Nova Esteira: AI agents from enrollment to teaching for 360k students

Turn a Board commitment into measured value in the core of a higher-education group serving about 360,000 students in Brazil. In June 2025 the Board approved a program with a goal of BRL 100 million in measurable impact (revenue and efficiency) by 2028, split into four value streams along the student lifecycle: Enrollment+, Collections+, Service+ and Teaching+.

Each stream has a business VP as sponsor and a technology leader as operational owner.

Goals for 2026:

1. deliver BRL 20 million of measured impact in the year; 

2. raise enrollment conversion with an AI sales agent; 

3. move debt renegotiation to self-service; 

4. resolve student requests with AI agents instead of human queues; 

5. cut the unit cost of academic production and grading; 

6. report every result with period, baseline, method and owner, without double counting.

BRL 35.9 million of measured impact in 2026 up to 5 October, against a BRL 20 million target for the year (179%): Enrollment+ 29.6, Collections+ 2.3, Service+ 3.8, Teaching+ 0.2.
Enrollment: AI interaction to enrollment conversion from 11.0% to 12.5%; enrollments in the two main entry routes +15.1% with 8.6% fewer applicants.
Collections: BRL 2.26 million net P&L effect, January to August. Service, brands with AI agents: bot containment from 13.3% to 44.0%, first-contact resolution from 27.7% to 49.5%, satisfaction from 33.7% to 58.6%.
Teaching: essay grading cost from BRL 4.90 to BRL 0.11 per essay across 34,286 essays; 96,000 simultaneous users on the first day of classes with no incidents.
2026 figures are pre-close and subject to year-end reconciliation with Finance.

AI: Anthropic and OpenAI models orchestrated through Amazon Bedrock on AWS. Built on them: AI sales agent on WhatsApp, AI student service agents, AI validation of academic requests, AI essay grading, an AI tutor and AI content creation inside the learning platform. CRM and service console: Salesforce. Messaging channel: WhatsApp. Student request workflows: Sydle. Data and analytics: Databricks and Power BI. Collaboration: Microsoft Teams. Core systems: in-house academic, financial and learning platforms, a self-service debt renegotiation portal, instant payment (Pix) at enrollment, and 13 automations in student financing and regulatory processes.

Four business VPs and three technology owners had to commit to one number, under rules that cut some of their own reported gains. The program runs on legacy academic and financial systems shared by 26+ brands, with peaks such as 96,000 concurrent users on the first day of classes and enrollment windows that do not move. Federal and state student-financing programs impose deadlines and document rules that every automation must follow. Mid-program, the contract with the partner building the pricing and quoting engine ended, forcing a re-plan without stopping enrollment. AI agents had to beat the 30% containment ceiling of the previous bot before scaling, and each rollout is gated by performance and token budget. Measurement was the hardest part: sales, collections and service gains overlap, so each case has one P&L line, and items were removed from the count when attribution could not be proven.

Initial situation: A higher-education group with 26+ brands and about 360,000 students in Brazil ran enrollment, collections, student service and teaching as separate initiatives, each reporting its own gains. Results could not be compared or added up. In June 2025 the Board set one goal: BRL 100 million in measurable impact (revenue and efficiency) by 2028, with BRL 20 million due in 2026. Constraints: legacy academic and financial systems shared by all brands, fixed enrollment windows, and federal and state student-financing rules.

Implemented solution: Four value streams along the student lifecycle, each with a business VP as sponsor and a technology owner. Enrollment+: an AI sales agent on WhatsApp that answers, qualifies and enrolls candidates, plus self-service enrollment and instant payment. Collections+: a self-service debt renegotiation portal and early-stage collection. Service+: AI agents replacing a linear bot, 13 automations in financing and regulatory processes, and academic requests handled with no human touch. Teaching+: AI essay grading, an AI tutor and AI content creation inside the learning platform.

Implementation specifics: The method mattered as much as the technology. Every case belongs to one stream and one primary P&L line; a predictive model never claims value, only the unit that acts on it does. Every number carries a class: observed result, partial estimate, avoided cost, estimated saving, operational indicator or volume. Gains that were not incremental or not attributable were removed, more than BRL 19 million in 2026. Agents scale only after beating the 30% containment ceiling of the previous bot, and each rollout is gated by performance and token budget. The hardest part was organizational: four VPs agreeing on one number under rules that cut their own reported gains.

Changes after implementation: Candidates enroll and pay without waiting for a person: self-service enrollments rose from 6.9% to 10.3% and instant payment reached 78.4% of candidate payments. About 75% of active defaulting students negotiate directly in the portal. In the brands with AI service agents, response time fell from 61 to 24 seconds. Staff time moved from repetitive financing tasks (40,407 hours a year freed) to exceptions. Management changed most: a transverse value office consolidates one program number, reported to the executive committee and the Board.

Brazil, nationwide. The program covers the group's 26+ higher-education brands, with campuses and distance-learning hubs across the country, serving about 360,000 students. Rollout is phased by brand. The AI sales agent runs in the 10 largest brands, with expansion to 18 brands and graduate programs on the roadmap. AI student service agents went live in two brands, one in the Northeast and one in Rio de Janeiro (July to September 2026), and are being extended to all brands.