{
  "format": "minglabs/v1",
  "surface": "insights-article",
  "slug": "the-last-mile",
  "kind": "article",
  "subtype": "position-paper",
  "url": "https://www.minglabs.com/insights/articles/the-last-mile",
  "htmlUrl": "https://www.minglabs.com/insights/articles/the-last-mile",
  "partOf": "https://www.minglabs.com/insights/articles",
  "title": "We made our agents email people. That's when the AI started working.",
  "subtitle": "95% of enterprise AI pilots show no return. The gap isn't the model. It's the last mile — and the last mile is boring, which is why almost everyone skips it.",
  "hook": "Most enterprise AI pilots don't fail in the model. They fail in the last mile — where the work has to leave a human's hands and land somewhere useful. That mile is unglamorous, so it gets skipped. We made our agents email people. That decision moved more than any model could.",
  "summary": "MIT's 2025 study: 95% of enterprise GenAI pilots produced no measurable P&L impact — and the cause was implementation, not model quality. The 'last mile' is everything between a capable model and actual use: which channel the work lands on, who owns the outcome, and whether anyone gives it longer than two weeks to improve.",
  "category": "Hybrid Organisation",
  "pillar": "Hybrid Organisation",
  "author": {
    "name": "Sebastian Mueller",
    "role": "Founding Partner, MING Labs"
  },
  "datePublished": "2026-05-27",
  "dateModified": "2026-05-27",
  "freshness": {
    "updated": "May 2026",
    "nextReview": "November 2026"
  },
  "evidenceTier": "experiential",
  "confidence": "B",
  "sources": [
    {
      "id": "S1",
      "title": "The GenAI Divide: State of AI in Business 2025",
      "publisher": "MIT NANDA",
      "date": "2025-12-01",
      "supports": [
        "95% of enterprise pilots show no measurable P&L impact",
        "60%→20%→5% evaluate-pilot-production funnel",
        "ROI concentrates in back-office and operations",
        "Divide driven by implementation, not model quality"
      ]
    },
    {
      "id": "S2",
      "title": "Internal MING Labs operations log, January–May 2026",
      "publisher": "MING Labs",
      "date": "2026-05-27",
      "supports": [
        "Agent channel design — email, Telegram, pushed briefings",
        "Observed adoption pattern"
      ]
    },
    {
      "id": "S3",
      "title": "Internal agent audit — decommissioned coordinator agent",
      "publisher": "MING Labs",
      "date": "2026-03-30",
      "supports": [
        "Correct-kill counter-example for week-one-output kill signal"
      ],
      "n": 764
    }
  ],
  "faqs": [
    {
      "q": "Why do most enterprise AI pilots fail to deliver ROI?",
      "a": "MIT's 2025 study found 95% of enterprise generative-AI pilots produced no measurable P&L impact, and that the cause was implementation, not model quality. In our experience the failure sits in the 'last mile' — embedding the output where work happens and giving the agent long enough to improve — not in the model."
    },
    {
      "q": "What does 'the last mile' mean for AI adoption?",
      "a": "It's everything between a capable model and actual use: which channel the work is delivered on, who owns the outcome once it arrives, and whether anyone acts on it. A tool people have to remember to open rarely gets used; a colleague who emails you the work does."
    },
    {
      "q": "Should you kill an AI pilot that underperforms in the first weeks?",
      "a": "Not on output quality alone. An agent that learns from corrections looks ordinary early and improves later — week-one polish is a poor kill signal. Judge instead on whether the work is owned, whether anyone acts on it, and whether corrections are sticking."
    }
  ],
  "relatedConceptSlugs": [
    "what-is-hybrid-organisation",
    "what-is-the-abc-framework"
  ],
  "relatedArticleSlugs": [
    "we-fired-an-ai-agent"
  ]
}