Mauricio Tapia.
Course index

Chapter 13 / 14 · Apply it to business

Evaluate quality, time, cost and contribution

A convincing demonstration is not a comparison with the work being improved.

Reading: 3 min · Suggested practice: 15–25 min

By the end of this chapter

Design an evaluation that supports a decision on continuing a pilot.

In this chapter

Goals become criteria

Define report requirements: all IDs, exact dates, sources, open questions and actions within permissions. Separate critical errors from presentation defects. Attractive formatting does not offset an invented order.

Evaluations use cases and criteria to check behavior. This course supplies a business rubric rather than mandating a tool. Manual review, automated checks and additional evaluators can be combined.

Representative cases

Include normal, incomplete, conflicting and duplicate cases and tool failures. Keep some cases aside for new-version checks rather than tuning prompts to every example. Process experts define expected answers.

Initial samples reveal defects without proving universal reliability. Record sample size, selection and limits. One difficult case does not support broad conclusions.

Complementary measures

Date accuracy is correct fields divided by reviewed fields; coverage is treated IDs divided by input IDs. Total time includes preparation, execution, review and correction. Include service, maintenance and operating costs when known.

In a fictional sample, 18 correct dates out of 20 equals 90%. If both errors create false commitments, the average cannot justify approval. Zero critical errors in a pilot sample can be an acceptance condition without guaranteeing future perfection.

Productivity and business value

Four released weekly hours are capacity. Realized economics need evidence of how they were used: more customer service, lower contracted effort or better fulfillment. Do not automatically convert hours to expected sales or attribute all business changes to AI.

Compare similar work and periods, document other changes and unintended effects and use a comparison group where feasible. Conclusions must match evidence.

Explicit decisions

Set continue, adjust and stop conditions before the pilot. Report denominators, failures and limits. Recheck relevant cases after changing instructions, models, sources or permissions. Evaluation continues during operation.

Worked case

Andina Equipos tests twenty briefs: all IDs covered, eighteen correct dates and two missing technical restrictions. Mean total time falls from 60 to 40 minutes. Speed improves but the no-material-omission criterion fails.

Adjust extraction and review, retest affected cases and keep manual operation meanwhile. Do not claim complete success or increased sales.

Practice instructions

Evaluate coverage, date accuracy, evidence, restrictions and permissions.
Report numerators, denominators and critical errors by category.
Measure total time and separate released capacity from realized economic benefits.
Recommend continue, adjust or stop against pre-existing criteria.

Your turn

Create a ten-case, five-criterion rubric and a stopping error. Explain what to do if time falls 25% but two false commitments appear, and what to test next.

Show the commented solution

False commitments can stop a pilot regardless of speed. Prevent converting expectations into agreements and verify supporting evidence. Retest difficult and normal cases to detect new omissions. Report sample limits instead of claiming general reliability.

Evaluate your work

  • Criteria precede measurement.
  • Denominators are explicit.
  • Critical errors remain visible.
  • Economic value has independent evidence.

Sources and technical reading

Sources support technical concepts. Business cases, rubrics and practice instructions are teaching proposals developed for this course.

Before moving on

Check your exercise against the criteria. Keep sources and your changes: they will support the final project.

Editorial review: September 29, 2026 · All case examples and figures are fictional. Study times are estimates.