AI is not helping companies as much as we expected. According to a 2026 HBR Analytic Services survey of AI decision-makers, just 16% reported a "high degree" of measurable value from their AI investments. Why? One big reason: AI is used simply as a side tool.
Among the 385 respondents, only 18% said that AI had been built into most of the processes that use it. Those who reported getting more value from AI were twice as likely to have done so, yet still relatively low at 24% (vs. 12%). For research and marketing teams, I believe that the way forward starts with a simple change.
Stop asking which roles AI will replace, and instead look for the specific tasks that AI can do, ideally within a larger process. Here are steps a team can take this month:
- Pick a familiar project type (e.g., a concept test) with the goal of choosing a relatively low-risk discrete task within that workflow. Then list the steps from the brief to the final output such as a presentation.
- Ask three questions about each of the steps you identified:
- Can the task be executed by AI?
- Can the output be easily checked/verified by our team?
- For this task, what does a mistake cost?
- Sort the steps into three groups.
- AI leads where output is easy to check and mistakes are less costly, such as data cleaning or perhaps building tables.
- AI assists where a person reviews and decides, such as a questionnaire draft or a very early stage of the analysis.
- People lead where the output is harder to check and mistakes become more costly, such as study design, interpretation, or final recommendations.
- Write down who reviews what in a grid. In the survey, among orgs using or exploring AI agents, 92% said that agents needed rules-based guardrails, yet only 48% said that their org had defined them. This gets back to one of my previous posts on setting policies around AI. The sorting can feed directly into that policy.
- Measure outcomes. In research, the obvious one is completion time for AI vs. traditional approaches. In the HBR survey, most organizations tracked "efficiency" (57%), but this term is subjective. At a high level, possible KPIs include the number of projects completed per quarter, feedback from internal teams like marketing, or how often recommendations hold up in market.
- Continuously test your approach on future projects of the same type (i.e., all future concept tests). As your process improves for a specific task use case, the approach can be expanded to related tasks, such as product tests or trackers.
I call this "raising the floor" vs. "lowering the ceiling." Researchers become freer to focus on higher-value, thought-oriented work. This change probably benefits senior researchers more, as the tasks that AI can take over once filled most of a researcher's week. The tasks that remain depend on experience: deciding what the business needs to know next, what to study, and how results compare with previous work over time. With AI doing more rote tasks, an experienced researcher can now do what once took a team.
That advantage comes with a responsibility. In the same survey, 64% agreed that some tasks will never be appropriate for AI. Someone has to teach the next generation how to do them.