What are we willing to have AI do for us? Meet The Human-AI Risk-Reward Matrix (HARRM)
The Ipsos Consumer Tracker, fielded on Ipsos' Omnibus platform, asks Americans questions about culture, the economy and the forces that shape our lives. Here's one thing we learned this week.

Why we asked: For the latest issue of Ipsos’ award-winning magazine, What the Future: Shopping we got to wondering about the agentic futures potentially on our present and future horizons. We were thinking about the meme about how people want AI to do the dishes so that they have more time to be creative, they don’t want AI to do the creative bits. And yet… we got to thinking about that tension between things you enjoy doing (or don’t) and things you trust AI to do (or don’t). Trusting AI, of course, is very different from WANTING AI to do something; that’s a matrix for another time.
But thinking through those two tensions, we plotted them on a 2x2 grid that we call the Human-AI Risk-Reward Matrix (HARRM). What we found in the Shopping issue was that for mostly shopping-related tasks, no items dropped in the Don’t Like/Do Trust quadrant. That’s amazing white space for marketers and developers today, and we also expect that as people grow more accustomed to AI and agents, more things will slide up into that quadrant of opportunity.
It got us wondering if we could come up with some things that would land there today if we added some items to the HARRM matrix.
What we found: Nope. Not at all. We put 10 more items through, including things that feel kind of low-risk and that we really don’t like doing. Looking at the chart, it’s clear we tried really hard. All but one item landed on the Don’t Like side of the chart. But also, all but one item landed in Don’t Trust.
The one item that bucked the trend … is not really agentic. We saw in the Shopping issue that people are much more trusting of recommendations than they are of having an AI actually execute a decision. It’s perhaps not surprising that recommending gifts for friends or family is something we’d trust AI to do. However it also landed (even farther) on the Like side of the ledger.
So still no magic Don’t Like/Don’t Trust activities that we’re already ready to hand over to AI agents. Stay tuned, that could well change. Or it could get exacerbated if early agentic experiments don’t land with consumers.
More insights from this wave of the Ipsos Consumer Tracker:
The Ipsos Creep-o-Meter: What Americans think is creepy vs. useful in online life
Investing is still seen as best way to build wealth
Holiday happiness is mixed with significant stress
Four in ten say they are hoarding more in case of emergencies
The Ipsos Care-o-Meter: What does America know about vs. what does America care about?