The system that rates and reports your specialist schemes

A typical business running one of these has 20 to 120 staff, a handful of specialist schemes (motor trade, taxis, holiday lets, specialist contractors, classic vehicles) and a few hundred to a few thousand policies a month, quoted and renewed by the phone team, introducers and a web quote form. Those are composite figures. Most brokers put their open-market business through a bought broking platform. This page is about the system beside it, written for the schemes the platform cannot rate, and most common at specialist brokers and managing general agents writing under delegated authority from one or two insurers. Written in 2006 or in 2020, it does the same job.

In the scheme manager's words, it rates each risk from the scheme's tables with its loadings and discounts, refers what falls outside the binding authority to the insurer's underwriter, issues the schedule and certificate, handles mid-term changes, cancellations and renewals, takes the premium or the instalments, works out the insurance premium tax and commission, and builds each insurer's bordereau at the month end, with claims where the broker handles them.

The rules in it belong to the schemes:

• A risk with recent claims, a value above the scheme's limit or a postcode in a band the insurer watches goes to the insurer's underwriter, and the office cannot bind it.
• New rates apply only to quotes from their effective date, and an issued quote keeps its rates until it expires.
• A mid-term change is charged or refunded pro rata to the day, with the premium tax worked out again on the difference.
• Each insurer sets its own bordereau columns, codes and cut-off, and a policy bound after the cut-off goes on next month's.
• A renewal on a retail home or motor scheme may not be priced above what the same customer would be quoted as new business.

Written somewhere between 2002 and 2013 by a contractor who understood the first scheme, then grown one scheme at a time. The quote and policy screens are a Windows program in Visual Basic 6 or a Microsoft Access database, the data sits in SQL Server, and the schedule and certificate merge into Word templates. Between 2010 and 2013 the introducers were given a quote form of their own, an ASP.NET Web Forms site on the .NET Framework reading the same database. The bordereau is a query pasted into Excel on the last working day and reshaped by hand for each insurer. All of it runs on a server in the office.

The contractor has since retired or moved on, and the rates for each scheme ended up split between tables in the database and figures typed into the code. When the insurer agrees new rates, someone edits the tables, runs a handful of test quotes against last year's prices, and trusts the code has nothing else to add.

The same scheme system built between 2016 and 2022, usually because the broker wanted customers to buy a policy without ringing. A small agency or a freelancer wrote it in PHP or React, Angular or Node, or someone put it together in Bubble. From the office it looked as though the schemes had caught up: a customer gets a price late at night, pays by card or signs up for instalments, and has the schedule and certificate by email before the phone team is in.

The rates are what pin it down. The rating tables were written into the quote journey's code, so every rate change the insurer agrees joins a developer's queue, and new rates can go live after the date they were meant to start. Then one insurer changed its bordereau format, the export stopped matching the columns it now expects, and the month end went back to a spreadsheet. Since 2023 some brokers have tried again in an AI app builder, and the attempt stalled at the referral rules. These days the scheme manager asks each insurer for new rates to start a month or more out, to give the developer time.

It prices a specialist risk on the call, inside the authority the insurer granted, and reports every policy to the insurer that carries it.

A specialist scheme is priced on questions a general platform does not ask, and the scheme system holds those questions and the rates behind them. A new handler quotes a risk the way the scheme manager would, and the office writes the business the insurer has given it authority for without phoning the underwriter each time. The referral rules protect that authority: what should go to the underwriter does, and the rest binds while the customer is still on the line.

The money is the other half. Premium, tax, commission and instalments are worked out once, on the policy, and the bordereau is built from the same records, so what the insurer receives matches what was sold. The renewal run picks up each month's renewals without anyone keeping a diary.

The quoting still works. The pressure comes from the regulator, the insurer and the server cupboard.

The Consumer Duty wants outcomes it never recorded. The FCA's Consumer Duty has applied to open products since 31 July 2023 and to closed books since 31 July 2024, and it asks for fair value and good customer outcomes to be evidenced, with an annual assessment by the board. Old program or agency build, the system holds the policy and the premium but nothing on cancellations, claims, complaints or renewal retention by customer group, so the evidence is pieced together from exports.

The auditor asks for history the system overwrote. The insurer audits the delegated authority and asks which version of the rates priced a given policy, and what became of each referral. The old tables were replaced when the new rates went in, and the referrals are in someone's inbox.

The bordereau is late, or it comes back. Each insurer's file is built from a query and a spreadsheet in the last days of the month, and a rejected row means finding the policy behind it and sending the file again. Late and rejected bordereaux are a standing item between broker and insurer.

The server has an end date. Often that is Windows Server 2016 over a SQL Server 2016 database. SQL Server 2016 stopped receiving updates in July 2026, Windows Server 2016 follows in January 2027, and any still on Windows Server 2012 have been without support since October 2023.

The same schemes and screens, with rates the office can change, a trail the auditor can follow, and bordereaux built from the policies.

Rates the scheme manager loads. Each scheme's tables come out of the code as versions with an effective date, loaded by the office and tried on test quotes before the date arrives. Every quote records the version that priced it and keeps it until it expires.

One price, however the risk arrives. The phone team, the introducers and the web quote read the same tables, so a rate change reaches all three on its date.

Referrals on the record. The trigger, the underwriter's answer and the name of whoever bound the policy, kept against the quote. The next audit gets its referral log from a report.

Bordereaux at the cut-off. Each insurer's columns, codes and cut-off are held as settings, and the file is produced from the policy records and checked against that insurer's rules before it is sent. A new format is an edit to a setting.

Outcomes kept against each policy. Cancellations, claims, complaints and renewals recorded by scheme and customer group, so the fair value review and the board's annual assessment start from a report.

Off the 2016 server. The database on a version still supported, on a server that is too, with a backup that has been restored on purpose.

Nothing here needs the old program or the agency's site turned off. Each piece goes in beside it, on the policies already there.

The reading and matching around each quote and bordereau, with a person deciding any price or referral.

Proposal forms read into draft quotes. An introducer emails a proposal form, a schedule of vehicles or a few paragraphs about the risk, and a handler types it in. AI fills in a draft quote from it, marking any answer it could not find, for the handler to check and rate. At about 30 emailed submissions a day and five minutes' typing saved on each, that comes to two and a half hours a day across the quote team.

Bordereau rejections traced to the rows. The insurer's rejection report lists the rows that failed and why. AI matches each one to the policy behind it and drafts the fix, a missing postcode or a tax figure a penny out, for whoever builds the bordereau to approve. If 60 rows come back from a file of a few hundred to a few thousand, tracing and fixing each at seven or eight minutes adds up to a working day of lookups; approving the drafted fixes at a minute apiece takes an hour.

Quotes whose answers moved. A re-quote where the declared claims dropped after a high price, or a mid-term change that moves the postcode into a cheaper band, is flagged for a person to review; AI decides nothing. At about 30 re-quotes and mid-term changes a day, comparing each with its earlier answers by hand would take half an hour at a minute apiece; five or so flags at three minutes each come to a quarter of an hour each morning.

The book asked questions in plain words. "Which renewals next month had a claim this year?" "Which scheme loses the most policies in the first month?" AI answers from the policy records and lists every policy it counted. Three or four such questions a week, at half an hour to write and check each query, comes to an hour and a half to two hours a week for whoever writes them.

Not worth doing: AI setting rates or deciding a referral. The rates are agreed with the insurer and the referral belongs to its underwriter under the binding authority.

Each reads the policy records or drafts into them, so none starts before the first stage makes the scheme system safe to change. The assessment maps where AI would pay back on the broker's own schemes.

Safe to change first, then the rates, then the month end, with the old system quoting throughout.

1. Stabilise. The VB6 project, the database file, the Web Forms site or the agency's code found and matched to what the office runs, the database on a version still supported, a backup restored on purpose, and a test copy that can rate a quote without binding anything. This goes first because every rate change the insurer agrees passes through this system, and because Windows Server 2016 support ends in January 2027.
2. The rating tables. Each scheme's rates come out of the code into versioned tables, and a sample of recent quotes is rated both ways until the prices agree. The screens do not change.
3. The bordereaux. Each insurer's file produced from the policy records in its own format, run beside the spreadsheet until the insurer accepts it first time.
4. The referral log, the outcome data, the quote screens in a browser. Each a stage, each priced on its own.
5. Retire the old program when nothing still depends on it. The renewal run usually goes last, because the whole book passes through it.

The assessment is from £395 + VAT, sized on a free one-hour consultation, with an exact price before you commit. For a scheme system that means a few days with the scheme manager, the accounts team and the code, finding the rates, the referral rules and the risks, and a report with costed options and a fixed price for each. Each stage after that is priced before you commit to it. How we work, what drives the price and the payment terms each have a page.