evidoria

← Back to browse

Good practice

Santa by Riiid — AI-adaptive TOEIC assessment and score-prediction platform

South Korea · Seoul · See the South Korea profile

Seoul-based Riiid Inc. applies deep reinforcement learning to adaptive TOEIC prep: score predicted in 6–10 questions; 2.5 million users (TechCrunch 2021); beta study 2014–17 (n=1,500, 7.4 h) found 56-point improvement; later cohorts averaged 165 points gain after 20 h.

56 points
Average TOEIC score improvement (beta study) (2014-2017, mean 7.4 hours of study, n≈1,500)
165 points
Average TOEIC score improvement (later cohorts, CEO-reported) (after 20 hours of study, cited May 2021)
2.5 million users
Confirmed users (Korea and Japan) (as of May 2021)
0.7914 AUC
Prediction accuracy of SAINT+ knowledge-tracing model (LAK21, 2021)

Details

Maturity
Established
Promoter
Riiid Inc.
Period
2017–present
Keywords
edtech, standardised testing, adaptive assessment, AI

Context

Riiid Inc. was founded in Seoul in 2014 and launched Santa in 2017 as an AI-powered preparation platform for the TOEIC (Test of English for International Communication), the most widely taken English proficiency examination in South Korea and Japan. The platform uses deep reinforcement learning and knowledge-tracing algorithms to model each learner's competence profile, predict their current TOEIC score from as few as six to ten answered questions, and generate a personalised daily study plan targeting identified weaknesses.

Activities

In 2020 Riiid expanded internationally with Riiid Labs (USA) and partnered with Kaplan for GMAT preparation. The company rebranded as Socra.ai in September 2025 while the Santa consumer app continues operating; the basic level-test and study plan are free, with premium daily features on subscription.

Results

A company-sponsored beta study (2014-2017, n≈1,500) recorded an average 56-point TOEIC score improvement after a mean of 7.4 hours of study. CEO Young-jun Choi cited average gains of 165 points after 20 hours in a Korea Herald interview (May 2021), drawing on over 300 million TOEIC learning interactions. TechCrunch independently confirmed 2.5 million users in Korea and Japan around the time of Riiid's USD 175 million SoftBank Series D round (May 2021). The EdNet dataset released by Riiid, covering 784,309 unique users, underpins the peer-reviewed SAINT+ knowledge-tracing paper (LAK21, 2021), which reports an AUC of 0.7914 for the platform's core prediction technology.

Conclusions

Santa is a long-running, internationally-expanding commercial platform with a large user base and an independently-confirmed funding/adoption history, and its core prediction technology has peer-reviewed academic grounding. However, its headline score-improvement figures come from company-sponsored or CEO-reported sources rather than an independent outcome evaluation.

Implementation
Indicative cost
Low (< €50k) — Consumer-facing freemium model: basic level test and study plan are free; premium daily features are subscription-based. Corporate-level funding (USD 175 million Series D, May 2021, led by SoftBank) capitalised the company but is not a per-deployment implementation cost.
Time to results
Long (> 3 years) — Founded 2014, Santa launched 2017, international expansion (Riiid Labs USA, Kaplan partnership) from 2020, USD 175M funding round May 2021, corporate rebrand to Socra.ai September 2025 with the Santa app continuing to operate.
Staffing & skills
Riiid Inc. (Seoul, platform developer and operator), Riiid Labs (USA, established 2020 for international expansion), Kaplan (GMAT-preparation partnership, from 2020)

Conditions for success

  • Large underlying training dataset (over 300 million TOEIC learning interactions; EdNet dataset of 784,309 unique users) enabling robust knowledge-tracing models
  • Peer-reviewed algorithmic foundation (SAINT+, LAK21) lending technical credibility beyond marketing claims
  • Significant venture funding (USD 175 million SoftBank Series D, 2021) supporting international expansion and partnerships

Common failure modes

  • Headline score-improvement figures (56-point beta study, 165-point CEO claim) are company-sponsored or self-reported rather than independently verified, so the true causal learning effect is uncertain
  • The peer-reviewed academic evidence (SAINT+) validates prediction accuracy of the underlying algorithm, not that using the platform causes the claimed score gains

Where it fits

Governance type
private commercial edtech company, consumer-facing product
Scale
international (Korea, Japan, expanding to USA via Riiid Labs)
Income level
high-income (Korea, Japan, USA)

Data sources

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

Similar practices you may find useful