Skip to content
Home

Tools

Demo tools for actuarial and insurance work, each built so its answer can be checked: an AI proposes, and code you can read does the computing and the testing. Data Science is open to anyone; the other tools need a link from me, since most of them call a paid model: ask for access and I'll send one.

What do you want to do?

All 12 tools.

Check Data

Close the Gap

Reconcile

Why two files disagree, traced row by row, to the penny.

Two sets of data that should agree and do not. It matches them, measures the gap, tests explanations against the actual rows, and bridges one total to the other.

More about Reconcile

Twelve detectors look for the ordinary causes (leading zeros, amounts in pence, a missing filter, duplicate rows, gross against net), each measured on your rows before you see it, and a model can propose more under the same test. On the motor example, taking 12% IPT off 518 premiums explains £36,312; on the annuity one, multiplying 57 incomes by 12 explains £136,727 of the difference. A model sees a profile of your columns, never your rows. Accept the causes you believe and it draws the bridge from one total to the other, with whatever is left standing in its own colour.

Every actuarial team reconciles, and the afternoon goes on hunting for the reason. Drop in two files and it pairs the columns, joins on the key and splits the difference by whatever you group on. Then twelve detectors look for the ordinary causes (leading zeros, amounts in pence, a missing filter, duplicate rows, gross against net), each applied to your rows and measured before you see it. A model can propose more under the same test, so on the worked example what you read is “multiplying these 57 annuity incomes by 12 explains £136,727 of the difference, and here they are”. Accept the ones you believe and it draws the bridge, with whatever is left standing in its own colour. Nothing is uploaded: a model sees a profile of your columns, never your rows.

  • Runs in your browser, AI on a server
  • Data
  • Validation
  • Automation
Open ReconcileBy invitation

Build or Recover a Calculation

Find the Formula

Backsolve

Give it inputs and answers; it finds the calculation and proves it on unseen examples.

Give it worked examples, inputs in and answers out, and it works out the calculation behind them, then hands it over as a spreadsheet, as Python and as a write-up.

More about Backsolve

A model proposes the calculation step by step and records every assumption; the Python and the spreadsheet are run against every example, and a miss goes back to be fixed. Examples it was never shown are scored once, at the end. Everything that computes runs in your browser; only the prompt leaves the machine. An input the examples never vary is reported as untested rather than guessed, and one that changes nothing, as the driver’s age does in the motor example, is dropped and said so.

Two models share the work: one proposes the calculation and a critic from a different provider reviews it. Your examples go to them through OpenRouter, and Run lets you replace the numbers with stand-ins, and the column names with placeholders, before they are sent. You also set how many attempts it gets, a spending limit, how close an answer must be to count as right, and whether about a third of the examples are held back for the final test. The Excel can be laid out with periods across the columns or down the rows.

  • Runs in your browser, AI on a server
  • Modelling
  • Automation
  • Validation
Open BacksolveBy invitation
Watch How It Works· 3:24

Words to Working Model

Calc Builder

Describe a calculation in words; leave with tested Python and Excel that agree.

Describe an actuarial calculation that does not exist yet and leave with a specification you confirmed, tests you read, Python and an Excel model that agree on every policy, and a validation summary structured to TAS 100.

More about Calc Builder

Two models each write the specification and you settle where they differ. Tests are written from it before any code exists, in four kinds only, and the code, written without sight of them, runs against them in your browser, then over every policy in your file. An Excel model built from the specification with no AI is checked against the Python policy by policy. Then you try it on figures of your own, see a map of how its parts link, and accept its acceptance tests.

Backsolve works out a calculation that already happened; this builds one that has not been written yet. Two models each write the specification; you settle where they differ, or take the best guesses. Tests are written from it before any code exists, in four kinds only, and the code, written without sight of them, runs against them in your browser, then over every policy in your file. An Excel model built from the specification with no AI is checked against the Python policy by policy. At Accept you try the finished calculation on figures of your own, worked out again as you type by the code and by the Excel model, see a map of how its parts link with each figure in its box, and accept or reject acceptance tests made from the specification, adding cases of your own. You leave with the specification, code, tests, workbook, the accepted acceptance tests, the map and a summary structured to TAS 100.

  • Runs in your browser, AI on a server
  • Modelling
  • Testing
  • Validation
  • Automation
Watch How It Works· 3:30

Speed It Up

Accelerator

Your Excel model on a GPU, checked against Excel on every policy.

Bring a bulk actuarial workbook and leave with it in Python: every formula carried across by fixed rules, every policy checked against Excel’s own answer, and the whole book run on the GPU in this browser.

More about Accelerator

Where the rules refuse a formula or misread the layout, the AI can rewrite it in formulas they carry, each proved against Excel’s own answers; the translation itself is rules, with Excel’s own rounding and lookups. The Python matches Excel to nine significant figures and the GPU to one part in 100,000. A million policies of a 480-month term assurance run in about a quarter of a second on a desktop GPU. What cannot be carried across (macros, circular references, volatile functions) is refused, naming the cell.

A bulk calculation in Excel is an array program in disguise: a formula filled down is one calculation over every policy, a formula filled right is a step through time. Accelerator reads the workbook in the browser, finds the policies (a sheet that works one policy at a time, the policy picked by a cell, or a row per policy), finds the lines through time and the answer, and carries every formula into Python by fixed rules, with Excel’s own rounding, INT, MOD and lookups. The Python runs here on every policy under Pyodide, and the same translation runs as a WebGPU kernel on the visitor’s own GPU, each checked against Excel’s answer policy by policy. The same run takes a little over a minute in Python alone, against the quarter of a second above on the GPU. What-if tables are refused too, naming the cell. The download is the Python with its tests, the inputs as CSV, a reconciliation workbook, the mapping from each Excel range to its line of code, an audit of the workbook, and a summary set out as TAS 100 asks.

  • Fixed rules, run on your GPU
  • Modelling
  • Validation
  • Automation
Open AcceleratorBy invitation
Watch How It Works· 3:14

Fit a Model

Looped Agent Pricing

Auto Research

An AI improves a pricing GLM one change at a time, keeping only what holds.

An agent proposes a GLM, a fixed harness fits and scores it, and only improvements are kept, on eleven example files with structure planted in the data, or on a file of your own.

More about Auto Research

Each example has a factor carrying no signal, to show whether the agent buys parameters that earn nothing. Each change is scored on rows the model was not fitted on, and charged for every parameter it adds. The holdout is scored on every fit but shown only at the end. The stepwise search, which asks no AI, ties with the agent on mortality: the value is the reasoning and the write-up, not the last decimal of deviance. Every fit runs in your browser.

A harness the agent cannot touch fits each model, scores it against a cost function you can edit, and keeps a change only if the score fell. The charge is a bar of 3.84 per parameter, the 5% point of a chi-squared test, so a change smaller than noise is not kept; Customise offers AIC, BIC, ridge, lasso or a charge of your own. When a run ends, each change it kept is scored again by cross-validation, and Results says where one was kept on a single split but not confirmed. The examples are taken just as a file of yours would be. Download gives the models as an Excel workbook of live formulas, checked as Excel works them out, and as plain Python.

  • An agent edits the model
  • Modelling
  • Pricing
  • Automation
Watch How It Works· 3:39

AI Data Science Analysis

Analyse

Load a policy table; get a scored pricing review and a TAS 100 paper.

Load actuarial data of any shape and get a first, high-level analysis: an AI model maps the file into a table to analyse, each column with its meaning and a reason, and Python runs its mapping over every row and proves it, totals reconciled.

More about Analyse

Python then checks the data and runs what suits it from ninety-five analyses across pricing, basis fitting, reserving, capital, data science, validation and conduct, or one you describe in words. Findings are merged, scored for importance and graded for reliability on rows held back, every p-value corrected for the tests run. Tick the findings that matter and the AI drafts a paper to TAS 100, every figure placed by Python; an audit folder reruns it all.

A first look at a new table usually means a week of one-way tables, a GLM and a chain of emails. Here the columns decide what runs: a yes-or-no event with exposure gets rates, actual against expected, a percentage of the table, graduation and its tests of fit, a Poisson GLM, a trend and survival curves; claims with their cost and premium get frequency and severity, the model against the current price, banding and smoothing; claims transactions get triangles, the chain ladder with Mack, Bornhuetter–Ferguson and a bootstrap; amounts get heavy-tailed fits and an aggregate loss with its value at risk. Nothing is built for one kind of data, so motor claims and renewals, home claims, mortality, lapses, income protection and take-up all go through the same list, and an analysis it lacks can be described in words: built from its own tested blocks, or written in Python by the AI and checked. The Python that made every figure goes into an audit folder with a log of every choice, and a script there reruns the analysis and checks every figure matches.

  • Runs in your browser, AI on a server
  • Data
  • Modelling
  • Pricing
Open AnalyseBy invitation
Watch How It Works· 3:32

Review and Control

Guard-Railed Sales Call

Callflow

An AI takes an annuity call; plain code decides, prices and blocks unsafe sales.

A customer has rung about an annuity, a pension pot swapped for an income for life, and a voice agent sells it by following a fixed script. The AI model only works out what the caller said. Ordinary code decides what happens next, works out the income, and will not let a sale through until every warning has been said.

More about Callflow

Sixteen obligations are tied to the words that discharge them, and thousands of fuzzed calls have never sold past one. The same engine sells term life cover and car insurance, each a file of data with its own facts, steps, wording and rating table, and the whole call opens as a flow chart to redraw, check and walk through with buttons.

The usual voice sales agent is a prompt describing the product and the rules, and hope. This inverts it. Each turn the model is handed one sentence and returns a filled-in form (an intent, an objection, a vulnerability flag, some numbers) and everything else is code: it never chooses the next node, never sees the quote, and is not told whether the sale is going well. Sixteen obligations are tied to the nodes whose words discharge them, and a fuzzer has run two thousand calls without reaching a sale with one outstanding. Any sign that the caller is not in a position to decide stops the sale. The same engine runs term life cover and car insurance, each written as data and priced from a rating table, and the whole call, whichever product, opens as a flow chart: change the words, add a question or an arrow, change the price, check it, walk through it with buttons, and save it to a file. The quote is a real annuity calculation: a Makeham survival curve, escalation, a guarantee period, a joint life and health priced as an age rating. On the cheapest model a whole call costs about 8p, natural voice included, and anyone can name the firm, reword the questions or add their own before they ring.

  • Voice agent, scripted engine
  • Pricing
  • Automation
  • Validation
Open CallflowBy invitation
Watch How It Works· 2:33

Standards Check

Doc Solve

Check a report against TAS and the Code; get the fixes as tracked changes.

Review an actuarial report against the standards it must meet: a Gap Score for what is missing, and suggested changes written into your own Word file as tracked changes.

More about Doc Solve

Pick the standards to hold it to, from TAS 100 and the Actuaries’ Code to Solvency UK, Consumer Duty and IFRS 17, or add your own. The AI gives every check a verdict with the words that show it, and each quote is looked for in the document; the app adds up its figures itself. Changes for the gaps leave any fact only the author knows in square brackets, to accept, reject or reword. The download is your own file with the accepted changes tracked and each gap a comment, a clean copy, and the checklist as a workbook.

Load a Word document, say what it is for and who reads it, and click the standards to hold it to: TAS 100 and TAS 200, the Actuaries’ Code and the APSs, the PRA’s Solvency UK reports, the FCA’s Consumer Duty, product governance and pricing rules, Lloyd’s, IFRS 17 and good practice, eighteen packs checked against their sources, or your own, typed, loaded from a file or drafted by the AI. A requirement several standards share is one check. The AI gives every check a verdict with the words that show it, and each quote is looked for in the document; the app adds up the document’s own figures, each table’s totals and the summary against the tables, with no AI; a Gap Score counts what is missing by fixed points, and a verdict you overrule changes it at no cost. The AI then writes changes for the gaps, leaving any fact only the author has in square brackets, and they are shown as Word shows tracked changes, to accept, reject or reword. The download is your own file with the accepted changes tracked and each gap a comment, a clean copy, the checklist as a workbook and the review as a page.

  • AI judges, rules score
  • Validation
  • Automation
Open Doc SolveBy invitation
Watch How It Works· 3:30

Solvency in Real Time

Live Balance

Drag a slider; the whole annuity balance sheet revalues in a fraction of a second.

A UK annuity writer’s Solvency UK balance sheet, revalued on your GPU as you move the sliders. Larkfield Retirement, a made-up writer with 100,000 pensioners, 30,000 deferred members and the assets behind them, is valued on the PRA’s published curve and fundamental spreads, under the standard formula as the Rulebook stands today.

More about Live Balance

Move rates, spreads, credit, mortality, reinsurance, hedging or capital, or set the PRA’s 2025 stress test or the 2022 gilt crisis in one click, and every member is valued again in a fraction of a second, every chart following. It runs a reverse stress test, simulates a year against the formula, projects five years and prices a buy-in, and leaves the figures as files. Or load a book of your own.

  • Runs on your GPU, AI explains
  • Capital
  • Reserving
  • Modelling
Watch How It Works· 2:52

Process Watchdog

Delta Change

Compare each run of a spreadsheet process: what moved, why, and what looks wrong.

Check a process run period after period, made of several linked spreadsheets: see how the files connect, what changed since the last run, and what looks wrong.

More about Delta Change

Drop each run’s files, month-end, quarter-end or year-end, into its own zone. The files are paired across the periods, mapped from their formulas and compared cell by cell. Forty checks look for what goes wrong at a period end: a link still reading last period’s file, a formula typed over, a total that stops short of the new rows, an assumption changed with no note, records dropped or copied twice. A bridge recalculates the files to show why the headline figure moved. The AI names the files, judges the flags and writes the report; the files stay on your computer.

Drop the files of a monthly process into one zone per month: a policy extract, the pricing assumptions, conversion and retention by channel, a renewal pricing check and the board summary, say, for two or more month ends. Each file is paired with its partners in the other months by its name with the dates taken out and by its sheets, and the process is drawn as a map from the files themselves: a formula reading another workbook proves a link, and says which month’s file it reached, so a link left pointing at last month’s file shows at once. Buttons over the map pick out one kind of problem at a time, and Trace Back lights everything that feeds a figure. Each month is compared with the one before, cell by cell, with formulas read relative to their own cells so a range grown to take in new rows is not a change. Forty checks, each of which can be switched off, given a severity or threshold and reworded, look for stale and broken links, typed-over formulas, totals that stop short, checks no longer nil, history restated, duplicates, dropped records and assumptions changed with nothing to say why, and you can add checks of your own in words. A bridge recalculates the files together, swapping one month’s file for the next at a time, so every step of the headline’s movement is a file and what the files do not explain is its own step. The download is an Excel workpaper with the bridge in live formulas, a Word review and the map.

  • Rules flag, AI judges
  • Validation
  • Data
  • Automation
Watch How It Works· 3:31

Learn

Data Science Training

Data Science

Data science for pricing, each idea a level scored on unseen policies.

Supervised learning for pricing work as a map of explainers (splines instead of bands, ridge as credibility, the lasso), each with levels in the style of a learning game.

More about Data Science

Every level is dressed in a pricing decision, such as a lapse model or an age curve an underwriter will sign, and scored against unseen policies. Every number is computed in the browser from the same book the levels use; no model is called and nothing is uploaded. The branches run from learning from data and linear models, through classification, smooth curves and choosing a model, to additive models, trees, and inference and averaging.

The map is drawn four ways and you can switch between them: Rings on a wide screen, Metro on a phone, Route, and the Hub it began as. Each branch starts from a plain question, such as why a 0/1 flag breaks least squares, what splines do that bands cannot, or how optimistic a training error is. The lasso is tried on a lapse flag, local fits on claim counts, and credibility is read as Bayes, with expectation-maximisation, Gibbs sampling, bagging and stacking at the far end.

  • Explainers with levels
  • Training
  • Modelling
  • Pricing

Keep Up

Actuarial Newsroom

News

The week’s insurance and AI news, filed by practice area, with an AI to ask.

The latest insurance and AI news from the trade press, the regulators and the labs, filed the way an actuary’s week is, with a model on hand to answer a question about the areas you choose.

More about News

Thirty-seven feeds are read about twice a day and committed as one list, so the page is instant. It keeps no more than eight hundred stories, usually about a week; the same story from four outlets becomes one, and the most-told sit at the top. A question is answered from the newest four hundred items in the areas switched on, with the items it drew on.

Thirty-seven feeds are read about twice a day by a scheduled job and committed as one list, so a visit fetches nothing and the page is instant: the trade press, the FCA, the PRA and the Bank of England where they touch insurance or pensions, and searches by subject, from bulk annuities and the CMI to Ogden and AI in claims. The list keeps no more than eight hundred stories, usually about a week, and says the dates it covers. The same story from four outlets becomes one story with four tellers, and the most-told stories sit at the top. Switch on one practice area or several and ask a question: a model reads the newest four hundred of their items and answers with those it drew on, each checked against the list. Copy any view as a Markdown briefing for the Monday note.

  • Refreshed about twice a day
  • News
  • Automation
Open NewsBy invitation
Watch How It Works· 3:03

© 2026 Alex Aicken

Home