New Member Data Extraction now runs as an Excel add-in. See the data quality map →

AI for pensions actuaries

Every answer traced to source

Sourcestone works inside Excel. It can read every document in the scheme and provides a clear map of your data from benefit specifications to member data. Every answer links to the document, page and clause it came from.

Runs inside Excel Built by actuaries
Reading scheme documents…
Data quality map 33 fields, one member
12 issues found 1 signed off 20 no issue
Example Company Pension Scheme.xlsx
HomeInsertFormulas DataReviewAdd-ins Sourcestone
C7fx  
Illustrative benefit specification with answers traced to source documents
RowBenefitBrief answerDetailed answerReference
5General
6Scheme established 6 April 1974 Established by Interim Trust Deed dated 6 April 1974, now governed by the Definitive Trust Deed and Rules dated 1 October 2011. Deed · p.1
7Status Closed to future accrual Closed to new entrants 1 April 2002 and to future accrual 31 March 2011. 3rd Deed of Amdt · p.2
8Contracted out status COSR to 5 Apr 2016 Contracted out on a salary related basis to abolition. GMP accrued to 5 April 1997, s.9(2B) rights thereafter. Rule 3 · p.7
9GMP equalisation Not in the documents Not addressed in the deeds or amendments provided. Refer to legal advice. No source
10Pensionable service definitions
11Pensionable Service Years and complete months Membership to the earlier of leaving service and 31 March 2011, in years and complete months. Rule 8 · p.14
12Pensionable pay definitions
13Final Pensionable Earnings Best 12 of last 36 months Highest 12 consecutive months of Pensionable Salary in the 36 months to leaving, less the Lower Earnings Limit. Rule 1 · p.5
14Normal retirement benefits
15Normal Retiring Date 65th birthday 65 for all members. Barber window service, 17 May 1990 to 30 June 1991, retains NRD 60 for male members. Rule 12 · p.19
16Pension at retirement 1/60th accrual 1/60th of Final Pensionable Earnings for each year of Pensionable Service. Rule 11(1) · p.17
17Commutation of pension Excess over GMP Rule 22(1) · p.47
18Commutation factors Age related, Actuary set Factors set by the Actuary and reviewed at each valuation. Table dated 1 April 2023 in force. Sch 3 · p.61
19Early retirement from active service
20Eligibility From 55, with consent From age 55 with Trustee and Employer consent, reduced as the Actuary determines. Rule 16(2) · p.28
21Early retirement reduction 4% simple a year 4% a year, pro rata for complete months, between the early retirement date and NRD. Sch 3 · p.62
22Benefits on leaving pensionable service prior to NRA
23Deferred pension at DOL Accrued to date of leaving Pension accrued to the date of leaving, calculated as at that date and revalued to NRD. Rule 19(1) · p.36
24Increases in Deferment
25Increases in deferment Statutory revaluation Excess over GMP revalued per the Revaluation Orders. GMP fixed rate revaluation by leaving date. Rule 19(3) · p.37
26Pension Increases (in payment)
27Increases in payment By tranche Pre-1988 GMP nil; 1988 to 1997 GMP CPI capped 3%; 1997 to 2005 LPI 5%; from 2005 LPI 2.5%. Rule 24 · p.51
28Death after retirement
29Widow(er) / civil partner's pension 50% of member's pension 50% of the pension before commutation, plus a 5 year guarantee from the date of retirement. Rule 26(1) · p.55
Benefit Specification Summary Line by line

Illustrative example, laid out as the real output is. Not a real scheme.

Answer traced to source

Benefit Specification.xlsxExcel
Sourcestone
17Commutation of pensionExcess over GMP
Source documentPage 47

Definitive Trust Deed and Rules

At the time a pension becomes payable, a Member may exchange any part of it in excess of the Guaranteed Minimum Pension for a lump sum payment.
Clause 22(1)✓ Source verified

Illustrative example. Not a real scheme.

It's a significant time saving and the references provide a clear audit trail. Chris Bunford FIAPartner & Senior Actuary, Goddard Perry
What it does

Understand your member data.

A single scheme can hide forty years of amendments, spread across deeds, consolidations and announcements that were never built to be searched. Sourcestone reads all of it and shows you where every answer came from. Fundamental to understanding member data are two jobs.

Benefit Specifications

Complex scheme rules, turned into a clear specification.

Sourcestone works through the full document history and produces a benefit specification in your own template, with every section referenced back to the clause that supports it.

  • Your own template, or one of ours
  • Every answer referenced to document, page and clause
  • Member categories handled separately where the rules differ
  • First drafts in minutes rather than days
  • A brief answer and a detailed answer on every line, so reviewers choose how deep to go
Book a demo
BenefitBrief / detailed answerReference
Normal retirement age65; NRA 60 for Barber window serviceRule 12 · p.19
Early retirementFrom 55, consent requiredRule 16(2) · p.28
CommutationExcess over GMP; no cap in the RuleRule 22(1) · p.47
Increases in paymentBy tranche: nil / CPI 3% / LPI 5% / LPI 2.5%Rule 24 · p.51
Spouse's pension50% of pension before commutationRule 27 · p.58
GMP equalisationNot addressed in the documentsRefer to legal
Fully referenced
Click through to the clause
Member Data Extraction

A field-by-field map of your member data quality.

Choose the members and the fields, and Sourcestone can read every document you hold on them. You get back a colour-coded map of what is confirmed, what is missing and what contradicts itself, with the source alongside each answer.

  • Select members by ID, by CSV list, or let it detect them for you
  • Choose from personal, contact, employment and pension fields
  • Every field either referenced to its source, flagged as missing, or flagged as a conflict
  • Blanks are left blank, and said to be blank
  • Sign-off is a tick your reviewer makes, not a status we award
  • Built for the work data quality gates: TPR data scores, and insurer due diligence before a buy-in
Why it works this way
Trustees and scheme managers remain accountable for decisions and outcomes even when they delegate activities to providers or advisers. The Pensions Regulator, AI plan, May 2026

So Sourcestone describes the conflict and shows you both sources. The decision stays with your actuary, and so does the accountability.

The kind of thing it finds

This member's date of birth appears twice in the file, a year apart. The two give different GMP payable ages. Both sources sit in the sheet, so an actuary settles it in minutes rather than finding it during buyout due diligence.

Map one scheme's member data
FieldExtracted valueReference
last_nameKerrKerr_statement · p.1
date_of_birthTwo dates: 1962 and 1963Conflict
national_insurance_numberNot held in any documentNot found
date_of_leaving31 March 2004Kerr_statement
gmp_at_date_of_leavingNot held in any documentNot found
final_pensionable_salary£53,582Kerr_statement
Reference text · 2 documents
Date of birth: 10 February 1962Deferred benefit statement, 2019
D.O.B. 10.02.1963Membership record, 1998

One engine, helping you understand your data. If your team currently spends a lot of answering a question by trawling through the scheme's documents, tell us about it.

How it works

Agents do the reading.
You keep the judgement.

1

It opens where you already work

Sourcestone installs as an Excel add-in and sits in a task pane beside your workbook. Point it at the scheme's documents in your own SharePoint or OneDrive. No new system, and nothing to migrate.

2

The agents read everything

Every deed, amendment, consolidation, announcement and member record, in a single pass, including scanned documents with no text layer.

3

You review against source

Results land in the workbook with the source document alongside them in the pane. Every answer carries a link to the document, page and clause it came from, so review runs against the source instead of starting from a blank page.

Sourcestone agent Example Co. Pension Scheme
  • Definitive Trust Deed & Rules (2004) 312 pp
  • Deeds of Amendment (1997 to 2019) 9 docs
  • Member records and correspondence 1,240 docs
  • Cross-referencing benefit categories done
  • Writing results into the workbook done
Every answer linked to its source document
Assurance

Three things regulated AI has to get right.

A chatbot that guesses and cites nothing is no use to a profession that has to show its working.

Provenance

A clear audit trail for every answer, traced back to source and shown directly in the Excel for fast, accurate review.

Expert judgement

Our AI follows the process you already work to, embedding your standards and your internal sign-offs.

Data security

All data is secure in transit and at rest, and you keep control of your files. They stay where they already live.

These are the same controls your own work already runs on: standards, internal sign-off and peer review.

How access works

One folder, for one job

You give us read access to the scheme documents you want read. Nothing else.

Revoke it whenever you like

Access is yours to withdraw in SharePoint or OneDrive, without asking us.

Deleted when the job is done

We do not keep your documents once the work has been delivered.

Never used to train models

Nothing you give us is used to train or improve any AI model.

Customer story · Goddard Perry
From weeks to hours on a first draft, with every line referenced.
CB Chris Bunford FIAPartner & Senior Actuary, Goddard Perry

The challenge

  • Manual review of multiple legacy PDFs by numerous actuaries
  • Decades-old documents with no search functionality
  • Qualified actuarial resource absorbed by routine document review

What changed

  • The document reading and first drafting handled by Sourcestone
  • Fully linked source references maintaining clear provenance
  • Review conducted side by side against the original documents
  • Fitted into the team's existing actuarial workflow

From weeks to hours

Time to a first draft of a benefit specification, depending on the size of the scheme.

Every line referenced

Document, page and clause, so review runs beside the original.

Actuaries back on advice

Routine document review handed off. The judgement stays with the team.

We are not the only ones saying this.

Themes from an industry roundtable on AI in pensions convened by The Pensions Regulator in July 2026, which we took part in alongside trustees, administrators, consultants and lawyers. The Chatham House Rule applied, so nothing here is attributed to any individual or firm.

01

Data quality was the barrier

Across every breakout room, the constraint on AI in pensions was not model capability and it was not regulation. It was data. The consensus was to cleanse the data first, and to use AI to do it.

02

Provenance was non-negotiable

Trustees have to be able to prove every answer back to source, for clients, sponsors and beneficiaries alike. An audit trail was treated as a precondition of using AI at all.

03

Human oversight is crucial

The room was clear: augmentation over uncontrolled autonomy. What is needed is ai with controls, tools that speed up workflows but keep the decision making in human hands.

We built Sourcestone on these three priniciples before that room ever met.

Get started

Run it on one of your own schemes.

Point us at one scheme's documents and see the referenced output on your own rules and your own template. Read access to one folder, revoked whenever you like.

  • One scheme, your documents, your template
  • Output you can review clause by clause against source
  • Run by an actuary, not a sales engineer