AI for Business Entrepreneurs

Entrepreneurs in emerging economies often lack access to affordable, contextually relevant business advice. This study tests Ujuzi (“knowledge/skill” in Swahili), an AI business advisor chatbot, with informal-sector market traders in Ghana, using a three-arm individually randomized controlled trial. We seek to answer the question: does culturally localizing an AI advisor’s underlying values change its adoption, use, and downstream impact, beyond simply offering entrepreneurs any AI advisor at all? Participants are randomized (balanced by product type and market size) into three groups: a waitlist Control (no advisor during the study; they get Ujuzi once the study ends), a Western-framed advisor (direct, transactional, focused on individual profit and fast decisions), or a Ghana-localized advisor (relationship-first, consensus-based, drawing on Ubuntu, Nyansapo, and Sankofa). Both advisor arms share identical underlying knowledge and business content – access to capital, competition, supplier challenges, operation challenges, help with bureaucracy, managing employees and dispute resolution. What differs is the values behind the advice: Western_LLM leans toward individual profit and fast, transactional decisions; Local_LLM leans toward preserving relationships and building consensus. Participants in the two advisor arms get an in-person demo in week 1, an in-person check-in in week 2, and phone check-ins in weeks 3 and 4. Specifically, participants receive in-person check-in to support or address questions, the next week a phone check-in and the last week an in-person check (identical structure across arms). All three arms get an in-person endline survey about four weeks after onboarding for the two advisor arms, or the same point in the timeline for Control. Everyone answers questions on confidence, business performance, and life satisfaction, since H6 needs those to compare all three arms. Only the two advisor arms answer questions on usage, trust, and cultural fit, since Control never had an advisor to use or judge. We expect the localized advisor to be adopted and used more, trusted more, and to produce larger self-efficacy and business gains than the Western-framed advisor, with both advisor arms beating the no-advisor control.