Four companies released surveys on November 10 explaining why AI investments fall short. Each one reaches a different diagnosis, and each diagnosis matches what the sponsor sells. That does not make the data worthless. It means the surveys work best read together, as four views of one problem, with the sample sizes in front of you.
FranklinCovey: leaders are not telling people the truth
FranklinCovey (NYSE: FC) asked workers in a June 2026 global survey which leadership behaviors they want during AI adoption, and how many get them. The top request was honesty about how AI will change their work. Only 36% say they receive it. 46% get access to AI learning, 59% have a safe space to experiment, 41% get help connecting AI to their own workflows and 46% hear a clear vision.
The release does not give a sample size, and the full study is due later this fall. FranklinCovey sells leadership training, including an AI readiness program launched alongside the findings.
Latch: the AI is missing context
Latch surveyed 742 US working professionals for its State of AI Automation at Work 2026. Among those whose AI project fell short, 57% blamed missing context, such as bad data or a system that did not understand how the work is actually done. Only 8% said the technology was not capable enough. Respondents estimate only 41% of their recurring work is documented well enough to hand off.
Latch launched an accounting automation product the same day that learns tasks by watching staff and asking questions.
Convertr: governed data triples the hit rate
Convertr surveyed 176 marketing and revenue operations leaders at US companies. 84% of teams with documented data governance hit or exceeded their expected AI return, against 27% of teams with significant governance gaps or none at all. Convertr sells contact data governance software. The sample is small and limited to go-to-market teams, and teams organized enough to document governance may simply be better run across the board.
Federato: legacy core systems block the gains
Federato surveyed 245 London Market insurance professionals across Lloyd’s managing agents, company market insurers, MGAs, reinsurers and brokers. 58% already use AI in production, yet 76% expect their core systems will need significant integration, modernisation or transformation to support future AI and underwriting needs. Only 24% think gradual improvement of current systems will be enough, and 89% say operational speed cost their organisation business in the past year. Federato sells an AI-native core platform for insurers.
What the four agree on
Set the sponsors aside and the overlap is plain. None of the four finds the models themselves to be the main constraint. Latch’s 8% figure says it directly. The bottleneck sits in what surrounds the model: how work is documented, how clean the data is, what systems the AI must plug into, and whether managers explain what is changing. That matches Lenovo’s workplace readiness study, released the same day, which found only 10% of companies have redesigned the workplace around AI.
Where the surveys disagree is on which fix comes first, and that is the part most shaped by who paid for the research.
