{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreibxsckidilubhswxk7ffrebyl3trciaxvcrghy2n45wwwfa23mor4",
    "uri": "at://did:plc:qz6ohvpdsdvv5kniizyfz25y/app.bsky.feed.post/3merv5c2nay32"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreibaetk2rlfqvpzj45j7a4ehzt6y7nzw4f44obmwgy5nhw7nkl3dqe"
    },
    "mimeType": "image/jpeg",
    "size": 3182039
  },
  "path": "/article/4132176/solving-enterprise-ais-roi-problem.html",
  "publishedAt": "2026-02-13T17:31:39.000Z",
  "site": "https://www.cio.com",
  "tags": [
    "Artificial Intelligence",
    "MIT figure",
    "Behavioral Agent Automation Platform",
    "here"
  ],
  "textContent": "By now, everyone has seen the MIT figure: Despite $30-40 billion in enterprise investment into GenAI so far, “95% of organizations are seeing zero return.” Boards want to see results; CFOs are asking hard questions. But the promised productivity gains have remained largely theoretical so far. The reasons are twofold: the**sequencing problem** and the**scaling problem.**\n\nAgentic automation represents one of the greatest levers for unlocking new enterprise value at scale. But today, the implementation of agentic AI forces IT leaders into an impossible choice between two flawed approaches: “off-the-rack” point solutions that require predicting which workflows matter most, or “DIY” custom frameworks that dump complex automation responsibilities onto already-stretched teams.\n\nBoth of these approaches have strengths on their own, but each faces the same two fundamental challenges:\n\n  1. They force organizations to design solutions before understanding the actual problem\n  2. Neither can solve for both sequencing _and_ scaling at the same time\n\n\n\n**The sequencing problem**\n\nTraditional automation requires CIOs to be fortune tellers. Which processes deserve automation first? Where does friction and/or opportunity actually exist? What workflows genuinely drive inefficiency?\n\nMost organizations have to essentially guess. They survey employees, map theoretical workflows, and build automations based on assumptions (or at best, self-reported interviews) about how people work.\n\nThe fundamental issue is sequencing: enterprises are designing against problems they can’t accurately observe using traditional methods. Design predates discovery when it should be the other way around.\n\n**The scaling problem**\n\nDIY custom frameworks have the opposite problem: they’re imminently customizable, but at significant cost. Even if someone did accurately identify the best candidates for automation, solving it requires deep domain knowledge of agentic tools and capabilities, plus the know-how to assemble, deploy, and maintain custom workflows in perpetuity (all while those employees have their regular jobs to do).\n\nThis approach doesn’t scale across enterprises where every department and team has unique workflows, different tools, and varying levels of technical sophistication. Custom solutions become technical debt faster than they create value.\n\nOff-the-rack solutions scale, but don’t customize. DIY frameworks are indeed custom, but they don’t scale.\n\n**A different approach: Intelligent automation based on observed behavior**\n\nLeading organizations are flipping the script. Instead of predicting workflows, they’re implementing a Behavioral Agent Automation Platform (BAAP). A BAAP not only observes how people actually work with GenAI, but it can then _automatically_ surface and build automations based on proof, not prediction.\n\nThe shift is from hypothesis-driven to data-driven automation. This is why BAAPs represent the future of enterprise agentic AI — they solve the prediction > proof, sequencing, and scaling problems in unison.\n\nThe building blocks that make Behavioral Agent Automation possible:\n\n  * A secure, horizontal platform for accessing and querying GenAI\n  * Model-agnostic access and orchestration (so you’re not locked into one foundation model)\n  * Enterprise-wide data access that feeds into the platform (Slack, email, SharePoint, Drive, CRMs… wherever knowledge actually lives)\n  * Behavioral observability within the platform itself\n  * An insights engine capable of interpreting observability data, identifying friction, and surfacing automation candidates proactively\n  * Governance infrastructure with human-in-the-loop approvals\n  * Autonomous assembly and deployment of agentic workflows (following human approval)\n  * Real-time telemetry on what’s working, what’s unused, and where friction persists\n  * Continuous adaptation so the deployed automations improve over time, automatically\n\n\n\n**New evaluation criteria**\n\nFor CIOs, the evaluation criteria are shifting. The question is no longer whether a platform can execute workflows… It’s whether a platform can discover the biggest inefficiencies and solve them automatically.\n\n  * Can the platform observe how work actually happens across your fragmented systems?\n  * Can it identify friction from behavioral signals rather than requiring you to guess?\n  * Can it assemble and deploy individualized automation at scale?\n  * Can it do that while also maintaining the governance and control your organization requires?\n  * And perhaps most importantly, does it reduce the burden on IT teams or simply create a new technical dependency?\n\n\n\n**Moving past pilots**\n\nOrganizations seeing real ROI share a common insight: they stopped trying to predict the future of work and started observing the present reality.\n\nThis doesn’t mean abandoning strategic planning or IT governance. It means building automation strategy on evidence rather than assumptions. It means deploying tools that learn from behavior rather than requiring employees to become solution architects.\n\nFor details on how Behavioral Agent Automation Platforms are solving the $40 billion enterprise AI value question, you’ll find the deep dive here.",
  "title": "Solving enterprise AI’s ROI problem"
}