Product Teams Hide AI Use From Stakeholders While Building Trust Carefully
Product managers strategically conceal when they use generative AI for prioritization decisions, according to new research. Teams build trust in AI recommendations through trial-and-error rather than systematic processes—and trust varies wildly by individual, raising questions about consistency and accountability in resource allocation.
Originaltitel: Should Teams Trust GenAI for Product Prioritization?
**Produktteam behöver strukturerade rutiner för AI-stöd i prioritering** Produktchefer som använder generativ AI för resursfördelning måste bygga explicita förtroendeprocesser — inte förlita sig på AI:s rekommendationer utan kritisk granskning. En kvalitativ studie med åtta produkthanterare i Indonesien och Estland identifierade fem tilltroendepraktiker: AI som assistent, kalibrering genom AI-fel, informell verifiering, datakvalitetsbedömning baserat på individuell bedömning och gemensam granskning. Förtroendet varierar kraftigt mellan roller och expertis — det finns ingen universell acceptansnivå. Betydande fynd: organisationer skiljer mellan beslut om AI-användning och dess transparens. Många team döljer strategiskt AI-medverkan för att hantera intressenters uppfattningar. För offentlig sektor och kommuner innebär detta risker vid AI-stödda resursbeslut. Beslutsfattare måste kräva tydlig dokumentation av AI-inflytande och mänsklig ansvar, särskilt i prioriteringar som påverkar medborgare och samhällsresurser.
Product prioritization is a collaborative process that determines how resources and effort are allocated. While product teams adopt GenAI for recommendations, the factors that calibrate trust in AI outputs remain underexplored. This qualitative study examined how product managers describe team-level trust practices through semi-structured interviews with eight practitioners in Indonesia and Estonia. Using reflexive thematic analysis, this study identified five trust practices: (1) GenAI as an assistant, (2) calibration through AI errors, (3) informal verification, (4) data governance based on individual judgment, and (5) collaborative review. The findings indicate that trust levels are not uniform but vary by specific roles and team members' expertise. In addition, there is a distinction between the decision to use GenAI and its disclosure, which is done strategically to manage the perceptions of various stakeholders. These findings offer insights for future research on the social dynamics of AI integration within organizations.