Highdigenous Tech

digital fabrication

Kabakoo’s AI onboarding tutorial recorded higher first-action rates—not deeper conversations

By · · EN

In an A/B test across 827 learners, Kabakoo’s AI-generated tutorial recorded a 31% relative lift in first actions among message readers, while subsequent conversation depth was effectively unchanged. Other findings show why activation and longer-term outcomes need separate examination.

Kabakoo team members work on laptops at La Maison de l’Étonnement in Bamako in 2024. Context photograph.

An onboarding test with a specific result

Kabakoo’s AI-generated onboarding tutorial recorded a higher first-action rate than a human-recorded version, but not deeper conversations afterward. In an A/B test across 827 learners, among those who read the message, 14.4% receiving the AI version took a first action within 24 hours, compared with 11.0% receiving the human version—a 31% relative lift.

Subsequent conversation depth was effectively identical: 6.93 messages against 6.91. The experiment therefore offers a specific finding about the first action, rather than evidence of a longer conversation. It sits within Kabakoo’s work on its WhatsApp support agent.

A first challenge that ended without another attempt

A different encounter with the AI Mentor shows what happened at one learner’s first assessment. In March 2026, a learner in Bamako recorded the introduction video required for her first challenge in Kabakoo’s upskilling program. The Mentor assessed the submission, found that it did not meet the criteria, and rejected it. The learner never tried again.

Her experience concerns a rejected assignment and no further attempt; the onboarding experiment concerns a first action after a tutorial. These are distinct observations, not interchangeable measures of a learner’s progress.

Activity fell sharply in a cohort analysis

Kabakoo’s June 2026 cohort analysis recorded learner activity at 84% in week zero, 29% in week one, and fewer than 10% by week four. Those figures describe activity over time, separately from the onboarding experiment’s comparison of tutorial versions.

The tutorial result concerns action within 24 hours among message readers. The cohort analysis follows activity across weeks. Neither finding establishes the AI Mentor’s isolated contribution to longer-term livelihood outcomes: Kabakoo does not yet have causal Level 4 evidence for the Mentor on its own.

Longer-term findings belong to the broader program

Kabakoo’s longer-term evidence concerns its broader hybrid program, within which AI is increasingly important. Following learners over time and comparing them with randomized counterfactuals, Kabakoo found baseline-adjusted growth-mindset gains of about 0.14 to 0.25 standard deviations and an increase of 0.33 in the Kabakoo Behavioral Index.

In the July 2026 analyses, 46% of the clean comparison group remained in the lowest monthly income bracket, at or below 40,000 FCFA, compared with 38% of program participants. These are findings for the broader program, not the onboarding tutorial or AI Mentor alone.

Separating the contribution of AI from peers, facilitators and in-person experiences remains work ahead. The onboarding experiment’s higher first-action rate stands alongside unchanged conversation depth, declining cohort activity and longer-term findings with a different scope.

The Kabakoo newsletter

Research, field notes, and stories from Kabakoo in your inbox.