evidoria

← Back to browse

Good practice Imported

PAIbot — UNAM's AI Chatbot for Distance-Learning Admissions Support

Mexico · Mexico City · See the Mexico profile · See the Mexico City profile

Evidence: Observational / pre–post Top 89% 27/100 · Ask Evidence Copilot about this practice

UNAM's Dialogflow chatbot answers admissions questions for its distance-learning bachelor's programme. A peer-reviewed study found conversations rose 52% and correct answers rose from 0.4% to 47.1%, though incorrect answers also rose.

52.3 %
Increase in chatbot conversations, PAIbot 2.0 vs 1.0
0.4 % (4 of 998 questions)
Correct-response rate, PAIbot 1.0
47.1 % (378 of 802 questions)
Correct-response rate, PAIbot 2.0
23.4 %
Incorrect-response rate, PAIbot 1.0
37.5 %
Incorrect-response rate, PAIbot 2.0
4 cycles
Admission cycles PAIbot has run across
PAIbot — UNAM's AI Chatbot for Distance-Learning Admissions Support

Details

Maturity
Established
Promoter
Universidad Nacional Autónoma de México (UNAM) — CUAIEED
Period
2021–ongoing
Keywords
higher education, admissions, chatbot, distance learning

Context

UNAM's Coordinación de Universidad Abierta, Innovación Educativa y Educación a Distancia (CUAIEED) built PAIbot, a Dialogflow ES-based chatbot that answers applicants' administrative questions — deadlines, procedures, requirements — for the university's open and distance-learning bachelor's admissions programme (PAI).

Objectives

Provide accurate, always-available answers to applicants' administrative questions during the admissions process, reducing reliance on staff for routine queries.

Activities

A 2021 pilot (PAIbot 1.0) was followed by a redesigned PAIbot 2.0, which added a custom interface, multimedia responses and a user feedback mechanism. A peer-reviewed study compared conversation volume and answer accuracy between the two versions.

Results

Conversations rose 52.3% despite a 25% shorter operating window, and the correct-response rate climbed from 0.4% (4 of 998 questions) to 47.1% (378 of 802). The incorrect-response rate also rose, from 23.4% to 37.5%. PAIbot has now run across four admission cycles.

Conclusions

The rise in incorrect responses alongside correct ones is a real limitation, attributed partly by the study's authors to a changed feedback mechanism — this record reports that trade-off rather than smoothing over it.

Implementation
Indicative cost
Low (< €50k)
Time to results
Long (> 3 years) — 2021 pilot (PAIbot 1.0), redesigned to PAIbot 2.0, now run across four admission cycles.
Staffing & skills
UNAM CUAIEED (Coordinación de Universidad Abierta, Innovación Educativa y Educación a Distancia) development team

Conditions for success

  • Built on the Google Dialogflow ES platform, avoiding custom NLP infrastructure
  • Iterative redesign (1.0 → 2.0) adding a custom interface, multimedia responses and a user feedback mechanism
  • Scoped narrowly to administrative admissions queries for a specific programme (PAI) rather than general-purpose Q&A

Common failure modes

  • Incorrect-response rate rose alongside the correct-response rate (23.4% → 37.5%), partly attributed by the study authors to the changed feedback mechanism

Where it fits

Governance type
university-led
Scale
institutional (single university admissions programme)

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Replication kit

Reusable artefacts from this practice — as published by their sources.

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