Case Study

Deciding Whether to Add a Commercial OCR Module

Case context: Your product team is thinking about spending several weeks connecting a paid OCR service to recognize text in scanned receipts. A few teammates want to begin right away because they expect it to eliminate many of the current recognition mistakes.

Question: As the technical lead, how would you apply error analysis to decide whether this integration is worth doing, and what would you compare before making the final call?

Sample answer: I would first review a representative set of recent mistakes and count how many of them come from poor text recognition on receipts. That count gives a practical estimate of the largest performance improvement the OCR service could deliver. If the expected gain is large enough, then the integration may be worth the engineering time; if the gain looks small, I would prioritize work on the error sources that appear more often or are cheaper to fix.

Key points:

  • Use error analysis to estimate the likely benefit of the integration.
  • Tally how many existing errors are caused by the targeted problem.
  • Treat the result as a quantitative estimate of value.
  • Compare the expected gain with the time required and with other higher-impact work.

Rubric: The response must describe checking the potential improvement before investing development effort, and it must say that the estimate should be used to judge whether the work is worth more than alternative tasks.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI