Evidence · strength 2 of 5 · searched to 2 Nov 2026
Does AI help with consulting and analysis work?
AI for consulting and analysis tasks may work for tasks that sit inside what the AI does well. Across 2 studies, time per task changed by −25.1% to −16.4%.
What field experiments with consultants and corporate teams found: faster, better work on tasks AI handles well, and worse answers on tasks just beyond it.
VerdictMay work for tasks that sit inside what the AI does well
Works for some
ImpactModerate gain
Across 2 studies, time per task changed by −25.1% to −16.4%.
Evidence strength2 of 5
We have low confidence in this estimate.
Cost per user a year$$
The studies used general AI assistants; business plans list about $20–$30 a user a month (checked 2 Nov 2026).
Pilot rating under our v0.3 rules. The rules are published on our method page but not yet independently reviewed, so ratings may change as the method is checked and new studies arrive. Our method.
What the studies found
Time (headline)
Moderate gain−25.1% to −16.4%- Dell'Acqua et al. (2026): −25.1% Time to reach the final question (first 17 of 18 tasks inside the frontier); -22.5% GPT + overview, -27.6% GPT only, about -25.1% on average.
subjects using AI outperformed those not using AI, completing 12.2% more tasks and completing them 25.1% more quickly on average while also delivering solutions of significantly improved quality.
- Dell'Acqua et al. (2026): −16.4% Time spent on the solution vs individuals without AI.
individuals with AI spent 16.4% less time than the control group, while teams with AI spent 12.7% less time.
Output
Gain- Dell'Acqua et al. (2026): +12.2% Tasks completed, 18 consulting tasks inside the AI frontier.
subjects using AI outperformed those not using AI, completing 12.2% more tasks and completing them 25.1% more quickly on average while also delivering solutions of significantly improved quality.
- Dell'Acqua et al. (2026): +9.2 points Chance of a top-10% solution, team with AI vs control mean of 5.8%.
teams with AI were 9.2 percentage points more likely to produce solutions in the top decile compared to the control mean of 5.8%
Quality
Gain- Dell'Acqua et al. (2026): +29.9% Graded quality of solutions inside the frontier (1.482 (SE 0.061) and 1.307 (SE 0.067) points on a 1-10 grade, control mean 4.375; p < 0.01 (Table 4, col. 1)).
GPT + overview leads to a 1.48 increase in scores over the control mean of 4.37, representing a 33.9% increase, with GPT only producing a 1.31 (or 29.9%) increase.
- Dell'Acqua et al. (2026): −19 points Correct solutions on a complex managerial task chosen to be outside the frontier (84.5% correct (control) vs 60% (GPT + overview) and 70.6% (GPT only); coefficients -0.245 (SE 0.054) and -0.139 (SE 0.053), p < 0.01 (Table 7, col. 1)).
for a complex managerial task selected to be outside the frontier, subjects using AI were 19% less likely to produce correct solutions compared with those without AI
- Dell'Acqua et al. (2026): +0.37 SD Expert-rated solution quality, individual with AI vs individual without AI (SE 0.106 (Table 2, col. 1)).
individuals with AI demonstrate a 0.37 standard deviation increase (p < 0.01), while teams with AI show a 0.39 standard deviation improvement (p < 0.01).
- Dell'Acqua et al. (2026): +0.39 SD Expert-rated solution quality, two-person team with AI vs individual without AI (SE 0.122 (Table 2, col. 1)).
individuals with AI demonstrate a 0.37 standard deviation increase (p < 0.01), while teams with AI show a 0.39 standard deviation improvement (p < 0.01).
Wellbeing
Moderate gain- Dell'Acqua et al. (2026): +0.46 SD Positive emotions (excitement, energy, enthusiasm), individual with AI vs control (p < 0.01). Self-reported.
individuals with AI showed a 0.457 standard deviation increase in positive emotions
A minus sign on time means the task took less time. Bands are provisional; see how impact is banded. Effects are not pooled yet: each band is set from one result per study, and where studies use different units we show only the direction.
How sure are we?
- Starting level3 of 5, from 2 grade A or B studies.
- Down one: old models0 of 2 studies show they used a model released since May 2025 (the 18-month window); the rest used older models or do not say which.
- Consistencywith 2 studies, never downgraded on this test alone.
- Publication biasnot testable with fewer than 10 studies.
- Result2 of 5. We have low confidence in this estimate.
Behind the average
- Bottom-half vs top-half skill (assessment task) (Dell'Acqua et al. (2026)): Both halves improved with GPT-4; the bottom half gained more. No percentages in the published text.
- R&D vs Commercial professionals (Dell'Acqua et al. (2026)): Without AI each function leaned to its own kind of solution; with AI both produced balanced solutions.
- Teams without AI (benchmark) (Dell'Acqua et al. (2026)): Teams without AI were +0.24 SD over individuals without AI, about the same as individuals with AI.
How, where and why it works
Why it works
2 of 4. One experiment suggests AI lets people work beyond their own speciality, so individuals with AI did about as well as teams without it.
For whom and where
3 of 4. Task type matters most: on a task chosen to be just beyond the AI's abilities, people with AI were less likely to get it right.
What it takes
1 of 4. Studies gave short training; they say little about rollout over months.
The full cost
1 of 4. Time includes checking within the task; costs are not counted.
Four separate scores from 0 to 4, never added up: how much the studies tell us about why it works, for whom, what it takes and what it really costs (the EMMIE method).
What to do next
Measure your pilot →If you are trying it yourself
- Test the AI on a past task where you know the right answer.
- Use it to widen your options, then check the reasoning yourself.
If you are rolling it out
- Map which of your tasks are inside and outside what the AI does well.
- Pilot on real tasks with expert, blind grading.
What it takes
- Teach people where the AI is weak, not only how to prompt it.
- Keep an independent check on conclusions the AI helped reach.
What can go wrong
- On a task just outside the AI's abilities, consultants using AI were about 19 percentage points less likely to be correct.
- Work can look polished and still be wrong.
The studies
| Study | Grade | Design | People | AI model (released) | Company role |
|---|---|---|---|---|---|
| Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality Fabrizio Dell'Acqua, Edward McFowland III, Ethan Mollick, Hila Lifshitz, Katherine C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, Karim R. Lakhani, 2026. Organization Science 37 (journal) | B | RCT, realistic task | 758 knowledge workers (Boston Consulting Group) | GPT-4 (Mar 2023) | none |
| The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork Fabrizio Dell'Acqua, Charles Ayoubi, Hila Lifshitz, Raffaella Sadun, Ethan Mollick, Lilach Mollick, Yi Han, Jeff Goldman, Hari Nair, Stewart Taub, Karim Lakhani, 2026. Organization Science 37 (journal) | B | field RCT, realistic task | 791 professionals (published abstract); working paper: 811 participants, analyses on 776 | GPT-4 via Microsoft Azure (Mar 2023); GPT-4o (July workshop) (May 2024) | none |
Grades: A, a well-run randomised trial in real work; B, a strong natural experiment in real work or a well-run trial on realistic tasks; C, a trial with clear limits or a matched comparison; D, before-and-after, surveys or case studies. How studies are graded.
Does the rating hold up?
- Run without company-run studiesNo change: no such studies.
- Run without working papers and preprintsNo change: no such studies.
How up to date is this evidence?
Method →Evidence searched to 2 Nov 2026. Studies cover models released Mar 2023 to May 2024. Rating computed 5 Nov 2026 under Module 15 v0.3 (pilot).
How this page was made: AI research agents searched for studies and read each paper; a second pass checked every number on this page against the paper itself, with the quote and where it appears. The rating is computed by fixed rules, not chosen. No person outside AI Adoption has reviewed it yet. Editorial owner: AI Adoption editors. Spot a mistake? Corrections.
Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work. No company pays for or has any say in a rating.