Your company already runs on an OS
It is just fragmented across forty tools that do not talk to each other. That was survivable when humans did the joining. It stops being survivable the moment you want AI to run on top.
The shape of the problem
Building tools is now cheap. Connecting them is the whole job.
Which is exactly why buying more of them stops working: every purchase adds another island, another definition of “customer”, another export nobody reconciles.
What gets built
Three layers. The middle one is the part nobody sells you and the part everything else depends on.
In what order
Decisions first, definitions second, plumbing third, surfaces last. Reversing that order is how companies end up with dashboards nobody opens.
- 01Map the decisionsWhich decisions each domain owns, and on what cadence. Everything else is derived from this.
- 02Fix the definitionsOne meaning per metric, ratified by the domain that owns it, not by the analysts.
- 03Wire the coreIdentity, event log, contracts between domains. The part nobody sells you and everything depends on.
- 04Build the surfacesViews, tools and agents on top of the core. Now cheap, because the hard part is done.
- 05Close the loopDecisions produce events, events feed the core, the core sharpens the next decision.
Why it compounds
A fragmented company loses the loop. Decisions get made, and the outcome is recorded somewhere that never reaches whoever makes the next one. Nothing accumulates.
Once the core exists, every decision leaves a trace in the same place, and the next decision starts better informed than the last. That is the actual return, and it is why this is infrastructure rather than a project.
It is also the precondition for AI being useful rather than impressive. An agent pointed at forty disconnected systems produces confident nonsense. Pointed at a coherent core, it can do real work.
“Tools are easy to build now. Why do I need you?”
Fair, and half right. Generating a tool is genuinely easy now, and if that were the job I would not be worth hiring.
The hard part was never the tool. It is deciding what a customer is when four systems disagree, getting a definition ratified by the people who own the number, keeping history when a source system overwrites it, and knowing that most dramatic movements in a business series are artifacts of the systems recording them rather than the business. That work does not get cheaper because generation got cheaper. It gets more valuable, because the cost of building on top of a wrong foundation just dropped to nearly zero and everyone is about to do it very fast.
The method is written up in full here - function by function, with the failure modes named. Read it and decide whether it sounds like someone who has done this before.
How an engagement starts
Decision inventory and source audit. You get the map whether or not you build anything with me.
The connective layer: identity, definitions, event log, contracts between domains. This is where the leverage is.
Views, tools and agents on top. Fast, because the hard part is behind you, and disposable, because they are no longer where the truth lives.
Step one is deliberately separable. If the map says you do not need this, that is a useful answer and a cheap one.