evidoria

← Back to browse

Good practice Imported

ICFES Cuts National Exam Grading Time 91.6% with AWS/Blend AI-Assisted Coding of Open Responses

Colombia · Bogota · See the Colombia profile · See the Bogota profile

Evidence: Observational / pre–post Top 56% 47/100 · Ask Evidence Copilot about this practice

Colombia's national exam body ICFES, working with AWS and Blend since 2022, automated coding of open-ended Saber-exam responses - cutting grading time 91.6% and results-delivery time about 50% - and used Amazon Rekognition to verify identity and proctor over 383,000 virtual Saber

91.6 %
Reduction in grading time for open-ended exam responses (2022-2025)
~50 %
Reduction in exam results delivery time (2022-2025)
383,547 tests
Virtual exams processed with Rekognition-based identity verification (2017-2025)
ICFES Cuts National Exam Grading Time 91.6% with AWS/Blend AI-Assisted Coding of Open Responses

Details

Maturity
Scaling
Promoter
Instituto Colombiano para la Evaluacion de la Educacion (ICFES)
Period
2017-2025 (AWS partnership since 2017; AI-grading transformation since 2022, announced Sept 2025)
Keywords
national examinations, automated grading, virtual proctoring, public-sector AI

Context

Colombia's national testing agency ICFES administers large-scale standardized exams (Saber 11, Saber Pro, TyT Saber Pro) and partnered with AWS and Blend since 2017, expanding since 2022 to use AI for automatically coding open-ended exam responses.

Objectives

To reduce grading time, optimise resources, and speed up delivery of results while using AI (Amazon Rekognition) for remote identity verification and proctoring.

Activities

AI-assisted coding of open-ended responses; Rekognition-based proctoring and identity verification for virtual exams (383,547+ tests processed); a geolocation algorithm for exam-site assignment; and a Data Lake with generative-AI report generation.

Results

Grading time was cut by 91.6% and results-delivery time by roughly 50%, according to the September 2025 announcement.

Conclusions

The reported gains come from the implementing companies (AWS/Blend) rather than an independent audit, and no human-review process for the automated grading itself (beyond identity verification) is described.

Implementation
Indicative cost
High (€500k–€5M) — Multi-year AWS partnership (since 2017) with cloud infrastructure (Rekognition, Data Lake) supporting a national exam system; cost not itemised in public sources.
Time to results
Long (> 3 years) — AWS partnership began 2017; AI-assisted grading transformation phased in from 2022; publicly announced September 2025.
Staffing & skills
AWS cloud engineering/partnership team, Blend implementation team, ICFES exam operations staff

Conditions for success

  • Partnership with established cloud/AI vendors (AWS, Blend)
  • Geolocation algorithm for equitable exam-site assignment
  • Data Lake infrastructure for report generation

Common failure modes

  • Reliance on vendor-reported metrics without independent audit
  • No disclosed human-review procedure for automated grading itself (only identity verification)

Where it fits

Governance type
national government agency
Scale
national
Income level
upper-middle-income

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful