The AI budget is buying document systems

Members stopped asking how to govern their AI agents. They started asking what the agents are supposed to read.

Six member companies raised it in July. Six different industries: pharmaceuticals, hospitals, manufacturing, software, IT services, and transportation. Five of them named the same category of system. Not a model. Not a security control. Enterprise content management.

Two want to replace a legacy ECM outright. One does not have a formal one at all, and said its agents cannot pull usable context out of SharePoint, Confluence, and file shares. One is consolidating ERP, R&D, and manufacturing data into an ECM so agents have something reliable to read. One asked for help scoping the project from scratch.

Before July, almost nobody in the DoGood network raised this. One company brought it up in April. Nobody did in May.

Here is the part worth your attention. Every one of these is being written up internally as an AI initiative. The thing being bought is document infrastructure that predates the AI budget by twenty years. The AI program is now paying for a content migration that never cleared a business case on its own merits.

Three buyers, three industries, three different budget lines. The third one states the problem the other two are already spending against.

Two raw asks pulled directly from member submissions, unedited:

Neither member is shopping for AI. Both are asking a vendor to fix what sits underneath it.

This week's Signal is about the plumbing under the AI project, and Scot Burdette built his organization around exactly that. He is Global Division CIO for Measurement and Analytics at ABB, where one of his three leaders owns data cleansing and usage outright. He spent a year connecting the path from customer order to factory floor. He described the result in his DoGood member spotlight, "All streamlined into an integrated workflow. Which has been really transformational for us."

Read his full story →

The headlines are catching up to what the network already knew. On August 10, BigDATAwire argued that enterprise AI keeps stalling at what it calls the first mile. That means finding the data, classifying it, and deciding what should never reach a model at all. It cites Komprise survey data. In it, 56% of IT infrastructure directors named classifying and tagging unstructured data their top AI-prep challenge. That is up from 41% the year before. Governance and security came second at 46%.

The argument is that the industry built for the last mile. Chunk the documents, generate embeddings, load a vector database. All of that assumes the content arriving is already clean, classified, and governed. Members were writing that assumption down as a blocker in July.

Bottom Line: The first mile is an ownership problem before it is a data problem. The team that runs your content systems is almost never in the room where the AI budget gets set.

Ask your AI pilot team which system actually holds the documents the agent reads. Then pull the funding history on that system for the last three years. If the answer is a platform nobody has invested in since the last migration, you have found your pilot's real timeline. It is not a model problem.

The CXO Brief is powered by the DoGood network, 5,000+ IT leaders sharing what they are actually working on.

Know a CIO who needs this? Forward it and they can subscribe here.

Enterprise IT leader at a $100M+ company? Apply to join DoGood.