Split mid-term changes by what they touch. An agent can apply non-rating changes, such as a corrected address, at depth 3 once certified. Changes that alter premium go through the rating engine and stay at depth 2 until proven. Backdating, underwriting referrals, cancellations and premium refunds stay with a named person, and every change records the policy before and after.
What counts as a mid-term change, and who owns it?
A mid-term change, often called a mid-term adjustment, is any amendment to a live policy between inception and renewal. Some are clerical: a new phone number, a corrected postcode, a change of bank details for collection. Some change the risk: a new driver, a different vehicle, an extension built on the insured property, a higher sum insured. Some change the contract itself: removing cover, adding a section, cancelling.
The policy administration system holds the contract. Underwriting owns the rules for which changes are acceptable and how they are rated. The operations or servicing team owns the process of taking the request and applying it. Finance owns any money that moves as a result. The agent sits inside the servicing process and must respect all three owners.
Which changes should the agent make, and at what depth?
Connectivity comes first, and connecting an AI agent to your policy administration system covers the service account, test environment and read-only start. Once the agent can read policies, grant actions in this order.
- Agent, depth 1: reads the request and the policy. It identifies the policy, the change requested and the effective date, and checks the request came from the policyholder or an authorised party.
- Agent, depth 1: classifies the change. Clerical, rating, contractual or cancellation. The classification decides everything that follows.
- Agent, depth 3: applies clerical changes that do not touch rating factors, effective from today.
- Agent, depth 2: prepares rating changes. It enters the change in quote mode, runs the rating engine and drafts the customer message with the additional or returned premium. A person reviews and applies it. Once Attest evidence supports it, routine rating changes under a premium threshold can move to depth 3.
- Agent, depth 1: routes referrals. Where the rating engine or underwriting rules refer the change, the agent stops and hands the case to an underwriter with the facts gathered.
- Person, underwriter: decides referred changes, backdated effective dates and any change the rules do not cover.
- Agent, depth 2: drafts the endorsement and schedule after a rating change is applied. The documents come from the system's own templates.
- Person, finance or servicing lead: approves return premiums. Refunds move money, which is depth 4 work, and stay held.
- Refused for the agent: cancellations, removing cover, and changing the policyholder.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Read and verify | Identifies policy, change, requester | 1 | Authority disputes |
| Classify | Clerical, rating, contractual, cancellation | 1 | Nothing |
| Clerical change | Applies it | 3 | Nothing |
| Rating change | Quotes and drafts | 2, later 3 under threshold | Applying above threshold |
| Referral | Routes with facts | 1 | Underwriting decision |
| Documents | Drafts endorsement | 2 | Sending |
| Return premium | Prepares the refund | held | Approval |
| Cancel, remove cover | Nothing | refused | All of it |
Why do backdated changes stay with a person?
A change effective today only affects the future. A backdated change rewrites the past. If a driver was added effective last month and a claim happened in between, the backdate decides whether that claim is covered. That is an underwriting and claims decision, not servicing.
The agent should treat any effective date before today as a referral, whatever the change. It can gather the facts, including whether any claim was opened in the gap, and hand them over. If your agent also works on first notification of loss, the boundary in letting an AI agent register claims should be drawn so that the same agent never both backdates a policy and registers a claim against the gap.
What is recorded for each change?
The request and who made it. The original message, the channel and the identity check. A request from a broker or a family member records the authority relied on.
The policy before and after. Every field that changed, old and new values. This is the record you will need when a customer disputes what they asked for.
The rating run. Rating engine version, factors used, premium before and after. If the agent drafted and a person applied, both names.
The verdict and the deciding bound. A change held because it exceeded the premium threshold shows the threshold. A change refused because cancellation is out of scope shows the role scope.
Documents issued. The endorsement and schedule versions, and when they were sent.
What should you measure?
- Clerical changes reworked after the agent applied them. Target: none.
- Rating drafts amended by reviewers, and by how much premium.
- Referrals the agent missed, found by underwriters in sampling.
- Elapsed time from request to applied change, clerical and rating separately.
- Customer disputes about what was changed.
Where does it usually go wrong?
Clerical changes that are not clerical. A change of address can change the rating territory. A change of vehicle registration can be a different vehicle. Classification must check whether a field feeds rating, not whether it looks administrative.
Declared information is treated as verified. A customer saying the new driver has no convictions is a declared fact. The agent can record it and the rating engine can use it, but the record must show it as declared so underwriting knows what was checked.
Life and health lines are treated like motor. Under the EU AI Act, risk assessment and pricing for life and health insurance is listed as high-risk. If your agent influences re-rating on those lines, the obligations are different; read what changes when the system acts before widening its depth.
Training cases are too clean. Real requests often mix three changes in one email. Certification should include those, as the sequence in train, shadow, certify, widen recommends.
Where Pryme fits
In Pryme, clerical changes, rating changes and documents are separate capabilities, each with its own depth. Attest replays the agent against your past adjustments, and the customer's own premium threshold becomes a limit in the minimum chain. A change above it is held for a named person, and the record names that limit as the deciding bound.
Questions people ask
Can the agent handle changes that come through brokers?
Yes, if the broker's authority to act for the policyholder is recorded in the system. The agent checks that authority before acting and routes the request to a person if it is missing.
Should the customer be told an agent made the change?
Yes. The confirmation should say the change was processed by an AI system and name the team the customer can contact if anything is wrong.
What if the customer changes their mind after a rating change?
Reversing a change is a new mid-term change. The agent prepares it like any other, and the record shows both the original change and the reversal.