Five questions to ask before you buy an AI benefits agent
AI that runs benefits administration is arriving. These five questions test whether any of it reaches your people.
Reconciliation, provider files and renewal administration have been slow and manual for years, and that work should go. An agent can take that admin away without changing one screen an employee uses. Some suppliers call this agentic benefits. Ask these before you sign.
Written so you can put them to any supplier, including us.
The short answer
Before you buy an AI agent that runs employee benefits, ask five things: what your people will see that they don't see today, who answers when someone is stuck, whether it will tell you what got used, what it knows about each person and who else sees that, and whose saving it is.
The five questions
Each one has what a good answer sounds like, what a weak one sounds like and what to ask next. Send them before the demo and ask for the answers in writing. Then ask to see each answer on screen, with one of your own policy documents if the supplier will allow it.
1What will our people see that they don't see today?
An agent that works for the benefits team can take the administration away without changing anything for the person with the cover. Staff still have the same login, the same policy booklet and the same inbox to ask.
A good answer
Names something a member of staff can do on day one that they couldn't do before, and shows it on a phone.
A weak answer
"Your team will have more time for employees." That is a hope about your team's time. Ask what the product itself puts in front of staff.
Then ask
Can I see the employee's screen before and after?
2When someone is stuck, who answers them?
The days that matter are a diagnosis, a claim that has been turned down or a bereavement. On those days people want someone who knows the policy and can ring the insurer.
A good answer
Names a person or a team, says how quickly they reply, and lets your people reach them without getting past an agent first.
A weak answer
"The assistant handles most questions and passes on the rest." Ask who decides what counts as the rest.
Then ask
On the day of a diagnosis, what is the first thing my employee sees?
3Will it tell us what got used?
Administration data shows who is enrolled and what it cost. It doesn't show whether anyone rang the GP line or used the physiotherapy that came with the policy.
A good answer
Shows the report of what was used, benefit by benefit, and says what happens when something isn't.
A weak answer
A dashboard of enrolment and cost. That is the administration again.
Then ask
Which of our benefits was used least last year, and what would you do about it?
4What does it know about each person, and who else sees it?
An agent needs data to act. Policy documents and member lists carry names, dates of birth and salaries, and staff questions can include health details.
A good answer
Lists what the agent reads, names the AI model and where it runs, says in writing whether your data trains a model, and reports to managers in groups only.
A weak answer
"It's all secure and compliant." Ask for the list.
Then ask
If I ask it about one named employee, what will it tell me?
5Whose saving is it?
Time saved has to go somewhere. Ask how it was measured, and where it lands. If none of it comes back to you, in your price or in better cover, your people won't notice.
A good answer
Says how any time saved was measured, and how much of the saving comes back to you, in fees or in better cover or support for your people.
A weak answer
A figure for hours saved, with no word on how it was measured and no change in what you pay.
Then ask
How did you measure the time saved, and what happens to our price if it holds up?
What passing looks like
A tool that passes all five is worth buying. One that only takes the admin away has changed nothing for your staff.
Our own answers
We are an employee benefits broker with our own platform, so we have an interest here.
What will our people see? Every benefit they belong to, what it covers in plain English and how to use it, with the support services confirmed for each policy.
Who answers them? A person replies, usually within a working day. Never a bot. Our AI doesn't write or draft replies to staff. When we're your broker, that person is Rachel or someone on her team. On the free platform, staff get plain guides to their own cover.
Will it tell us what got used? Partly. We count how many people open each benefit in the app and show you the count each quarter, benefit by benefit, shown only where five or more people are counted, never who. That shows interest, not claims. We're switching it on company by company, starting with our own.
What does it know, and who sees it? Only what a job needs: the policy document you upload. Your admin switches AI on first, on a plain page that says what it reads and what it never sees. If your admin connects your own assistant, it sees check-in results only as averages of five or more people. Ask us in writing which AI model reads it, where it runs, whether your data trains it and how long a document is kept, and we'll answer in writing before your admin switches it on.
Whose saving is it? There is no separate charge for the agent, so any time your team saves is yours.
Which AI model reads our documents, and where it runs.
Whether our data is used to train any model.
How long documents and messages are kept.
A list of what the agent can change without a person approving it.
Where the log of what the agent did is kept, and who can read it.
What happens to our data if we leave.
The price, and what would change it.
Common questions
What is an AI benefits agent?
Software in which an AI agent carries out the administration of employee benefits, such as enrolment changes, provider files and reconciliation, usually with a person approving anything that changes a record. Some suppliers call this agentic benefits.
Does an AI benefits agent change anything for employees?
Not by itself. An agent built for the benefits team takes administration off the team's desk. Whether employees find and use their cover depends on what they are shown and who helps them, which is what questions 1 and 2 test.
Who should ask these questions?
Anyone being shown an AI tool for employee benefits: HR and people leads, reward managers, and the finance director signing it off.
When should we ask them?
Before the demo, in writing, so the answers come back in writing. Then ask to see each answer on screen.
Is it safe to give an AI agent our employees' data?
That depends on the supplier's answer to question 4. You need to know what the agent reads, which AI model reads it and where, whether your data trains a model, and what the agent can change without a person approving it.
Can we use an AI benefits agent without a benefits team?
Some are built for employers with a benefits team to run them. Ask the supplier who checks and approves the agent's work when nobody at your company has benefits as their job, and the price for a company like yours.
Do these questions apply to Alltoogether?
Yes. Our own answers to all five are on this page.
Your insured cover, the support that comes with it, your perks and your policies in one free staff app, with your logo on it. Free for your staff, for ever. No card.