Use of Technology How to build a repeatable process for using AI

“It’s important to think about how artificial intelligence (AI) could help you as an advisor, and then consider how to build a process that allows you to use it again and again,” says Brock Sutton, head of emerging client capabilities at Capital Group. He sees a repeatable process as one of the most important foundations for using AI effectively. The goal is not just to experiment with a tool once, but to identify a workflow, understand how it works today, apply AI in the right places and then measure whether it improves the business.

A simple framework includes five steps:

1. Prioritize the right workflows

Start by writing down the main activities required to run the business (e.g., portfolio construction, client onboarding, client review meetings, communication, marketing, etc.). This should include the work done by the advisor and the broader team. Then score each activity on two dimensions:

• Total time spent
• Business value

Use a simple 1-10 scale for each category. This does not need to be overly scientific; the scoring is relative. The goal is to identify where the team spends meaningful time and where the activity creates meaningful value.

The best place to start is usually the workflow with the highest combined score — the one closest to 20. That usually represents a process where the firm is spending a lot of time and where improvement could create real business impact.

“The two biggest use cases we see are admin tasks, such as automating client meeting prep or client onboarding, and investment research and analysis, such as applying it to asset manager due diligence or fund selection process,” says Sutton.

2. Document the current process

Once a workflow has been selected, document how it works today. For example, if the priority workflow is a client review meeting, write down each step involved in preparing for and completing that review. For each step, capture:

• What the step is
• What the desired outcome is
• What “good” looks like
• Who owns the step
• How long it typically takes
• What data is used
• Where that data lives

This creates a clear map of the existing workflow before AI is introduced.

3. Map AI tools to the process

Next, look at the AI tools the firm already has access to and determine where each tool fits within the workflow.

The key is to match the right tool to the right job. Some tools may be better for drafting communications. Others may be better for summarizing documents, analyzing data, preparing meeting materials or organizing notes.

This step prevents firms from using AI randomly. Instead, AI is applied intentionally to specific parts of a defined process.

4. Implement and test

After mapping the tools to the workflow, begin building the actual AI-enabled process. This includes:

• Creating the prompts needed for each step
• Building templates or repeatable instructions
• Testing the outputs
• Creating evaluation criteria
• Using scoring rubrics to judge quality

Evaluation is especially important. Teams should define what a good output looks like and compare AI-generated work against that standard. The goal is to improve the process before using it more broadly.

5. Scale and measure

Once the workflow has been tested and refined, the firm can begin using it more consistently in production. At that point, measurement becomes critical. The firm should consider tracking questions such as:

• How much time is being saved?
• Which parts of the workflow improved the most?
• What additional work can the team now do?
• Where is the greatest business value being created?

Those insights should feed back into the prioritization process. For example, if the firm finds that AI saves the most time in data analysis, the next workflow to prioritize may be another process that relies heavily on data analysis.

What's the bottom line?

The most effective AI implementation starts with the business process, not the tool. Identify high-value workflows, document how they work today, apply AI intentionally, test the output and then scale what works.