The goal: a first AI tool that compliance could approve
When someone leaves a job, they have to decide what happens to their 401(k): roll it into an IRA, leave it in the old plan, move it to a new employer’s plan, or take the cash. It is one of the largest financial decisions many people make, and it often comes with questions they would rather ask on their own time than wait to ask a person.
American Century Investments saw a place where AI could help, and wanted it to be the firm’s first AI tool. The hard part was not the technology. It was making sure an AI assistant would protect participants’ data, stay within the firm’s compliance standards, and never cross from education into personalized advice.
What we built together
American Century and Athena built a rollover decision guide, live at retirement.americancentury.com, under American Century’s brand.
A short guided path
A few plain-language questions about the person’s situation, such as their age when they left and whether a new employer offers a plan.
The options, explained
The guide lays out each choice — rolling over to an IRA, staying in the former plan, moving to a new employer’s plan, or taking a cash distribution — with what each one means for their situation.
An AI assistant alongside
A Rollover Assistant sits beside the guide and answers questions in plain language as the person works through it.
Clear disclosures
Results are presented as education, with the disclosures American Century requires, not as a personalized recommendation.
Built for compliance from the first line
Compliance was part of the design from the start, not a review at the end. Every concern American Century raised shaped how the assistant was built.
Data isolated in Athena’s AWS
Participant data stays inside Athena’s secure infrastructure on Amazon Web Services, not in a third-party AI service.
AI on Amazon Bedrock
The assistant runs on Amazon Bedrock inside that same infrastructure, so conversations never leave it.
No training on participant data
Conversations are never used to train AI models. What a participant types is used only to answer them.
Guardrails that limit advice
Guardrails keep the assistant on the rollover decision and on education. It declines requests for personalized advice and off-topic questions.
Every conversation is recorded, including the questions the assistant declined, so American Century can see exactly what participants asked and how the assistant answered.
One question, start to finish
Here is how a typical question moves through the assistant. The question is an example.
A saver asks
“What’s the difference between an IRA and leaving it in my old plan?”
The assistant explains
Costs, choices, and access, in plain language
Guardrails hold
Education only; personal recommendations are declined
The conversation is kept
Recorded for American Century to review
If the saver then asks “So which one should I pick?”, the assistant doesn’t choose for them. It explains what to weigh and points them to the guide’s next step, so the decision stays theirs and the answer stays within American Century’s rules.
What it means for your firm
Advisory firms get the same care. Athena AI answers your clients from their own household data and your firm’s investment policy, under your brand, and every conversation lives on the household in Mission Control for your team and your compliance staff to read. When a firm needs its AI isolated the way American Century did, Athena can run it on Amazon Bedrock with guardrails inside its own AWS infrastructure.
See how the answers stay grounded on Athena AI, read how the platform is secured on Security, or book a demo. More stories are in our case studies.