This guide looks at agentic AI in investment and portfolio capabilities. For the bank-wide picture, our overview of agentic AI use cases in banking describes nine illustrative banking workflows where agentic AI could apply.
This guide focuses on answering the question: Wich portfolio workflows deserve a close look before a pilot, and how do you assess them?
A CDO will likely open this guide first. The Head of Wealth Management or portfolio leader who owns the workflow is the natural co-reader. Every workflow below is a candidate for assessment. Every measure is an example for your firm to define.
What wealth management means in this guide
Wealth Management is a capability domain inside the platform: portfolio intelligence and reporting, investment advice and execution, mandates, ESG. It serves Private Banking and reaches well beyond it.
Private Banking is the segment: UHNW/HNW clients, relationship-led and RM-fronted, low volume, high complexity.
This guide evaluates Wealth Management capabilities that can matter to private banks and to other wealth firms. Its scope follows the capabilities and extends past the Private Banking segment.
For client segmentation questions, start with our article on sorting wealth clients by AUM.
Four agentic AI workflows in wealth management to assess
1. Portfolio intelligence and reporting
Portfolio reporting follows a rhythm. Teams gather holdings and performance data, then explain what drove results in client commentary. Parts of that preparation repeat each cycle, which makes the workflow worth assessing. The advisor's judgment matters most in the commentary itself. In this candidate, an agent prepares a draft and a short list of points for the advisor.
- Trigger: A scheduled reporting cycle or a client request for a portfolio update.
- Required information: Holdings, performance data, and approved reporting templates.
- Candidate agent task: Prepare a draft summary of performance drivers, plus items that need advisor attention.
- Human decision boundary: The advisor or portfolio manager reviews the commentary and decides what reaches the client.
- Exceptions: Missing or conflicting data, or source-system breaks, go to a named owner.
- Firm-defined pilot measure: Reporting preparation time and the proportion of drafts edited, compared with the firm's baseline.
2. Investment advice preparation and review
A good client review depends on preparation. The advisor recalls the client's goals, risk profile, and current holdings. The firm's house view adds another reference point. Checking a proposal against recorded objectives has clear reference points, which makes it a candidate for assessment. The recommendation and the conversation stay with the advisor throughout.
- Trigger: A scheduled client review or a change in the client's circumstances.
- Required information: Client goals, risk profile, current holdings, and the firm's house view.
- Candidate agent task: Prepare a meeting brief and a draft proposal rationale, checked against the client's recorded objectives.
- Human decision boundary: The advisor owns the recommendation and the conversation. A reviewer approves when firm policy requires it.
- Exceptions: A stale profile, conflicting household objectives, or a proposal outside the advisor's remit.
- Firm-defined pilot measure: Advisor preparation time and review comments per proposal, compared with the firm's baseline.
3. Mandate and ESG constraint checks
Constraints live in several places. Mandate terms set one boundary, the firm defines its ESG criteria, and rating sources supply the data. Checking a trade or proposal against all of them takes care. In this candidate, a pre-check surfaces matches, breaches, and items that need judgment before a reviewer looks. Interpreting those items stays with people.
- Trigger: A proposed trade, a new proposal, or a periodic mandate review.
- Required information: Mandate terms, firm-defined ESG criteria, current holdings, and ESG-rating sources.
- Candidate agent task: Pre-check constraints and flag matches, breaches, and items requiring judgment.
- Human decision boundary: Compliance or the portfolio manager interprets and rules on breaches. The agent does not waive constraints.
- Exceptions: Ambiguous mandate wording, inconsistent ESG sources, or constraints recorded only in documents.
- Firm-defined pilot measure: Breaches caught before approval and false flags, compared with the current process.
4. Investment execution preparation and approval
This workflow begins after the decision is made. The recommendation or rebalancing choice is already approved, so the candidate work is preparation. A ticket has to reflect account permissions, positions, order rules, and trading cut-offs. In this candidate, an agent assembles the draft and the checks, and gives the approver a clear summary. Any order is placed only after an authorized person approves it.
- Trigger: An approved recommendation or rebalancing decision.
- Required information: The approved decision, account permissions, positions, order rules, and trading cut-offs.
- Candidate agent task: Prepare a draft order ticket, pre-checks, and a summary for the approver.
- Human decision boundary: An authorized person approves before any order is placed, within existing firm controls.
- Exceptions: Insufficient cash, restricted instruments, or orders close to a deadline.
- Firm-defined pilot measure: Ticket rework and the time from approval to ready-to-place, compared with the firm's baseline.
To decide which wealth-management workflows to automate first, use the agentic banking use cases prioritization guide.
How to evaluate a portfolio workflow before a pilot
Once you have a shortlist, put each candidate through six checks. Together they show whether a workflow is ready for a pilot.
Data quality. An agent prepares work from the information it receives. Confirm that holdings and client profiles are complete, current and traceable to a source. Apply the same test to ESG data, and flag any fact that lives in two systems with two values.
Permissions. Define the data the agent may read and the actions it may take. Its data access should follow the entitlements your advisors already hold. Decide which steps stay read-only.
Documented rules. An agent can follow rules that exist on paper. If a mandate interpretation lives in one portfolio manager's head, write it down first.
Human approval. Name the approver for each workflow. Mark the point where the agent's output stops and a person's decision starts. Record who approved what.
Exception handling. List the cases the agent should leave alone. Give each one a destination and an owner, and set a time limit.
Evaluation. Decide before the pilot what good looks like. Review samples of agent output against expert judgment. Keep reviewing after launch.
Baseline first, then set your own measures
Start by measuring the current process. Record how long the workflow takes and how often work is redone. Note where errors appear. Each firm then sets its own measures and targets. A pilot result needs a baseline beside it to carry meaning.
The CFA Institute offers a useful lens for these checks. Its framework for ethical decision-making in AI for investment management covers obtaining input data. It also covers building, training and evaluating the model, and deploying the model and monitoring it. Data quality sits naturally with the first stage, and evaluation with the second. Approval, permissions and exception handling come into focus once a workflow is deployed and monitored.
Keep scoring brief
At this stage, a simple pass, hold or fail on each check is enough. For the full scoring approach, use the guide to agentic banking use cases. It keeps attractiveness and preparedness separate and adds readiness gates.
How this differs from the operating-model question
This guide assesses individual portfolio workflows. Our article on what a wealth management operating model is and why it breaks covers the broader question of how a wealth business is organised to run its work.
Where Backbase fits
Investment advice preparation and review is the workflow closest to Backbase. It connects to Relationship Intelligence, which supports advisors with Client 360, next-best-action, meeting prep, proposal co-creation and house-view summarisation.
The other three workflows are candidates for your own assessment. Portfolio intelligence and reporting is one. Mandate and ESG checks is another. Execution preparation and approval completes the set. All four are illustrative examples of how you might evaluate work in your firm. They say nothing about Backbase deployments.
You can read more on the Relationship Intelligence page. To rank candidates like these, see the guide to prioritizing agentic AI workflows.
