General AI answers ERISA questions confidently. That is exactly the problem.
General-purpose AI knows a lot about benefits law in general, but it does not always know how to think about ERISA, what law to focus on, or what your plan documents say. This makes it unreliable when you ask it a specific question where it needs to be able to think like an ERISA professional — by considering all the nuances of ERISA, and applying them to what your plan document says. We build the AI SKILLs that close these gaps, and we teach your benefits people to use them effectively. Start with the SKILLs Suite: twenty-six substantive ERISA practice areas, installed in the Claude account you already have — and your AI outputs will consider more of the relevant legal issues, will be more focused and will generate draft documents and analysis that is much closer to actionable than a generic AI tool can manage.
See the SKILLs Suite → Training →
"An employee sent an email to HR six months ago, asking them to change the employee's deferral rate from 3% to 5%. The employee did not change their deferral rate in the Plan's online portal and did not complete a deferral change form. What do we need to do now?"
Four ways general AI gets ERISA wrong.
Ask a leading model a textbook question and it will often answer well. Ask it the question your team actually has — the one turning on your plan's language, your service-crediting method, your committee's past practice — and the confident fluency becomes a liability. It will still answer confidently. It will just be answering a different plan's question, and it might not consider all of the relevant issues.
It does not know how to think about ERISA's myriad interlocking provisions
General AI answers from broad ERISA knowledge, but it might not consider all of the ERISA-related areas — like the interaction between a deferral or match correction and a testing failure. The risk is not that general AI gets something completely wrong. It is that it is confident about an issue it should not have considered, and does not consider an issue that is essential to the question asked.
What closes it: guidance for how to think about ERISA questions; what else to consider when one ERISA issue is raised, and step-by-step instructions for how to work through each ERISA question raised — the way ERISA professionals approach this work.
It doesn't know your plan
General AI knows a lot about 401(k) plans. But it has not read your adoption agreement. When eligibility, vesting, or a distribution provision turns on a design choice the statute permits several ways, it fills the gap with the most common answer rather than the correct one.
What closes it: your own documents. The getting-started SKILL walks you through setting up a workspace holding your plan documents, SPDs, and forms — and confirming they are actually being read by the AI.
It's stale in a field that moves
ERISA guidance does not stand still. Between SECURE 2.0 provisions phasing in, shifting agency positions, and litigation reshaping fiduciary practice, an answer that was right eighteen months ago may be wrong today. A general model's knowledge has a cutoff and does not necessarily keep up if the law changed.
What closes it: dated releases, currency checks built into the SKILLs, and honest disclosure. Our SKILLs specifically tell the AI to check for updated figures regarding items that change frequently — like contribution limits, ACA affordability numbers, and excise tax amounts. Each release also states the law as of a given date, and says so when currency matters. And we intend to publish updated releases on a regular schedule, though nothing in the license commits us to one — the release you buy is yours whether or not another appears.
It doesn't know when to stop
A generic model cannot spot the line between an administrative question it can handle and a judgment call that needs a lawyer.
What closes it: criteria for recognizing a judgment call, plus a failsafe notice that fires when one appears — telling you plainly not to act on an answer until ERISA counsel has reviewed it.
Twenty-six practice areas, plus setup and routing.
Organized and activated by the question you ask. You might not use all of them — a sponsor with a 401(k) and a health plan may only ever use six or seven — but they will be there when you need them, even if you don't know you need them.
Qualified Retirement Plan Compliance
Health & Welfare Benefits Compliance
Fiduciary Governance & Plan Administration
Corrections — When Something Goes Wrong
Executive & Equity Compensation
Corporate Transactions & Workforce Changes
Specialized Employers & Plan Structures
Life Events & Employee Leave
“I’m one person — I can’t set this up.”
One person absolutely can set this up and supercharge your benefits department. This is why we built one of the twenty-eight SKILLs to show you how to use the other twenty-seven. Installation takes about fifteen minutes with no developer tools and no special expertise required. Then you ask the SKILL to teach you, and it takes it from there.
What it walks you through
- Creating a workspace and writing the instructions that tell it how to treat your plans.
- Which documents to load first — plan document, SPD and SMMs, trust agreement, administrative forms, the last two Form 5500 filings.
- Confirming your documents are actually being read, with a test question and how to read the answer.
- Six prompting habits, each with a worked before-and-after example, so that you can get the most out of the SKILLs and the general AI.
What stays with you
- Assembling your own documents and deciding what goes in.
- Redacting participant data — the SKILL tells you how to test coverage without names.
- Clearing the whole thing against your employer's own data-handling and vendor policies.
- Obtaining review by ERISA counsel of your own choosing before acting on any output.
The tool is the easy part.
Nearly everyone who tries AI on benefits work has the same two experiences in the first week: an answer that saves an hour, and an answer that would have been embarrassing if they had sent it. The difference is almost never the model — it is how the question was asked, what context it was given, and whether anyone checked the output.
That skill is teachable and it transfers, which is why we teach it as its own thing rather than as a product demonstration. Sessions are online and are being scheduled now.
Five things we believe about AI in a benefits function.
Narrow beats general
A tool that knows one field deeply is worth more than one that knows every field shallowly. Everything we build is for employee benefits work and nothing else, which is why it can afford to be opinionated about how a question should be analyzed.
A tool has to know its own limits
The most dangerous output is a confident answer to a question that required judgment, or to a question that was not actually asked. Our SKILLs are built to address both limits.
You supply the facts
An answer is only as specific as the documents behind it. We do not guess at your plan design — we give you the means to put your own documents in front of the tool, we teach you to verify that it is reading them, and we teach you how to prompt the AI effectively to ensure it has the relevant facts before jumping to an answer.
Currency is part of the product
In a field that moves quickly, a tool that cannot tell you how old it is will eventually mislead you. Our SKILLs are designed to ensure the AI looks for the most current information. And each output includes the date the SKILL's content is current to, and says so unprompted when the question is time-sensitive.
Nobody is reviewing the output but you, or your chosen advisor
Our SKILLs will not imply otherwise. No attorney checks the AI's answers before you see them, and no law firm stands behind a given response. Every output is a draft for review by ERISA counsel of your own choosing.
Built by people who practice ERISA law.
Our SKILLs are not a general AI consultancy's reading of the benefits field. The content was written by practicing ERISA attorneys for use in their own work, and it exists because they needed it. The reference summaries are original syntheses of primary authority they prepare for their own use, rather than repackaged treatise material. And the worked examples built into the SKILLs — coverage testing, ADP correction, cross-testing rate groups, top-heavy, 415 aggregation — are the calculations a practitioner actually runs.
What that means, and what it does not: it describes who wrote the content. It does not make the product legal advice, and it does not make anyone your lawyer.
ERISA Benefits Tech, LLC is a technology company, owned by the professional entities of the two partners of an Arizona law firm that practices ERISA and employee benefits law. It is not itself a law firm: it does not practice law, and it does not represent you. Buying or using our products creates no attorney-client relationship. Output is unreviewed, is not legal advice, and must be reviewed by competent ERISA counsel of your own choosing before you act on it.
Everyone starts with the Suite. Some go deeper.
Where the Suite ends, a custom build begins: SKILLs grounded in your plan documents by us rather than by you, your house positions and communication preferences encoded into how each SKILL reasons, and — for employers — escalation criteria configured to route judgment calls to counsel, with ERISA legal review provided by ERISA Benefits Law Group, PLLC under a separate engagement.
That path starts with a fixed-fee readiness assessment and a written roadmap you keep whatever you decide. But it starts after the Suite, not instead of it — a few months of real use tells you more about where AI helps your function than any proposal can. Start a conversation when you are ready.
The Suite is not on sale yet.
We are finishing the license terms and the purchase path. Join the waitlist and we will tell you when it opens — and when the first training sessions are scheduled.