ROI Case File No.409: The Bottleneck Illuminated by Theory of Constraints
The Bottleneck Illuminated by Theory of Constraints
Chapter 1: The Toil Called Transcription
"We repeat the same work every day."
OCRTech's operations director said this with a tired expression. Before him were piled delivery notes and warranty certificates sent by customers, all containing handwritten information.
"Documents sent from customer sites contain handwritten information. Product names, quantities, dates, staff names—we visually confirm these one by one and transcribe them into the system."
The operations director picked up one delivery note showing "2024/11/15" written in somewhat distinctive characters.
"This work takes approximately three hours per day. Two staff members. Meaning six hours of human resources are spent daily on this transcription work."
His voice carried a tone resembling resignation.
"What's more problematic are transcription errors. Handwritten character misreading, input mistakes—to prevent these, another staff member visually checks. That checking work also takes two hours daily."
"Meaning, a total of eight hours. One person's work time consumed by transcription and checking."
The operations director spread out materials listing OCR tools previously considered.
"We consulted with multiple OCR companies at exhibitions. However, we were told handwritten character recognition accuracy was insufficient, and symbol identification was also difficult. And above all, we couldn't reach agreement on costs."
"Currently we use Windows standard marking tools, but accuracy is low, ultimately requiring manual rework."
The problem was clear, but solutions remained unfound. That was OCRTech's current state.
Chapter 2: The Weakest Link in the Chain
"For this case, the TOC Model approach is optimal."
Gemini drew a chain on the whiteboard with one clearly thin link in the center.
"TOC—Theory of Constraints," I began explaining, "is the idea that overall system performance is determined by the weakest part."
"A chain's strength is determined by its weakest link," Claude supplemented. "No matter how much you strengthen other links, if even one weak link exists, the chain breaks there."
The operations director tilted his head. "Then, what's the 'weak link' in our operations?"
"Identifying that is TOC Model's first step," I answered.
[Step 1: Identify the Bottleneck]
"Let's organize OCRTech's workflow," Gemini proposed.
The operations director began explaining. "First, documents arrive by mail from customers. We receive and sort them. Next, we attempt to read handwritten parts with OCR tools, but accuracy is low, so we ultimately transcribe manually. After transcription, another staff member visually checks, and if there are errors, corrects them. Finally, we register data in the system to complete."
"In this series of flows, which step takes the most time?" I asked.
"Manual transcription. Three hours daily × two people equals six hours. Next most time-consuming is visual checking at two hours."
"Meaning," Claude organized, "of the total eight hours, six are spent on transcription work. This is a clear bottleneck."
"What's more important," I pointed out, "is that when this transcription work's accuracy is low, subsequent checking work also increases. Meaning the bottleneck adversely affects following processes too."
The operations director nodded. "True, on days with many transcription errors, checking work can take over three hours."
"This is the TOC Model's core," Gemini emphasized. "Bottlenecks have ripple effects not just on their own problems but throughout the entire system."
[Step 2: Thoroughly Utilize the Bottleneck]
"Once you identify the bottleneck, next thoroughly utilize it," I explained.
The operations director looked puzzled. "Utilize? Not eliminate?"
"The TOC Model's interesting point," Claude answered, "is not trying to immediately eliminate the bottleneck, but first 'extracting maximum value from the current bottleneck.'"
"Specifically?"
"If manual transcription is unavoidable," Gemini proposed, "arrange the environment to perform that work most efficiently. For example, prioritize processing documents with readable characters. Eliminate factors disturbing transcription staff concentration. Design formats easy to transcribe."
"However," I added, "in OCRTech's case, a method exists to eliminate the bottleneck itself. That's introducing high-accuracy OCR tools."
The operations director smiled wryly. "But I mentioned we couldn't reach agreement on costs."
"That's because," I answered, "you haven't accurately grasped the 'cost' the bottleneck is generating."
[Step 3: Optimize the Entire System]
"To make investment decisions on bottleneck elimination requires accurate ROI calculation," Gemini began explaining.
"Currently, eight hours daily—one person's work time—is spent on transcription and checking. About 160 hours monthly. About 1,920 hours annually."
"Assuming that staff member's labor cost is 2,000 yen per hour," Claude calculated, "that's 3.84 million yen annually."
The operations director showed a surprised expression. "That much?"
"Furthermore," I continued, "including indirect costs like downstream process impacts from transcription errors, customer apology responses, reshipping costs—these probably total close to 5 million yen annually."
"Then, what's the introduction cost for high-accuracy OCR tools?"
The operations director checked materials. "Initial cost was said to be about 1.5 million yen, annual maintenance about 500,000 yen."
"Meaning first-year total cost is 2 million yen," Gemini organized. "Compared to current cost of 5 million yen, there's an annual reduction effect of 3 million yen."
"Cost-effectiveness is clear," Claude smiled.
"However," I cautioned, "introducing it doesn't automatically realize cost reduction. TOC Model's third step is 'optimize the entire system.'"
"What do you mean?"
"Even introducing OCR tools won't make reading accuracy 100%," I explained. "Probably about 95%. The remaining 5% needs manual correction."
"Meaning the entire workflow needs redesigning," Gemini supplemented. "OCR performs primary reading, people handle exceptions and checking, register in system—establishing this new flow and creating manuals is important."
[Step 4: Strengthen Support Systems]
"TOC's fourth step is building systems supporting bottleneck elimination," Claude explained.
"After introducing OCR tools, staff need time to master the new system. Please prepare training programs for that."
The operations director asked, "Specifically?"
"First, conduct training to learn basic tool operations," Gemini answered. "Next, perform trial operations using actual documents and record what types of misrecognition occur easily."
"And," I added, "based on those records, create 'guidelines to increase OCR reading accuracy.' For example, concrete directions like 'this type of handwritten character is easily misrecognized, so check intensively.'"
[Step 5: Continuous Improvement]
"The final step is continuous improvement," I emphasized.
"Even eliminating bottlenecks, new bottlenecks can appear," Claude explained.
The operations director thought deeply. "If transcription work is automated, what becomes the next bottleneck?"
"Probably exception handling," Gemini predicted. "Processing special characters and symbols OCR can't read. Or document sorting work. These could become new bottlenecks."
"The TOC Model doesn't end once," I emphasized. "Regularly review workflows, identify new bottlenecks, improve them. Continuing to turn that cycle is important."
Chapter 3: The One Who Masters Constraints
The operations director gazed at the chain diagram on the whiteboard.
"The essence of the TOC Model is 'seeing the whole.'"
"Exactly," I answered. "Many improvement activities fall into local optimization. 'Let's streamline this process' 'Let's automate this work'—but if that's not the bottleneck, overall system performance doesn't improve."
"For example," Claude showed an example, "even if document sorting work could be done in half the time, if transcription work remains the bottleneck, overall processing time barely changes."
"Conversely," Gemini continued, "improving transcription work as the bottleneck dramatically improves overall system throughput."
The operations director asked, "Then where should the first step begin?"
"First, I propose one month of pilot introduction," I answered. "Use a trial version of high-accuracy OCR tools to verify reading accuracy with actual documents."
"And," Claude added, "in that one month, record in detail how much time was reduced, what misrecognitions occurred, which process seems likely to become the next bottleneck."
"Those records become material for full introduction decisions," Gemini organized.
The operations director's expression brightened. "Understood. Starting next week, I'll begin pilot introduction preparations."
Chapter 4: Beyond Constraints
After he left, Claude said quietly.
"The TOC Model is simple but contains deep insights."
"Yes," I answered. "Many business problems seem like 'too many things to do.' But the TOC Model teaches—what's truly important is discerning that single constraint and eliminating it."
"And," Gemini added, "once that constraint is eliminated, the next constraint becomes visible. Improvement proceeds like climbing stairs, one step at a time."
Outside the window, winter sunlight illuminated the office.
One month later, a report arrived from OCRTech.
Pilot introduction results achieved 94% reading accuracy. Transcription work time reduced from six hours to 30 minutes, checking work also shortened from two hours to 45 minutes.
Moreover, document pre-sorting work had surfaced as a new bottleneck.
The constraint was eliminated, and the next constraint had become visible. That was exactly the path the TOC Model indicated.
"System improvement isn't simultaneously fixing all weaknesses. Discern the weakest link, the bottleneck, and concentrate there. Once eliminated, the next bottleneck becomes visible. The TOC Model teaches that certain path of breaking through bottlenecks one by one, named constraints."
