ROI Case File No.452 'The Silence the Camera Watched'
The Silence the Camera Watched
Chapter 1: The Answer in the Footage
"We have the camera footage. We just don't know what to look at."
Koichi Suzuki, Head of International Operations at TransGlobal Retail, set a tablet on the table as he spoke. The screen showed an overhead view of an overseas store: staff moving, a queue forming, registers processing — ordinary scenes, playing on loop without sound.
"It's been six months since we piloted AI cameras at three stores. The video data is piling up. But neither the floor managers nor headquarters can interpret what it means. The board is asking me to show results that justify the investment."
"What was the original purpose of the installation?" I asked.
"Improving and standardizing store operations. Longer-term, we hoped for sales improvement and higher QSC — quality, service, cleanliness. But—" Suzuki gave a wry smile. "We never clearly defined what was actually wrong with the stores before putting the cameras in. We should have decided what questions to answer before the cameras went in."
"The footage exists, but no question exists," Claude said quietly. "That's the same as having an answer with no problem."
Suzuki nodded. "Exactly. I need help figuring out where to start."
Chapter 2: How NPS Creates the Question
"This case calls for NPS."
Gemini drew a simple number line on the whiteboard — minus one hundred to plus one hundred.
"NPS — Net Promoter Score," I began, "asks a single question: 'How likely are you to recommend this store to a friend or colleague?' Customers answer on a scale of zero to ten. Subtract the percentage of Detractors — those who score six or below — from the percentage of Promoters — those who score nine or above. That's your NPS. Comparing what differs between high-NPS and low-NPS stores is what finally makes operational problems visible."
"Combining camera data with NPS," Suzuki repeated.
"More precisely: NPS creates the question, and the camera provides the answer," Claude continued. "First, run an NPS survey across all three stores. Ask Promoters and Detractors to put their reasoning into words. Those words will reveal which locations and which times of day the cameras need to examine."
"Let's use KOE-Score," Gemini proposed. "It collects customer feedback as text and automatically analyzes sentiment — positive and negative — along with key themes. It dramatically reduces the work of turning NPS numbers into language."
[NPS Survey Across Three Stores]
Three weeks later, the NPS data from all three stores was in. When entered into KOE-Score, a striking pattern emerged.
Store A: NPS +32. Store B: +11. Store C: -4.
"Look at the Detractor comments from Store C," Gemini said, pointing at the screen. "The most common keyword is 'wait time.' Second is 'can't find a staff member.' Third is 'order takes too long to arrive.'"
"And the Promoter comments from Store A?" I prompted.
"'Staff come right away,' 'no waiting' — a mirror image of Store C's complaints."
"Meaning," Claude summarized, "what the cameras need to watch is staff movement patterns and response times. At which times of day at Store C does staff flow break down? Once we know that, we have a target for improvement."
Chapter 3: The Silence Begins to Speak
"Let's use ROI Polygraph to design the camera analysis," I proposed. "We'll direct the questions NPS gave us at data that's been recording without direction for months. Location, time of day, metrics to measure — nail those three first, and the footage yields answers."
Entering the three stores' operational hours and peak customer arrival data, the tool surfaced Store C's bottleneck. During the lunch peak, seventy percent of staff were concentrated around the registers — the floor was understaffed. The times of day matched precisely with the timestamps on Detractor comments.
"The camera had been filming the answer all along," Suzuki murmured. "There was just no question."
"That's the point of NPS," I replied. "Customer voices tell you where on the floor to look. The camera records evidence — but for evidence to testify, there has to be a question first."
[Designing the Fix]
"Let's design improvements for Store C," Gemini continued. "Revise the shift allocation during the lunch peak to add one floor coverage staff member. Verify the impact the following week using camera footage and a new NPS snapshot. Running that cycle monthly will drive the three stores' operational quality to converge."
"On the gap between Store A and B," Claude added, "Store B's NPS of +11 is average, but Promoter comments lack the 'staff are warm' keyword that appears at Store A. Cleanliness and wait times have improved, but service quality differs. That's a training issue, not a scheduling issue."
"In other words, even though both stores need to raise NPS," Suzuki said, working it through, "Store C is an operations problem, Store B is a training problem — the prescription is different for each store."
"That's what separates NPS from a simple satisfaction survey," Claude said. "Looking only at the number makes you want to apply the same fix to every store. Breaking down the customer voice reveals a bespoke solution for each location."
Chapter 4: The Day the Camera Testifies
Suzuki took notes in silence, then said quietly:
"For six months the footage was piling up and I couldn't say anything. I was afraid to report to the board. But today, for the first time, I feel like I can speak in data."
"NPS is also a language for communicating with leadership," I replied. "When demonstrating ROI, a concrete number — how many points did customer advocacy score rise — reaches the board better than impressions. What proves the value of the camera investment isn't the volume of footage. It's the magnitude of the NPS change."
"Set a target for Store C: bring NPS from -4 to +15 or above by end of quarter," Gemini proposed. "Put in a mid-point measurement two weeks after the shift change, and track the numbers monthly. Show that trajectory to the board, and the investment justifies itself."
Suzuki stood, and bowed his head. "Thank you. I'll contact all three store managers this week and get the NPS surveys running."
Three months later, a report arrived from Suzuki.
Store C's NPS improved from -4 to +19. The shift change showed up in the numbers in week two. "Can't find a staff member" complaints fell to less than a third. Store B improved from +11 to +28 following a new training program. Store A maintained its +32 and ticked slightly higher still after late-night operational improvements.
In the board report, NPS improvement data was presented alongside before-and-after camera footage. Suzuki wrote in his report: "The cameras had been filming the answer for six months. When we came to them with a question, the footage finally started talking."
The day the silent camera finally gave its testimony.
"Data doesn't answer. What answers is not the data — it's the question itself. What NPS provides is the design blueprint of customer voice as question. The camera records phenomena; NPS asks about feeling. When those two combine, footage acquires meaning for the first time. Data without a question is a silent witness — it knows the truth but won't speak it."
Related Files
Tools Used
- KOE-Score — Customer voice text analysis and sentiment scoring
- ROI Polygraph — Camera data analysis design and time-of-day bottleneck visualization
