CPP Investments on how AI redefines core investing roles and processes

CPP Investments says its internal trials show the power of generative AI has advanced so quickly that traditional workflows across core investment tasks could eventually be eradicated – but only with strict governance and clear operating frameworks. 

The C$777.5 billion ($562 billion) Canadian asset owner ran three head-to-head experiments ranking autonomous agents – which work together like digital workers on complex tasks – against traditional processes:  

• Accurately reconciling a $522 million portfolio which contained hidden errors. 
• Conducting performance attribution. 
• Analysing a forward-looking tariff scenario. 

The strong results suggest that AI can now do more than speed up workflows and processes; it will eventually redefine roles, jobs, and organisations themselves, according to Jon Webster, senior managing director and chief operating officer at CPP Investments. 

“We have 2,100 colleagues who are engaging with the technology practically every day,” Webster tells Top1000funds.com. “As we see the opportunities moving forward, we are well positioned to listen to the team and then say, ‘this is now reaching the point where we could think about applying it in different ways than we have seen before’.” 

A five-gate framework

Webster says the industry is trying to build reliable systems with non-deterministic technology (potentially delivering different outputs from the same inputs), making a nuanced, tailored approach is essential. He advises using a five-gate system which can impose clear discipline and lift results: 

• Frame: specify the problem and context. 
• Input: clean and validate the numbers. 
• Model: develop and/or run an analysis and flag uncertainties. 
• Validate: cross-check results. 
• Narrate: it must document every decision. 

“If you are to naively copy off-the-shelf patterns that others are showing you, without engaging very diligently and thoughtfully in your context, with your requirements, with your reliability and consistency needs, that is likely to be the place where mistakes will be made and the results will be disappointing,” he says. 

Process and quality control are crucial 

This governance system and quality control – including verifying sources, normalising identifiers, and tracing every figure – proved crucial in lifting the AI’s performance in the position reconciliation test conducted by CPP. 

It was clear that AI only performed well when the right human inputs, guardrails and oversight were in place. 

While direct prompting caused AI models to disagree and miss critical errors, requiring the AI to rate its confidence in each finding (triggering human review when below a 70 per cent confidence level) meant accuracy was boosted by 45 per cent. Two AI models also cross-checked each other, and humans reviewed uncertain cases.  

The second test – performance and risk attribution – also showed strong results with the right framework. The AI agents performed well given a clear problem, clean data, and explicit methodology. Without such precision, the AI could choose the wrong analytical approach, like using a monthly calculation rather than a daily one. Humans framed the problem and verified the results. 

The third test required AI to model how semiconductor export restrictions would impact a globally diversified portfolio. Without clear human guardrails the AI predicted gains when there should be losses. A more careful combination of AI-human expertise produced a credible prediction of a 1.82 per cent portfolio decline. 

Webster said this tariff scenario was a good example of how AI could help organisations become better investors.  

“I don’t think we will become a better investor by simply doing our existing tariff scenarios better,” Webster says. “But maybe we’re able to run more scenarios, maybe we’re able to look at more edge cases, maybe we’re able to put more inputs into those to think about areas that we couldn’t have thought about before.” 

A redefined future for humans 

CPP Investments is not yet at the point where it will overhaul the structure of the organisation, Webster says, but the value of people in the investment chain will inevitably shift.  

“I think it’s premature to say where value will migrate to. But intuitively, we expect value to migrate into organisational EQ [emotional intelligence]: trusted relationships with others, your standing in the environment,” he says. 

Today, analysts do the research, portfolio managers decide, and risk teams validate. AI is compressing that sequence but may eventually reshape it completely. 

Webster says the future may involve smaller teams organised around outcomes rather than departments with people acting as facilitators (who manage AI workflows), architects (who frame problems), and leaders (who make judgment calls). 

Staff at CPP Investments continue to experiment, using AI tools from several major providers, particularly given the rapid pace of change and lack of existing roadmaps about how to use the technology. 

“Which ones – if indeed we anchor on one in the long term – I think is unclear. They may evolve into different product market fits, and different capabilities may suit different parts of our teams, our investment approach.”