Up to 50,000 handwritten forms a day was the plan. When almost 150,000 arrived, the bottleneck surfaced where nobody expected it.
The task was simple. For an Australian campaign, between 5,000 and 50,000 handwritten forms were expected to arrive each day and had to be processed within five hours — from the 11am mail delivery to the 4pm file submission deadline.
The challenge was equally straightforward. Open the envelopes and scan the mailed-in forms. Extract a mix of relevant data points — checkboxes, generic information, dates, addresses, personal information — within the timeframe and at high accuracy. Package the extracted data with a reference to the image file, and submit it for processing by the end client.
What we were working with
The forms were structured, fixed-format and printed, and they were always scanned at a consistent image quality.
The handwriting, however, varied significantly in legibility: neat block letters in clean lines, “doctor’s shorthand” or scribbles drawn by a shaky hand. It was all there.
The volatility in volume was hard to predict. The only expectation was that the first few days might be slower than the weeks to follow, with the final week of the four-week campaign tapering off again.
The customer’s plan was to keep costs under control, focus on mail processing and scanning, limit core staff attention to the tasks that mattered for compliance, augment the team with a few temporary staff for the peak weeks, and outsource whatever could be outsourced without issue.
The ScaleHub solution
Understand the form and the data points. Identify the complexity and risk around data quality, sensitivity and processing. Identify suitable means of processing with high efficiency, while maintaining maximum accuracy and compliance.
Fields were taken out of context — pseudonymised — so a date of birth became just a date, phone numbers became numbers, names were split, checkboxes were removed from their labels, and address blocks were cropped out of the page. What remained as direct PII (Personally Identifiable Information) were the email addresses.
From there, the work was routed three ways:
- Public Generic fields were assigned to a public microtasker workflow.
- Specialist Address blocks were assigned to a trusted specialist BPO (Business Process Outsourcer).
- Internal The sensitive email addresses were sent into a dedicated workflow for the customer’s internal core and augmented staff.
The P.S.I. model routes each data point to the workforce best suited to its sensitivity and value. Tap to enlarge.
P.S.I. — Public, Specialist, Internal — is designed to take pressure off internal staff when dealing with volatile volumes, or with high volumes of low-value but highly critical data entry, especially under time pressure.
ScaleHub’s managed service received the scanned images and orchestrated everything that followed: image registration, OCR extraction, evaluation, snippeting of data points, routing to P.S.I., stitching the verified results back together, and packaging everything in the requested format.
The challenge for ScaleHub
Only two weeks of preparation time, and no real samples for training the image registration engine or performing any meaningful testing.
We had mock-up samples with mock-up data. That was suitable for setting up and testing rules, but not very helpful for day one in production, and no help at all in identifying any unexpected exceptions or data combinations that could cause challenges for workers making a decision.
ScaleHub prepared as well as possible and was ready on day one to use the first real incoming forms to train the engine within the first hour on the job. Fine-tuning took place at the end of the day, after manually working through the exceptions encountered.
Low volumes meant slow reaction times from the public workflow, which caused slight panic for our customer, who had not experienced ScaleHub at scale before. Their immediate conclusion: if ScaleHub needs the maximum available time for such a low volume — just under 5,000 forms across day one and day two combined — what will that look like when the real anticipated volume hits?
Mentally, they were preparing for a tsunami.
That tsunami came, and it hit in a different way than anticipated.
The problem with scale
Scale is harder on people than most consider.
Remember the sensitive email addresses that were supposed to be processed internally? At the very beginning, the fewer than 5,000 email addresses received over two days were processed long before the jobs were ready for return by ScaleHub. But even with augmented part-time staff added to the team, that internal workflow did not scale beyond 30,000 forms per day. The internal staff became the bottleneck, despite single keying.
By comparison, here is what the ScaleHub side of the process for this campaign looked like:
- ScaleHub uses AI-enhanced OCR for data extraction, particularly for handwriting. Any OCR confidence level below 94% per field is rejected as an automated extraction result in favour of a human worker.
- For every extracted data point, we need two matching results from at least two independent workers — human or OCR — double keying.
- If a result is ambiguous, it goes to a tie-break, letting an experienced, highly ranked supervisor make the final decision.
- Finally, the entire form passed through a human QA check performed by ScaleHub if the extracted data did not fulfil the minimum requirements.
By week two, the customer was struggling with employee satisfaction and backlog pressure. Volume peaked at close to 150,000 forms on a single day. ScaleHub stepped in with a limited group from the specialist BPO to process the PII and cleared the backlog within a day. And, as it turned out, even at a slightly lower rate than the augmented staff cost in reality.
The result
After the initial unease of the first two days, the customer came to appreciate the consistent speed and quality regardless of volume. All turnaround and accuracy SLAs were met by ScaleHub.
The peak of almost triple the anticipated daily volume was caused by external factors — specifically, mail handling by the courier — creating immense pressure on the internal staff.
Experience pays. And first-hand experience leads to an appreciation of ScaleHub’s value.
Facing volumes you can’t predict? Talk to us about how P.S.I. routing can take the pressure off your internal team — or see how other customers scale.
