TECH Nov 02, 2026 · by Skills Team

AI Investment Returns: The 8% Reality Check for Buyers

Only 8% of asset managers report measurable improvements in investment returns from AI, even as 55% have adopted it. This buyer guide separates efficiency, insight, and returns so you approve AI budgets on the right metric.

AI Investment Returns: The 8% Reality Check for Buyers

The Short Answer

When a vendor promises that artificial intelligence will improve your investment returns, the numbers now exist to test that claim. The September 2026 Morgan Stanley and Oliver Wyman BluePaper, titled "Breaking from the Pack: The Race for Above-Market Growth", cites the Mercer 2026 AI in Asset Management Survey and reports that only 8% of asset…

When a vendor promises that artificial intelligence will improve your investment returns, the numbers now exist to test that claim. The September 2026 Morgan Stanley and Oliver Wyman BluePaper, titled "Breaking from the Pack: The Race for Above-Market Growth", cites the Mercer 2026 AI in Asset Management Survey and reports that only 8% of asset managers have seen measurable improvements in AI investment returns. That figure should change how every buyer evaluates an AI pitch.

This is not an anti-AI story. Adoption is real and accelerating: 55% of asset managers surveyed had already integrated AI into at least one part of the investment process, and another 27% were testing it. Only 18% reported no integration at all. The point of this article is measurement. A tool that makes research faster is valuable, but calling that value "returns" is how budgets get approved on the wrong metric. If you buy AI on a returns promise, you need a returns metric, a baseline, and a review date, before you sign.

What the survey actually found about AI investment returns

The Mercer survey, conducted in February 2026 among 131 asset managers globally, shows broad adoption and narrow financial payoff: a productivity story that has not yet become a performance story.

Reported benefitShare of asset managers
Improved operational efficiency69%
Better insights and decision-making55%
Reduced costs24%
Measurable improvement in investment returns8%
Reduced risk or volatility8%

Read that table the way a buyer should. Nearly seven in ten firms say AI made them more efficient, and more than half say it improved their insights. Fewer than one in ten can point to better returns or lower risk. The BluePaper states it plainly: "Notably, investment returns have not yet seen widespread impacts from AI, with just 8% of firms reporting measurable benefits." Faster research and improved productivity, the report cautions, do not necessarily translate into financial gains.

Three Kinds of Gain, and Why Buyers Must Separate Them

Most AI sales conversations blur three different kinds of gain into one promise. Separating them is the most useful discipline a buyer can adopt, because each is measured differently.

1. Efficiency gain: doing the same work faster

Efficiency is where AI is winning today. Summarizing earnings calls, cleaning datasets, drafting research notes, and automating routine reporting all save analyst hours. The 69% efficiency figure is credible precisely because it is easy to verify: hours saved, reports produced per analyst, and turnaround time are all countable. If a vendor promises efficiency, ask for hours saved per week and check whether that time was redeployed to something valuable or simply absorbed.

2. Insight gain: seeing more, and seeing it sooner

Insight is the second real gain. AI can scan more companies, more filings, and more unstructured data than any human team, and 55% of managers say their insights and decision-making improved. But an insight is not a return. An insight only becomes a return when it changes a portfolio decision, and that decision beats the alternative. Ask any vendor claiming insight gains a simple question: show me three decisions your tool changed, and what happened next. If the answer is a demo of a dashboard rather than a decision log, you are looking at a research aid, not a returns engine.

3. Returns gain: beating the benchmark, measurably

Returns are the hardest claim and the rarest result. A returns gain means risk-adjusted performance above a stated benchmark, over a stated period, net of the cost of the tool itself. Only 8% of surveyed managers report it. That does not mean AI can never improve returns. It means that, as of 2026, almost nobody can prove it has, and any vendor claiming otherwise carries the burden of proof. Demand the benchmark, the period, and the net figure after fees. A backtest is not a track record.

AI investment returns depend on where the AI actually sits

The survey explains the gap with one revealing detail: where AI sits in the workflow. Adoption is heaviest far from the actual investment decision and lightest exactly where returns are made.

Stage of the investment processShare with AI integrated
Investment research49%
Data processing40%
Portfolio construction13%
Trade execution7%

This is the structural reason AI investment returns remain elusive. AI is mostly a research assistant and a data janitor today, not a decision-maker. Most firms describe their AI use as operational (74%) or as a co-pilot (69%), while only 6% say AI is used for decision-making, and only about 5% have given AI autonomous or semi-autonomous authority over an investment decision. A research tool that never touches a portfolio decision cannot produce a return. It can only produce better-informed humans, which is valuable, but it is a different purchase.

For buyers outside asset management, the lesson transfers directly. Map exactly where any AI product sits in your workflow. A tool that speeds up research, reporting, or data preparation should be budgeted and judged as an efficiency purchase. A tool that claims to change outcomes must be judged on outcomes, with a control group or a before-and-after baseline. Confusing the two is how organizations pay outcome prices for efficiency tools.

A Worked Example: How the Wrong Metric Approves the Wrong Budget

Consider a mid-sized investment firm weighing an AI research platform. The vendor's pitch leads with "AI-driven outperformance" and a backtested portfolio beating its benchmark by 2% a year. The investment committee approves it as a performance investment, expecting returns to rise.

Six months later, the analysts love the tool. Research coverage has doubled and reports arrive in hours instead of days. But portfolio returns are unchanged, because the tool never changed a position: it changed how quickly analysts reached the same conclusions. The purchase was a success as an efficiency tool and a failure as a returns investment, and the firm only discovers this because it measured the wrong thing at the start.

Now replay the same purchase with the right metrics. The firm classifies the tool as an efficiency buy, sets a target of 15 analyst hours saved per week, and prices that time against analyst salaries. It separately asks the vendor for the returns claim in writing, with benchmark, period, and attribution, and sets a 12 month review date at which the returns claim must be evidenced or dropped. The tool is still bought, still loved, and still valuable, but the budget is justified honestly, and the vendor is held to a testable promise.

The Buyer Checklist: Five Questions Before You Sign

  1. Which gain are you selling me? Ask the vendor to name one: efficiency, insight, or returns. Refuse a pitch that claims all three without separate evidence.
  2. What is the metric, and what is the baseline? Efficiency needs hours saved against a measured current state. Insight needs decisions changed. Returns need a benchmark, a period, and a net-of-cost figure. No baseline means no proof is possible later.
  3. Where in the workflow does the AI sit? If it sits in research and data processing, like most tools today, judge it as a research and data tool. If it claims to sit in portfolio construction or execution, ask for the decision log.
  4. Who else has measured this, independently? Ask for named reference clients willing to discuss measured outcomes, not just satisfaction. A vendor claiming returns should be able to show its working.
  5. What is the review date, and what happens if the outcome is missed? Write the promised outcome, the measurement method, and the review date into the agreement, with a defined consequence, such as a fee adjustment or an exit option. A promise without a review date is a slogan.

What the 8% Figure Does Not Mean

Three cautions keep this reality check honest. First, 8% is a snapshot from early 2026, not a permanent verdict. AI capability is improving fast, 91% of surveyed managers plan to expand their AI use within twelve months, and today's research assistant may become tomorrow's decision support. The right response is skepticism about today's claims, not dismissal of the technology.

Second, the survey measures reported, measurable improvements, which sets a high bar. Some firms may be gaining returns they cannot yet attribute cleanly to AI, because attribution in investing is genuinely hard. That cuts both ways, though: if attribution is hard for the buyer, it is equally hard for the vendor, and unverified attribution should not be sold as fact.

Third, the biggest barrier named by managers is not regulation or talent, but data quality and access, cited by 69% of respondents. This matters because it reframes where the money should go first. An organization whose data is fragmented or poorly governed will get limited value from even the best AI model. Investment in data engineering and analytics foundations, clean pipelines, consistent definitions, and governed access, is often the prerequisite that makes the AI purchase work at all.

How to Buy AI Honestly: A Practical Framework

Start with the business problem, not the technology. Mercer frames due diligence as four questions: what specific problem does AI solve, where in the workflow is it applied, how is it operated and governed, and what counts as outperformance. Borrow that framework for any AI purchase.

Next, classify the purchase honestly. If the value is efficiency, build the case on time saved and track it monthly. If the value is insight, build the case on decision quality, and keep a log of decisions the tool changed. If the value is returns or revenue, insist on a controlled comparison and a review date, and be prepared to walk away if the evidence does not arrive. This kind of structured evaluation is exactly what professional IT consulting is for: an independent view of what a tool can actually do, where it fits, and how its promised outcome will be proven. A machine learning solution that accelerates analysis is worth paying for, but its business case should be written in hours saved or decisions improved, not in revenue it has never demonstrated.

Finally, treat the first year as a pilot with success criteria, even for a full purchase. Define the metric, the baseline, and the review date in writing, and report the result against the original promise, not against a new story invented after the fact. Firms that do this will buy AI successfully, because they will buy the right tool for the right reason, and they will know whether it worked.

For more on how analytics tools translate into better decisions, see how AI-powered analytics help businesses make smarter decisions, and for allocating technology budgets to the highest-value opportunities, read the high-ROI decision guide on where to invest your budget for maximum return.

Frequently Asked Questions

Does the 8% figure mean AI is useless for investors?

No. AI's proven value today is mainly efficiency and insight, not returns: 69% report better efficiency and 55% report better insights. Those are real gains, but they should be bought and measured as efficiency and insight, not as performance.

Why do so few firms report better returns from AI?

Because AI is mostly deployed in research (49%) and data processing (40%), not in portfolio construction (13%) or trade execution (7%). A tool that never changes a portfolio decision cannot directly produce a return, and whether greater research breadth translates into persistent outperformance remains unproven.

How should I evaluate a vendor that promises AI-driven outperformance?

Ask for the benchmark, the measurement period, the attribution method, and the net figure after the tool's cost. Ask for named reference clients who measured the same outcome, and set a written review date. A backtest alone is not evidence, and a satisfied customer is not the same as a measured return.

Is efficiency alone a good enough reason to buy AI?

Often, yes, provided the hours saved are real, measured, and redeployed to valuable work. The mistake is justifying an efficiency tool with a returns story, which sets it up to be judged a failure even when it succeeds.

What should come first, the AI tool or better data?

Better data, in most cases. 69% of surveyed managers named data quality or access as a significant barrier to AI adoption, the largest obstacle in the survey. Clean, governed, accessible data is the foundation that determines whether any AI investment can deliver on its promise.

Will the 8% figure improve over time?

Probably, but buyers should not pay today for tomorrow's improvement. 91% of managers plan to expand AI use within a year, and capabilities are advancing quickly. The smart approach is to buy for today's proven value, pilot outcome claims with strict success criteria, and expand when the evidence arrives.

The 8% reality check is good news for disciplined buyers: it shows where AI delivers today, where it does not yet, and what to demand before you sign. Buy the efficiency, verify the insight, and make any vendor who promises returns prove it, on an agreed metric, by an agreed date. If your organization is evaluating an AI investment and wants an independent assessment of the claims and the measurement plan, talk to our team and we will help you buy AI on evidence, not on enthusiasm.

#AI investment returns#AI in asset management#Mercer survey#AI buyer guide#IT consulting