Has AI in asset management finally gone mainstream? Yes. Adoption is no longer what separates firms; what they do with it is.
Northern Trust’s Driving Growth in Asset Management 2026 survey, run with InvestOps Insights, asked 300 senior executives at managers with up to $500bn in AUM how their operating models are changing. One finding stands out: every single respondent reports deploying AI in some form.
The same survey shows something just as important. Managers are pulling back from large internal transformations and handing more of their operations to specialist partners.
Most commentary treats these as two separate trends. We see them as one story. Outsourcing and AI are both answers to the same question every asset manager is now asking: what do we truly need to own?
In this article we look at what the data says about AI, where we believe the real value lies, and what CEOs, CIOs, CTOs and investment leaders should do next.
What is changing in how asset managers use AI?
AI adoption is universal, and mostly partner-led
The survey splits AI deployment into three models:
- 41% combine internal tools with external partnerships
- 33% work through partnerships with specialised providers
- 26% rely only on general-purpose tools such as Microsoft Copilot or ChatGPT
Put simply, roughly three in four asset managers already rely, at least in part, on external partners for AI. Only a quarter are going it alone with off-the-shelf assistants.
AI is wanted where accuracy and process matter most
Asked where AI could help most, managers point first to the operational chain that surrounds investment decisions:
- Data accuracy and quality control: 60%
- Document management and alternative investment processing: 55%
- Research (summarising reports, parsing materials): 50%
- Operational due diligence: 48%
- Client reporting and communications: 40%
- Investment due diligence (screening, analysis): 25%
The ranking is telling. The priority is not replacing investment judgement. It is making the information behind that judgement cleaner, faster and more reliable.
The front office wants AI too
The front-office numbers point the same way. Sourcing and aggregating varied investment analytics is the biggest front-office challenge, cited by 59%. Attribution at account, desk and strategy level follows at 41%, and cash forecasting and liquidity at 37% each.
To respond, 63% plan to implement AI, generative AI or machine learning. 51% will invest in data-specific decision-support tools.
The operating model is being redrawn at the same time
AI is arriving while managers rethink what they do in-house:
- Refining or rebuilding the target operating model fell from 75% to 46% since 2024
- Outsourcing non-core activities as a strategic lever rose from 18% to 39%
- Outsourcing as a cost-control lever rose from 27% to 42%
- Automation and technology, including AI, is a cost lever for 52%
The link is explicit: 40% say the availability and adoption of internal AI capabilities has increased their likelihood of outsourcing. Cybersecurity (44%) and staffing challenges (43%) push in the same direction.
Data remains the bottleneck
The biggest data challenge is consolidating information from disparate internal and external sources, cited by 49%. Managing data and cybersecurity costs follows at 43%.
The report draws its own conclusion: AI is moving from experimentation to implementation, but its value will be constrained by the quality of the data, controls and governance beneath it.
Our view: outsourcing and AI are the same decision
The question is no longer build vs buy, but own vs orchestrate
We believe the most important signal in the survey is not the 100% adoption figure. It is the combination of two trends: fewer firms rebuilding their operating models from within, and more firms turning to partners, for operations and for AI alike.
The build-versus-buy debate assumed a firm would end up owning everything, either by building it or by buying it. That is no longer the goal. The leading asset managers are deciding what they must own, and orchestrating everything else.
In our view, what an asset manager must own is clear:
- Investment judgement and the investment process
- Client relationships and the investor experience
- Accountability: oversight, risk appetite and regulatory responsibility
Almost everything around that core — data preparation, document processing, reconciliation, reporting, monitoring — can be run by specialist partners and AI agents for asset management, provided the firm keeps control of how it is done.
General-purpose tools are where AI starts, not where it scales
The 26% relying only on general-purpose assistants is, in our experience, the group most at risk of stalling.
Copilots are useful. They save time on emails, summaries and first drafts. But they rarely reach production in regulated, high-volume workflows such as operational due diligence, capital call processing or reconciliation breaks. Those workflows need agents that:
- work on the firm’s own data, systems and procedures
- follow defined steps, with clear limits on what they can and cannot do
- leave a complete audit trail of inputs, reasoning and outputs
- hand exceptions to a human at the right point
Agentic AI for asset managers is not one super-agent. It is a set of narrow, controlled agents, each doing one job well inside a governed workflow. We believe this is why three in four managers already work with partners: production-grade agents require engineering, domain knowledge and governance that a general-purpose licence does not provide.
You can outsource the work, not the accountability
This is the point most often missed in outsourcing debates, and it applies to AI with even more force.
Whether a process is run by a custodian, a technology partner or an AI agent, the regulator still holds the asset manager responsible. Supervisors across Europe — the Central Bank of Ireland, the CSSF in Luxembourg, FINMA in Switzerland and the FCA in the UK — expect firms to oversee what they delegate. The EU AI Act and DORA’s requirements on ICT third-party risk raise that bar further.
That is why we see AI governance for asset managers in Europe as an enabler, not a brake. Firms that can show who did what, on which data, with which controls, will scale AI faster, because they can answer the questions that stop everyone else. Many firms build agents. Far fewer know how to govern them.
Value compounds across connected workflows
Look closely at the survey. The functions managers want to outsource — data management, trade processing, reconciliation, middle office — and the areas where they want AI — data quality, documents, research, due diligence, reporting — are not separate lists. They are one chain.
Take alternative investments. A capital call notice arrives as a PDF. It is read, validated against fund documents, reconciled with cash positions, checked for compliance, reflected in reporting and communicated to investors. Automating one step gives a modest gain. Connecting agents across the whole chain compresses the entire cycle, and removes the hand-offs where errors usually appear.
Buying point solutions one by one recreates the vendor sprawl managers are trying to escape. The survey shows managers want fewer, deeper partners. We believe the same logic should apply to AI: shared data foundations, reusable components and one governance framework, rather than a collection of disconnected pilots.
Better outcomes, not just lower costs
Outsourcing is often framed as a cost exercise. The survey suggests managers want more than that: quality and accuracy is their top operational priority (53%), and data accuracy is the area where they expect AI to help most (60%).
In our experience the biggest gains are rarely headcount. They are fewer errors, faster cycle times, more consistent controls and teams that spend their time on judgement rather than reconciliation.
That matters given that 43% of managers cite staffing challenges. AI should not replace scarce expertise. It should amplify it, turning analysts and operations teams into supervisors of well-governed processes.
What should asset management leaders do now?
1. Redraw the core
Start with an explicit map of what the firm owns and what it orchestrates. Investment judgement, client relationships and accountability stay inside. For everything else, ask the same question for outsourcing and for AI: who can run this better, with what controls, and how will we oversee it?
2. Fix the data layer before scaling agents
With 49% of managers struggling to consolidate data from disparate sources, this is the real starting line. Agents amplify whatever data they are given. Prioritise:
- a single, reconciled view of positions, cash and reference data
- clear data ownership and quality checks
- access controls that agents can respect by design
3. Govern outsourced and AI-driven processes with one framework
Do not build one oversight model for vendors and another for AI. Use a single framework covering:
- audit trails for every decision and output
- explainability proportionate to the risk of the task
- human-in-the-loop checkpoints at defined thresholds
- ongoing monitoring of performance, drift and incidents
4. Start with connected workflows, not isolated tasks
Pick a chain where hand-offs are frequent and errors are costly: alternative investment documents, operational due diligence, reconciliation exceptions, or periodic client reporting. Map every step, then decide where agents act, where humans review, and how the steps connect.
5. Choose fewer, deeper partners, for operations and for AI
The survey shows managers consolidating providers to reduce complexity. Apply the same discipline to AI. Favour partners that can co-design, deploy into production and govern over time, rather than vendors selling another standalone tool.
6. Measure what matters
Leave demo metrics behind. Track the operational ones:
- cycle time per workflow
- error and exception rates
- time from idea to production
- share of components reused across use cases
- incidents and their resolution time
These are the numbers a board and a regulator will both understand.
Conclusion: from AI adoption to AI operating model
The Northern Trust survey describes an industry entering a more disciplined phase. Managers are concentrating on their core, outsourcing more of what sits around it, and deploying AI everywhere. We believe these are not three trends but one shift in the operating model.
AI in asset management has crossed the adoption line. The next phase will be decided by three things:
- Data foundations that agents can trust
- Governance that keeps accountability where regulators expect it
- Partnerships that deliver production-grade agents across connected workflows, not more pilots
The firms that get this right will not simply cut costs. They will run faster, more accurate and more controllable businesses, with their best people focused on what clients pay them for.
The next competitive edge in asset management won’t come from doing everything in-house. It will come from knowing exactly what to own, and governing everything else.


