How we built private AI to prepare a family’s annual insurance review in as little as 13 minutes

Hundreds of figures, copied by hand from every insurer’s statements.

Wealthy families hold several life policies, often in trusts. Every year or two, sometimes three, the firm reviews the lot.

For each family, the team gathers every policy’s annual statement and latest projection from the insurers, and every insurer lays its PDFs out differently. Someone copies the values into the firm’s Excel review workbook: a cover, a schedule, a page per policy and the disclosures. They carry last time’s wording forward, then hand it to the producer, the licensed adviser, who writes the commentary and approves it.

A typical review has 285 figures to fill. Every review needs two people to approve it.

Each policy checked separately.

We split the work the way the team already does it: policy by policy.

A team member selects the client in a web app. It suggests the statements and projections from that client’s folder, the latest review and the current template. Some families are reviewed only every two or three years, so it finds the latest review regardless of its date. The team member confirms the files, chooses the adviser and starts the review.

The system groups the statements and projections by policy. It reads each policy separately and records the source page and line beside every figure.

It uses those figures to update the previous workbook, keeping the firm’s layout and wording. For a first review, it fills the blank template. Our code writes each figure to the required cell and preserves the workbook’s formulas.

A separate check compares each policy with its sources and checks the whole workbook. Errors go back for correction before the team receives the draft. The firm’s rules and instructions for each insurer are written in plain English and can be edited by the team.

Every number, next to where it came from.

The team reads the draft with the source open beside it.

Click a premium, a death benefit or a cash value and the insurer’s PDF opens at that page with the line boxed. If the team wants changes, they list them together and get one clean new version back.

If a figure is missing, the system flags it for the team. It leaves a cash value unresolved when none of the supplied statements shows one.

The preparer and the licensed adviser approve the same exact version. The finished workbook lands in the client’s SharePoint folder with a receipt. The system never overwrites or deletes a file.

Checked against the team’s own work.

We compared the prepared review, figure by figure, with one the team had already made by hand.

Every hard number but one matched the team’s to the cent or the day: premiums, death benefits, policy dates and policy years. The remaining difference came from choosing between two valid source records.

Where nothing had changed since the previous review, the system kept the firm’s wording.

A returning client’s review was ready to check in as little as 13 minutes. In a separate run, the whole cycle, including a round of changes, cost as little as $7. The firm’s limits were 25 minutes and $15. Its own team has used the system to run a review.