannot do that unless the underlying information has first been made available to it in a coherent form.
That means the first step is not simply “use AI.” It is to make the business observable.
You need to collect the signals that describe what is actually happening: communications, transactions, documents, operational data, system activity, and the other information generated as work gets done. But collection by itself is not enough, because a pile of disconnected information is still just a pile.
The information also has to be connected.
An RFQ needs to be understood as the precursor to a particular quote, which may require a certain set of parts, which in turn depend on specific suppliers and available inventory. If that quote becomes an order, fulfilling it may depend on the capacity of a particular machine that is also supporting several other jobs, each with its own customer and promised delivery date. A delay in one part of that chain can therefore ripple through production schedules, purchasing decisions, customer commitments, and future quoting.
A client communication needs to be understood in the context of the matter it belongs to, the documents and evidence associated with that matter, the attorneys working on it, and the deadlines that govern what happens next. A newly discovered fact might change the research that needs to be done, affect the strategy being pursued, create work for another attorney, and ultimately influence what is communicated back to the client. What looks like a single email or document is therefore part of a much larger web of activity surrounding the matter.
A patient’s condition needs to be understood alongside their medical history, medications, recent lab results, imaging, clinical notes, and the care they are already receiving. A change in one of those signals may affect the physician’s treatment plan, require a new test, change the medications being administered, alter the patient’s expected discharge date, and in turn affect staffing or bed availability elsewhere in the hospital. What happens to one patient can therefore have consequences that extend well beyond a single chart or department.
The business itself is not organized as a collection of software applications. It is one interconnected system. If AI is going to operate intelligently inside that system, the data has to reflect those relationships.
The Common Operating Picture
Once that information has been collected and connected, it becomes possible to create what we think of as a common operating picture: a continuously updated digital representation of the current state of the business.
The point is not simply to put everything into one database. The point is to represent the business closely enough that a system can reason about it in context.
If a customer asks whether an order can arrive by Friday, the answer may depend on inventory, supplier lead times, production capacity, the status of other jobs, the health of a machine, and the history of promises already made to that customer. The question may sound simple, but answering it well requires a view across the operation rather than a single system.
If a new fact emerges in a matter, understanding what it means may require looking across the client’s communications, the relevant documents and evidence, prior research, deadlines, opposing arguments, and the strategy the legal team has already been pursuing. The significance of the fact comes not from the fact alone, but from how it changes the broader context of the matter.
If a physician is deciding whether a patient is ready for discharge, the answer may depend on vital signs, recent lab results, medications, imaging, clinical notes, mobility, follow-up care, home circumstances, and whether another member of the care team has raised a concern. No single record necessarily contains the answer; the decision depends on the picture formed by all of them together.
If those facts remain disconnected, then even a very capable AI model is forced to reason through a narrow window. A common operating picture widens that window by giving the AI a digital state of the world in which the decision is actually taking place.
And once that state exists, the value of intelligence layered on top of it increases dramatically.
Example: Uber’s Actionable Operating Picture
Uber is a useful example because its business depends on exactly this idea.
At any given moment, Uber has to know where drivers are, where riders are, where those riders want to go, what the surrounding road network looks like, where traffic is building, how long different routes are likely to take, and what supply and demand look like across a city. All of those signals are collected continuously and assembled into a constantly updating digital representation of the rideshare network.
Only then does the matching algorithm become useful.
The intelligence that pairs a rider with a nearby driver is valuable because it operates on top of a reasonably complete picture of the relevant world. It knows where both parties are, what the roads look like between them, how long the pickup is likely to take, and how those decisions interact with the rest of the network.
Without that picture, the algorithm has very little to work with.
Imagine someone had invented Uber’s driver-rider matching algorithm in 1997. It might have been an elegant idea, perhaps even a strong research paper, but there would have been almost no way to commercialize it. You could not continuously know where millions of drivers and riders were. Smartphones were not yet ubiquitous, GPS was not sitting in everyone’s pocket, mobile connectivity was limited, digital mapping was immature, and payments were far more cumbersome.
The intelligence could have existed. The infrastructure required to make the intelligence useful did not.
That is the distinction that matters for AI today.
We already have access to increasingly powerful models, and those models will continue to improve. But for many companies, the real bottleneck is no longer the intelligence itself. It is whether the business has been represented in a way that allows that intelligence to see what is happening, understand the relationships between different parts of the operation, and reason with enough context to be useful.
Building the Roads for AI
This is the foundation Overstand helps companies build.
We bring together the scattered information that describes how the company actually operates — customer conversations, RFQs, quotes, orders, ERP data, production schedules, machine telemetry, inventory, suppliers, and delivery commitments — and turn it into a coherent operating picture that AI can actually work with.
We bring together the scattered information that describes how the firm actually operates — matters, client communications, documents, contracts, research, deadlines, billing data, work product, and internal knowledge — and turn it into a coherent operating picture that AI can actually work with.
We bring together the scattered information that describes how the hospital actually operates — patient records, clinical notes, lab results, imaging, medications, staffing, scheduling, bed availability, billing data, and patient communications — and turn it into a coherent operating picture that AI can actually work with.
The goal is not simply to clean up data, and it is not to bolt an AI assistant onto another piece of software. It is to create the underlying environment in which AI can understand the business with enough context to become genuinely useful.
The models are already here.
The Ferrari has been invented.
What most companies are still missing are the roads.
Related
Frequently asked questions
Because a model is only one piece of a larger system. Like a Ferrari delivered to 1776 — extraordinary, but without roads, fuel, and mechanics it stays a curiosity. AI needs the environment around it to be built before its capability turns into durable business value: the information describing the business has to be collected, connected, and made available to the model in a coherent form.
The information that describes what is actually happening — communications, transactions, documents, operational data, system activity — is usually scattered across many systems and people's heads. A person can move between those systems and piece the story together; an AI system cannot, unless that information has first been collected and connected into a coherent representation of the business.
A continuously updated digital representation of the current state of the business — not just everything in one database, but the business represented closely enough, with its real relationships intact, that a system can reason about it in context. Once that state exists, the value of intelligence layered on top of it increases dramatically.
Uber's matching algorithm is valuable because it operates on a reasonably complete picture of its world — where drivers and riders are, what the roads look like, how decisions ripple through the network. The same algorithm invented in 1997 would have been nearly impossible to commercialize, because the infrastructure required to make the intelligence useful — smartphones, GPS, mobile connectivity, digital maps — didn't exist yet. For AI in most companies today, the missing piece is the same: not the intelligence, but the infrastructure that gives it context.