ROI Case File No.528: Contract Review Lived Inside One Person's Head
Contract Review Lived Inside One Person's Head
Chapter 1: When That Person Is Off, Contracts Stop
"Only one person handles legal review. When that person is off, contract checks stop."
Saori Ukita, administration headquarters manager at LexCorp, said this while pointing to a stack of contracts. "About twenty contracts arise a month. Heavy or light in content, every one needs a check. During the new-contract and annual-renewal periods they concentrate, and the load skews onto one person. We want to do something about the personnel dependency."
"Are you using AI currently?" Claude asked.
"The person in charge is personally using the free version of ChatGPT," Ukita answered. "They say it's convenient, but it's not something we deployed as an organization. There's a security anxiety about pasting contracts into an external AI. So as an organization, we want to bring in a generative AI we can use in a closed environment. That's how the talk went at our monthly review meeting."
"Is the way to proceed decided?" I confirmed.
"That's the worry," Ukita answered. "Do we bring in generative AI company-wide all at once, or start with just legal review? Which operation to begin with, and how to expand. We can't set the investment priority. There's no doubt legal review is urgent, but we can't draw the expansion map beyond it."
"Which operation to concentrate AI investment in, and how to grow it—you need to organize it with a portfolio," I responded. "Let's break it down with PPM."
Chapter 2: PPM Asks Which Quadrant to Concentrate Investment In
"This case needs PPM."
Claude drew four quadrants with vertical and horizontal axes on the whiteboard.
"PPM—Product Portfolio Management—is a framework proposed by the Boston Consulting Group that classifies businesses or measures into four quadrants by two axes, market growth rate and share, to decide investment allocation," I explained. "It transfers to AI deployment too. We sort each internal operation into quadrants by 'size of effect' and 'ease/maturity of deployment,' and design where to invest first and how to grow and expand. Which quadrant legal review sits in is the starting point of the investment judgment."
"First, let's measure the current cost," Gemini said, opening ROI Polygraph. He entered the data Ukita had provided.
"The monthly legal-review cost came out," Gemini read aloud. "Contract-review labor is twenty items a month, averaging three hours each, for 60 hours a month; at the person's hourly rate of 5,200 yen, 312,000 yen a month. The cost of overtime and delays from concentration during busy periods averages 250,000 yen a month. The expected value of the work-stoppage risk from dependency on one person averages 400,000 yen a month. The judgment-variance risk of review quality depending on the person averages 200,000 yen a month. The expected value of the information-leak risk from personally-used AI averages 350,000 yen a month. The total is 1,512,000 yen a month. Annualized, that's about 18.14 million yen."
Ukita stared at the figures. "I thought it was only the check labor. I had no idea the personnel-dependent work-stoppage risk and the leak risk of personally-used AI would become numbers too."
"Then let's design with PPM," I continued.
[Star—Large Effect and High Maturity: Legal Review]
"First, the positioning of legal review," Claude said. "Its effect is large, and it's also an AI strong suit. Extracting points of contention from contracts and detecting risk clauses are mature areas where generative AI already achieves high accuracy. We position this as a star and concentrate investment here first. We build it in a closed environment as the first move to resolve the personnel dependency."
[Question Mark—Large Effect but Low Maturity: Other Legal Operations]
"Next, the question-mark quadrant," Gemini continued. "Contract-drafting support and more advanced legal judgments have large effect but AI maturity is still low. Investing right now carries high failure risk. We hold them as candidates to grow in stages after seeing legal review succeed. A decision not to reach for everything at once."
[Cash Cow—A Steady Efficiency Area: Standard Document Processing]
"The cash-cow perspective," I continued. "Confirming the standard portions of contracts and referencing past cases are areas that steadily produce efficiency effects. Not flashy, but they steadily cut costs. Once the legal-review platform is in place, they can be expanded while holding down additional investment. Like stable income, an area that works diligently."
[Dog—Areas to Forgo Investment]
"Last, the dog quadrant," Claude continued. "We don't invest in operations where the effect is small and AI-fication is hard. Rather than trying to AI-fy everything, we include in the design the decision to exclude them from the investment targets. The essence of PPM is also deciding what not to do."
[The Phased-Deployment Design]
"With the four quadrants in mind, the order of deployment settles," I continued. "We concentrate investment in the star—legal review—and proceed in three phases in a closed environment: preparation, deployment, operation. After confirming success, we expand to the question-mark and cash-cow areas. A design that doesn't AI-fy all operations at once but grows from the area where effect is certain."
[Estimating the Investment Recovery]
"Let's estimate with ROI Proposal Generator," Gemini proposed.
- Initial cost: Closed-environment generative-AI construction, documentation of the legal-review process, legal dataset preparation, test operation, and internal training—5,800,000 yen total
- Monthly cost: AI-environment operation plus maintenance ongoing cost—200,000 yen a month combined
- Monthly reduction effect: Contract-review labor reduction = 220,000 yen a month (assuming 70% reduction), busy-period delay-cost reduction = 200,000 yen a month, reduction of personnel-dependent work-stoppage risk = 300,000 yen a month, reduction of information-leak risk = 350,000 yen a month—1,070,000 yen a month total
- Monthly net reduction: 1,070,000 yen − 200,000 yen = 870,000 yen a month
- Payback period: 5,800,000 yen ÷ 870,000 yen = about 6.7 months
"Recovery in a little over half a year," Gemini summarized. "What matters is concentrating investment in the star and not expanding to all operations at once. Because we produce certain results in the mature area of legal review before expanding, the investment is unlikely to be wasted. The resolution of information-leak risk through closed-environment-ization also works strongly."
Ukita confirmed the figures and said, "With a concept of bringing AI into all operations, I couldn't set priorities. When you split into four quadrants, where to invest first becomes clear."
"PPM is a tool for deciding where to concentrate limited investment," I responded.
Chapter 3: A Deployment Plan That Proceeds in Three Phases
"Let me organize how we'll proceed," I said, standing before the whiteboard.
"Preparation phase (months one and two)—detailed documentation of the current legal-review process, preparing a dataset to have the AI learn legal knowledge, and building the closed environment. Deployment phase (months three and four)—test operation of the generative AI in the closed environment and adjustments based on the person in charge's feedback. Operation phase (month five onward)—starting official operation, expanding so employees other than the person in charge can use it, and periodic performance reviews and improvement. Month seven onward—based on the success, considering phased expansion to the question-mark and cash-cow areas."
"Making it usable by people other than the person in charge is the key to resolving personnel dependency," Ukita confirmed.
"Exactly," Claude responded. "The closed AI extracts points of contention and detects risk clauses. With this, even someone other than the person in charge can do a first-pass check. The person in charge concentrates on the final judgment, and even when they're off, the first-pass response doesn't stop. Personnel dependency is the state of 'only that person can do it.' The AI turns the knowledge into shared property."
Taking notes, Ukita said, "Trying to do everything at once, we couldn't move. With the idea of starting from the star, it looks like we can move forward."
Chapter 4: The Day One Person's Head Became the Organization's Asset
Eight months later, a report arrived from Ukita.
Contract-review labor was cut 70% versus before, three months after the closed generative AI went live. "The AI extracts the points of contention and risk clauses first. The person in charge can concentrate on confirming them and the final judgment. Three hours per item became under an hour," Ukita wrote.
The personnel dependency was also resolved. Employees other than the person in charge could now do a first-pass check, and the work no longer depended on one person. "Even when the person in charge is off, contract checks no longer stop. The tightrope of 'things don't run without that person' is over," the report said.
The biggest change appeared in stability during busy periods. Even when new contracts and annual renewals concentrated, they could get through with the AI's first-pass processing. "The structure where the person in charge collapsed under overtime in busy periods disappeared. The work was leveled out," Ukita wrote.
Information-leak risk was also resolved. They moved from an operation where an individual pasted contracts into the free version of ChatGPT to organizational use in a closed environment. "The anxiety of putting confidential contracts into an external AI is gone. It became an environment we can use with peace of mind as an organization," the report said.
As a secondary effect, check quality was standardized. With the AI carrying the standard for extracting points of contention, variance in judgment by the person in charge shrank. "Whoever checks, the points the AI picks up are the same. The floor of quality was raised," Ukita wrote.
Phased expansion also began. Following the success of legal review, expansion to standard contract processing and the like began to be considered. "Because the star produced results, the investment judgment for the next area is easier. PPM's phased expansion is moving just as planned," the report said.
At the end of Ukita's report, she had written: "With a concept of bringing AI into all operations, I actually couldn't move. When I split into four quadrants with PPM, I could make the judgment to concentrate on the star of legal review. Because I narrowed down what to do, I could take the first step."
It was recorded that the day a company where contract review had lived inside one person's head turned that knowledge into the organization's asset, legal review changed from a personnel-dependent craft into a mechanism shared across the organization.
"A common stall in AI-deployment consultations is being unable to set priorities because you try to AI-fy all operations at once. The concept spreads to this and that, and in the end you can't invest anywhere. What PPM asks is which quadrant to concentrate limited investment in. Start from the star—large effect, high maturity—and grow the question marks and cash cows after seeing success. Also decide the areas you won't do. The day a company where contract review had lived inside one person's head turned that knowledge into the organization's asset, what changed was not the AI's performance but the very judgment of where to concentrate investment."
Related Files
Tools Used
- ROI Polygraph — Visualizing check labor, personnel-dependent work-stoppage risk, and the leak risk of personally-used AI
- ROI Proposal Generator — Investment-recovery simulation for star-concentrated generative-AI deployment
