# Chapter 5 β€” Local Wins Everywhere ## Local Wins Everywhere 🎧 [Listen to Chapter 5](https://thetwoproject.com/player.html?src=https%3A%2F%2Ftheowner-audio.s3.us-east-2.amazonaws.com%2FThe_Owner_Chapter_5.mp3\&chapter=5\&type=gitbook) The AI Progress Report arrived in Ethan's inbox on a Sunday night. Twelve pages, compiled by the project team for Karen Holt's monthly leadership review. Raj had forwarded it with a one-line note: *You should see this before the steering meeting.* Ethan opened it on his laptop in the hotel room. *GENAI TRANSFORMATION β€” PROGRESS REPORT, MONTH THREE* *Titanshield Insurance / Northstar Consulting* *HEADLINE: 82% of claims adjusters have logged at least one AI-assisted draft in the past 30 days.* *Month 1 adoption: 34%* *Month 2 adoption: 61%* *Month 3 adoption: 82%* *Caption: Strong adoption momentum across all regions.* He scrolled through the rest. Adoption by seniority band. Adoption by office location. A heat map showing which teams had the highest and lowest usage rates. Three pages of charts, all measuring the same thing: how many people were touching the tool and how often. The highest-usage team was in the Ohio regional office β€” the same office where Marcus had found the repealed regulation. Eighty-nine percent adoption. The team generating letters that cited a law no longer in effect was also the team with the best score on this chart. *** The next morning, Ethan showed the progress report to Angela in the hallway outside the conference room. Angela flipped to the adoption chart. Three seconds. "Who compiled this?" "Raj had it pulled together for the steering review." She handed it back. "Karen's going to ask where the ROI is. This report doesn't have an answer. It has a participation trophy." "What would you measure?" "Validation time. How long does my team spend reviewing a draft before they approve or reject it? That number tells you whether the tool is helping or creating work. If validation time is going down, the drafts are getting better and my people trust them. If it's going up, the drafts are worse or my people don't trust them and they're checking everything." "That number is going up," Ethan said. "I know. But nobody put it on the dashboard. They put adoption on the dashboard. Because adoption is easy to measure and it goes in the right direction." She looked at the conference room door. "Karen is going to walk in there and see eighty-two percent adoption and ask what she's getting for it. And the answer is: eighty-two percent of her team touching a tool that some of them don't trust, some of them are using wrong, and all of them are spending more time reviewing than they saved on drafting." "So the metric is a lie." "The metric is true." Angela looked at him. "Eighty-two percent of adjusters used the tool. That's a fact. But it's the wrong fact. It's like measuring how many people went to the gym without asking if anyone lost weight." Ethan put the report in his bag. The board decks Julia was preparing came to mind. The same green arrows. The same upward trend lines. Usage. Adoption. Engagement. All the metrics that proved people were doing the new thing without proving the new thing was working. *** The monthly Steering Committee at Titanshield Insurance started with a parade of green arrows. Angela Ruiz shared her screen to a room that already felt like a victory lap. "Claims Explanation Pilot," she said. "Complaint rate is down forty percent. Citation accuracy is up to ninety-nine percent. Zero hallucinations in the last three weeks." Michael Tran nodded from the end of the tableβ€”his version of a standing ovation. "Compliance citations are fully traceable. The audit trail is clean. We can defend every output in court." Ethan sat next to Angela, watching Karen Holt. The Chief Claims Officer was reviewing the printed deck, flipping pages with a rhythmic *snap, snap, snap*. She hadn't looked at the screen once. "Excellent," Ethan said. "We've stabilized the quality issues." Karen stopped flipping. She rested her hand on the table. "Quality is up," she agreed. "Complaints are down. Michael is happy." She looked up, her gaze landing hard on Ethan. "So where is my ROI?" The room went quiet. Karen let the silence sit. "I'm looking at the operational metrics," Karen continued, tapping a page in the appendix. "Average Handling Time is flat. Headcount is flat. Overtime is flat." Angela shifted in her seat. "We're spending less time drafting, Karen. But we're spending more time reviewing. Run the same claim twice and you get slightly different wording β€” sometimes a different citation order, sometimes a different emphasis. My team can't just glance and approve. They have to read every line." She paused. "The net is neutral." "Neutral," Karen repeated. "So I spent half a million dollars on a pilot to generate better letters at the exact same cost?" "We reduced risk," Michael interjected. "That has value." "Risk is a cost avoidance," Karen cut in. "I can't put 'we didn't get sued' on the P\&L. The board wants to know when GenAI is going to bend the cost curve. You told me this would make us efficient. Right now, it's just making us polite." Ethan leaned forward. "It's the first phase, Karen. We needed to establish trust before we pushed for speed." "We established trust," Karen said. "Now I need value. Until I see the unit cost go down, there is no expansion. The Underwriting pilot is on hold." Angela looked at Ethan. The Underwriting expansion β€” their revenue driver β€” was gone. "Bring me a business case, not a compliance report," Karen said, closing the binder. "Meeting adjourned." *** Three hours later, Ethan was on a video call with the Northstar leadership team. Julia Reyes sat in her New York office, looking at the same two charts Ethan had just seen rejected. Raj Mehta was on the line from Atlanta, where he was running the Swiftcurrent Logistics engagement. "Karen paused Underwriting?" Julia asked. "She wants ROI," Ethan said. "We improved quality, but we didn't touch speed. The review process is eating all the gains." "We prioritized safety," Lena Park said from the New York conference room. "That was the mandate." "It was *one* mandate," Julia said. She turned to the screen. "Raj. How is Swiftcurrent?" Raj smiled. He looked tired but satisfied. "Different story here. Paul Jensen is ecstatic. Decision turnaround in Exception Handling is down thirty-five percent. They are ripping through the backlog. Client satisfaction scores are the highest in two years." "And the risk?" Julia asked. Raj shrugged. "They're aggressive. They loosened the executive review gates. They aren't kicking margin exceptions to a human. Paul says he'll take the occasional error if it means clearing the queue. But Swiftcurrent is winning. Paul is ready to sign a renewal. We should be scaling this." "Scaling what?" Ethan said. "They turned off the gates. They're rubber-stamping outputs. Paul's numbers look great until a client gets a hallucinated promise." "Karen's numbers look safe until the board asks why we're spending half a million dollars to go sideways," Raj shot back. "So your answer is to move fast and hope nothing breaks?" "My answer is that we have a client who's actually *using* the tool, and you want to slow him down because Titanshield can't figure out a review process." Julia let them go for another thirty seconds. Then she cut in. "Stop. Listen to what you're both saying." She looked at Ethan. "You're defending safety that costs too much to scale." She looked at Raj. "You're defending speed that's one incident away from a lawsuit." She paused. "We told the board we have a 'GenAI Transformation Platform.' We don't. We have two bespoke consulting projects that are optimizing for opposite things." The room was quiet. "If we scale Titanshield's approach, we lose money β€” the service delivery costs too much," Julia continued. "If we scale Swiftcurrent's, we blow up a client. Those aren't two strategies. That's no strategy." "The technology can do both," Lena said. "But the workflows are breaking." "Then fix the workflows," Julia said. "No new clients. No new pilots. We are frozen until we can prove this model works without trading safety for solvency." "Karen is waiting for an answer," Ethan reminded her. "Tell her the answer is process redesign," Julia said. "Tell her we're not expanding to Underwriting until we fix the Claims workflow. We need to get the cost down." She looked at Raj. "And tell Paul to slow down. If Swiftcurrent crashes, it's on us." *** That evening, Ethan sat in the hotel bar in Chicago, staring at his laptop. He pulled up the logs Lena had shared. The "divergence" was clear in the data. At Titanshield, the line for "Time in Review" was creeping up. Every week, the adjusters were adding seconds to their checks. They were learning to mistrust the system, padding it with human oversight. At Swiftcurrent, the line for "Time in Review" was crashing. They were rubber-stamping. They were learning to ignore the system's warnings in favor of its speed. Two clients. Same tool. Opposite behaviors. "Local wins everywhere," Ethan muttered to himself. Angela was happy because she was safe. Raj was happy because he was fast. But Karen wasn't paying for safety, and Northstar couldn't afford the liability of unchecked speed. He closed the laptop. The problem wasn't the AI. The problem was that they hadn't decided what it was for β€” a shield or an engine. His phone buzzed. It was a text from Raj. *Paul wants to expand to Pricing. He says the 'Engine' is ready.* He typed back: *Hold him off. Julia meant it.* He put the phone down. He knew Raj wouldn't hold him off. Raj never did. Ethan signaled the bartender for a check. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://thetwoproject.gitbook.io/the-owner-preview/ch05-local-wins-everywhere.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.