ROI Case File No.593: The Features Were Decided, but the Risk of Being Shut Down Was Never Appraised
The Features Were Decided, but the Risk of Being Shut Down Was Never Appraised
Chapter 1: They Want to Build Auto-Posting, but Can't Commit
"We want to build an SNS auto-posting system for restaurants."
Kei Koriya, business development manager at Global Reach, said this as he laid out the situation. The company is an advertising agency whose main business is inbound customer acquisition. "We receive photos and menu information from the restaurant, and AI writes the post. Once the restaurant approves it, it goes out automatically. We want to offer it as a subscription for a few tens of thousands of yen a month. Restaurants can't keep up daily posting."
"So the shape you want to build is settled," Claude asked.
"It's settled," Koriya answered. "But as I looked into it, things started catching on me. On some platforms, automatic posting can get an account suspended. If that happens, our client's restaurant account is gone. That's the one thing I want to avoid. And I can't tell whether the economics work at a few tens of thousands a month."
"Have you appraised that risk, and whether the economics work?" I confirmed.
"...I haven't," Koriya answered. "All I've researched is whether we can build it. How likely a suspension is, how reliable the means of avoiding it are, and how easy those means are to implement. I've never laid them side by side."
"Being able to build something and being right to do it are different," I replied. "Let's break this down with ICE."
Chapter 2: ICE Asks—Impact, Confidence, and Ease
"This case calls for ICE."
Claude wrote "Impact / Confidence / Ease" on the whiteboard.
"ICE—Impact, Confidence, Ease—is a framework that scores ideas on three yardsticks to decide where to start," I explained. "The crux is not to judge on feasibility alone. How much does it move, how solid is the read, and how easily can it ship? Multiply the three and the first shape to release is decided. Confidence in particular is the column where risk gets appraised. Build with that column blank, and you will be stopped after you go live."
"First, let's measure the current cost," Gemini said, opening ROI Polygraph. The data Koriya had provided went in.
"The monthly cost is out," Gemini read off. "Labor for writing posts, organizing assets, and posting on behalf of the restaurants currently under management averages 160 hours a month; at ¥3,800 an hour, that's ¥608,000 a month. Revenue lost through cancellations by restaurants that can't sustain posting averages ¥440,000 a month. Development stalled because the account-suspension risk can't be appraised averages ¥380,000 a month. Back-and-forth caused by approvals scattered across phone calls and email averages ¥320,000 a month. Lack of evidence for proposals, because there's no mechanism to measure post performance, averages ¥280,000 a month. Total: ¥2,028,000 a month. Annualized, roughly ¥24,330,000."
Koriya stared at the figures. "I was only looking at the posting burden. Once you add the cancellations from restaurants that couldn't keep it up, and the development cost sitting frozen because nothing gets decided, it comes to this much."
"Then let's design it with ICE," I continued.
[Impact—narrow down where it moves the most]
"First, impact," Claude said. "The reason restaurants can't keep it up isn't the act of posting—it's that nothing comes to mind to write. Generating post drafts works on every restaurant, every day. Automatic posting itself is only the final nudge. The larger impact is on the side before the writing."
[Confidence—appraise the risk of being shut down]
"Next, confidence," Gemini continued. "Posting through each platform's official API falls within the terms of service and isn't subject to suspension. What's dangerous is driving the screen by machine. Draw a line between what the official APIs can cover and what they can't, and confidence turns into a number. Leaving that column blank while you build is the greatest risk."
[Ease—make the first release small]
"After confidence comes ease," I continued. "Draft generation and an approval screen alone can ship in a few weeks. Scheduled posting through official APIs comes next. Ship in order of ease, and you can decide what comes next while watching how restaurants respond."
[Multiply—multiply the three to decide the starting point]
"Last, multiplying," Claude continued. "Large impact, solid confidence, easy to ship. Multiply the three and draft generation comes first. Fully automatic posting has large impact, but it includes a portion where confidence can't be read. Line them up by the product, and the order settles itself."
[Simulating the investment recovery]
"Let's run the numbers with ROI Proposal Generator," Gemini proposed.
- Initial cost: Researching each platform's terms and appraising the risk, implementing the post-generation AI, building the approval feature, official API integration, and implementing performance measurement—¥4,600,000 in total
- Monthly cost: AI usage, API usage, plus ongoing operation and maintenance—¥210,000 a month
- Monthly savings: Automation of post writing and asset organization = ¥430,000 a month (assuming a 70% reduction), suppression of cancellation losses through better retention = ¥340,000 a month, elimination of approval back-and-forth = ¥240,000 a month, securing proposal evidence through performance measurement = ¥220,000 a month—¥1,230,000 a month in total
- Net monthly savings: ¥1,230,000 − ¥210,000 = ¥1,020,000 a month
- Payback period: ¥4,600,000 ÷ ¥1,020,000 = approximately 4.5 months
"That's a payback of four and a half months," Gemini summarized. "What works is that you don't stop at whether it can be built—you hold confidence as a column. Build full automation without appraising the risk, and after go-live a client's account disappears and you lose the business itself. Confine it to the range of the official APIs, and you can sell it with peace of mind. The investment doesn't swing at air."
Koriya looked over the numbers. "I thought researching whether we could build it was enough. Without appraising the risk, you can't even decide whether you're allowed to sell it."
"ICE is a tool for picking a starting point by impact, confidence, and ease," I replied.
Chapter 3: A Deployment Plan That Chooses by Three
"Let me lay out the approach," I said, standing at the whiteboard.
"Month one—research each platform's terms and appraise the suspension risk. Month two—score on the three yardsticks and fix the scope of the offering. Months three and four—implement post generation and the approval feature, and run it at pilot restaurants. Month five—implement scheduled posting through the official APIs. Month six—implement performance measurement and validate the monthly price. Month seven onward—revise the scoring and expand to more industries and more restaurants."
"What restaurants want is full automation. Won't something halfway fail to sell?" Koriya confirmed.
"Full automation that gets stopped sells far worse," Claude replied. "If even one account disappears, that fact travels through the industry in a day. Conversely, deliver the drafts reliably and the restaurants that couldn't keep it up start keeping it up. That track record becomes the foundation for selling the next feature. Confirming the range where you won't be stopped comes before promising everything."
Koriya took notes as he spoke. "Appraise the risk of being shut down before researching whether we can build it. I can see the sequence now."
Chapter 4: The Day the Risk Was Appraised and It Became Sellable
Ten months later, a report arrived from Koriya.
Posting consistency changed dramatically once draft generation was offered. "Restaurants that stalled because they couldn't decide what to write can now post by touching up the text that arrives. What used to break off after three days now continues," Koriya wrote.
No account suspensions occurred either. Confining the design to the range of the official APIs did it. "Because we built entirely inside the terms, I sleep without worrying. And when we explain it to clients, we can speak with evidence," the report said.
The largest change showed up in how they decided on plans. A state of judging by whether something could be built became a state of choosing by multiplying three. "We used to hold meetings purely about technical possibility. Once we scored impact, confidence, and ease, the release order settled in one pass," Koriya wrote.
The service price also gained grounding. Measurement figures became the backing for proposals. "It changed from 'we use it because it's cheap' to 'we keep it because this many more customers came.' Price negotiations dropped off," the report said.
As a side effect, they could explain what set them apart. The very fact of having appraised the risk became a selling point. "The restaurants most worried about safety are the ones who sign once we walk carefully through the terms of service. The topic we'd been avoiding turned out to be our strongest material," Koriya wrote.
The final line of Koriya's report read: "I thought the trouble with SNS automation was whether it was technically buildable. But the real problem was that the features were decided and the risk of being shut down was never appraised. The moment we scored the three with ICE, both the release order and the sales approach were settled. Before researching whether we could build it, measuring confidence came first."
The day a company that never appraised the risk of being shut down became a company that could pick a starting point from three, SNS auto-posting development had changed from confirming feasibility into a design that multiplies impact, confidence, and ease, the report noted.
"Consultations about a new service usually arrive in the form of 'can this be built?' But there is a question to ask before building. Will it be stopped? What ICE asks is impact, confidence, and ease. Confidence in particular is the column where risk is appraised. Multiply the three and the first shape to release is decided. The day a company that only researched feasibility could appraise its risk, what changed was not the level of the technology but the very perspective of picking a starting point by impact, confidence, and ease."
Related Files
Tools Used
- ROI Polygraph — Visualizing post-writing labor, cancellation losses from poor retention, and development stalled by unappraised risk
- ROI Proposal Generator — Investment-recovery simulation for an SNS auto-posting system built from selecting a starting point by impact, confidence, and ease
