Are today's Al-powered investment tools just a new flavor of hype, or do they have the ability to deliver real, measurable alpha?

Al is absolutely surrounded by hype. At the same time, the capability shift is real, and it matters in a very practical way for fundamental investing.

Alpha often comes from finding information that's not yet fully priced in. The challenge has never been that investors lack ideas. The challenge is the work required to prove or disprove them. That work lives in transcripts, filings, footnotes, internal notes, and years of history. When Al collapses the time it takes to search, connect, and summarize that material, it removes the bottleneck in the process. Teams can test hypotheses faster, go deeper on a company, and cover more of the opportunity set without lowering the bar on fundamental work.

On measurement, we believe alpha begins with the insights of our fundamental investment teams.

We don't use Al to make investment decisions. We use it to accelerate research and improve alignment; however, accountability stays with our investment professionals. The effectiveness of Al-powered investment tools might best be measured by leading indicators that are tightly linked to investment outcomes: reductions in time spent mining lengthy documents, faster processing of information, broader company coverage for a given team capacity, and faster responses during market-moving events.

Can Al-driven research really enable investment managers to play catch-up in real time during market-moving events and make better-informed decisions?

The short answer is yes. In fast-moving markets, time is a scarce resource. Everyone is trying to answer similar questions, and the edge comes from getting to the relevant context quickly and then applying judgment.

We saw this during "Liberation Day" last April, when the U.S. tariff regime drove market dislocations. We developed a semantic search tool that ingests company transcripts and related materials and makes them searchable. When the event hit, our investment teams were able to look across names in their portfolios, pull up management discussions from prior cycles about tariff exposure, and get an initial ranked view of which holdings were likely more exposed. That's a starting point, not a conclusion, but it's a powerful starting point. It helps portfolio managers and risk partners decide where to focus their attention and how to frame current volatility within the context of our long-term investment thesis.

When volatility spikes and markets move, human activation remains central to investing. Are we at a point where we can rely on agents to flag changes, analyze developments, and present them to investment managers?

We're getting closer. However, as I learned working in performance attribution earlier in my career, while the rearview mirror is always crystal clear, the view ahead is decidedly less so— the windshield is foggy, and the wipers might not be working.

In other words, the relevant themes and market-moving events are always changing. Today's major event might be a small hedge fund in China releasing a superior large-language model (LLM), tomorrow's might be political in nature.

What I'm most excited about is the integration of qualitative and quantitative work in a controlled way. LLMs are excellent at working with text, like transcripts, filings, and internal notes. Quantitative inputs such as tinancial models, macroeconomic measures of activity, and factor exposures can also be incorporated, but that'll require the LLM to call a specific tool for computation and alignment.

We're partnering with vendors who are at the forefront of this work, while also exploring applications in related contexts. From here, we might be able to see something like early-warning workflows, but they will still require controls, validation, and human review.

There's skepticism in the market. Is it overdone? Are we framing Al accurately when we talk about its ability to influence the search for alpha?

We have the privilege of working with investment analysts, who by definition are professional skeptics. In my view, we need to be skeptical. Al can feel like a catch-all for every problem, and when something is broadly useful, it's easy to start believing it can do everything. It cannot.

Al isn't magic fairy dust. It's a set of technologies that require inputs to get outputs. To honor our fiduciary obligations and fulfill regulatory requirements, we need to substantiate our investment decisions. That requires input data normalization, pipelines, controls, and clear governance. Al doesn't make that go away. Hallucinations are a real risk, especially on open context questions where there might not be an obvious source of truth for validation.

That is why we've set a hard boundary—we will not use Al for investment decisions.

More importantly, every Al solution we have in place today was proposed by analysts and portfolio managers. As a technical team, we weren't guessing what we hoped might be useful. We were co-creating solutions with our investment colleagues. Along the way, we were all alternatively delighted and disappointed, learning together.

How do you keep Al from amplifying groupthink and herding behavior?

It starts with the boundary. Al doesn't decide. Portfolio managers do.

Groupthink is a real risk because these models are optimized to produce plausible language that can look like consensus even when it isn't grounded in anything useful. The control here is context. We require traceability back to source material and keep expert review in the loop. When our portfolio managers and analysts ask questions, the questions are often specific to a theme that they're researching. We designed the tools so that they'll always provide citations in their output. This feature helps to ensure integrity in attributing insights from source material and assists the user in evaluating whether the tool has returned a likely but unsubstantiated result.

This is where our close partnership with investment teams plays a key role-portfolio managers and analysts will ask hard questions. They'll find hallucinations, and they'll keep you honest.

At the same time, groupthink is also something you want to understand, measure, and manage. Al can help identify where consensus is building, where exposures are concentrated, and where narratives are dominating price action. Those can be valuable inputs into portfolio construction and risk management. The key is to treat them as inputs and leave the decision-making to humans in the context of a broader governance framework.

Vendors are investing tens of millions to capture this. Why build anything internally at all?

The narrative we hear a lot in the industry is that if you can get everything you need from vendors and providers, you shouldn't build internally and generally this makes sense. However, it doesn't always apply.

The reason we still take a hybrid approach comes down to workflow integration and control. Providers build capabilities inside their own platforms. Investment teams work across many platforms. If Al and data become the foundation for a more unified investment framework where you connect systems and make them interoperable for portfolio management teams, you need an integration layer you control, especially if you care deeply about proprietary data security and governance.

The second reason is process specificity. We do have a platform, and it's customizable for different investment teams. And that's difficult for a vendor to do deeply without being embedded in our process. We also onboard Al-native vendors for capabilities that would be difficult for us to build well ourselves.

If you look back to the last 18 months, what's been the biggest challenge or setback? And looking ahead, what's your biggest concern?

In a highly regulated industry such as investment management, when you're thinking about launching an Al program from scratch, you might ask: who decides what foundational model you can use? How do we verify that model providers aren't using our inputs for training in ways that violate our expectations? How do we protect our data?

We benefited from Manulife's broader commitment to digital transformation and responsible Al. Since 2017, the firm has invested more than $1 Billion in this space, and the inaugural Evident Al Index for Insurance recognized Manulife as the #1 life insurance company for Al maturity in 2025¹. That institutional foundation, especially our published Responsible Al Principles², gave us the framework to move quickly. For asset management specifically, we had strong conviction that the program needed to be led by investment teams, and that Risk and Compliance oversight must be integrated into every step of the process.

It's an interesting paradox. Many believe that governance is the opposite of innovation. For us, it was a superpower. A robust governance program provided the guardrails needed to innovate with Al for asset management in the first place.

The biggest challenge then for us was how to balance experimentation with accountability. We're stewards of client capital and the firm's resources. Traditional ROl frameworks assume you can predict outcomes. With investment Al in particular, you often can't. The models move quickly, and some work will be throwaway. In our case, senior management already brought almost a decade of experience in leading Al initiatives across Manulife. And that leadership support has made all the difference. For us, investing in asset management Al was more of a calculated risk than a leap of faith.

We managed the risk by minimizing the cost of being wrong through a structured release program. A prototype sits with a couple of analysts. If it looks promising, it becomes a pilot with deeper risk, compliance, and legal review. We measure adoption and engagement, then it graduates to production, where it's supported within the broader technology framework.

This approach helped us to balance our commitments while still leaving space for exploration.

Do you have any worries about Al?

Yes, of course. I think we all do and it would be hard not to. Investment management is an apprenticeship business. I remember many nights working through a question, gathering data, formatting, aligning, and figuring out how to visualize my work. Much of this was pure drudgery. Along the way, I learned data modeling, computation, and analysis at an almost intuitive level. As Al has the potential to automate much of this early career work, I wonder what that means for the apprenticeship model, and by extension, how we, as an industry, will be training the next generation of analysts and portfolio managers.

1 https://www.manulife.com/ca/en/about-us/news/manulife-named-number-1-life-insurance-company-for-ai-maturity-by-evident 

2 https://www.manulife.com/content/dam/manulife-com/ca/financial-documents/pas/en/MFC_AI_2024_EN.pdf

This material originally appeared within Savvy Investor’s Signal Magazine, and is repurposed with permission. The views expressed are subject to change. Manulife Investment Management is not responsible for the comments by or views of anyone not affiliated with Manulife Investment Management.