TL;DR: Emma extracts transfer values, GARs, fund holdings, and charges from provider LOA packs and populates your firm's summary template, cutting review time by up to 80%. Initial template configuration is handled as a one-time onboarding step, guided by AdvisoryAI's ex-paraplanner team so your firm's document structure is built in from the start. LOA data feeds into suitability reports alongside meeting notes, fact-finds, and cashflow modelling. Colin checks the draft against FCA Consumer Duty and COBS requirements, on any suitability report regardless of origin.

The biggest delay in your advice process is not the provider's turnaround time. It is the manual data extraction that happens after the LOA pack arrives in your inbox. LOA pack review is one component of the broader documentation burden facing UK advice firms: 71.9% of firms spend 1-7 hours producing a single suitability report, with 43.3% of advisers reporting that paperwork and admin reduces time devoted to advice itself. A Letter of Authority (LOA) is a document your client signs to authorise you to request their policy information from a ceding provider.

When paraplanners talk about an LOA pack, they mean the bundle of documents a provider returns in response to that authority, arriving from Aviva, Aegon, Standard Life, or another provider in whatever format that provider has used for the past decade, and containing the policy details, transfer values, charges, fund holdings, guarantees, and beneficiary nominations that underpin the suitability report.

Providers send LOA information in varied formats, letters, emails, and PDFs, from which key data must be extracted manually. AI can change that equation by shifting the paraplanner from author to editor: instead of hunting for data, you verify a structured draft.

LOA Pack Contents and Review Challenges

LOA pack review typically sits at the start of suitability report workflows, and the manual burden compounds across every case. Understanding which fields create the most friction explains why AI extraction produces such a measurable time reduction.

Policy Details and Transfer Values

LOA packs contain the policy number, plan type, current transfer value, and effective date of that valuation as core data points. These fields appear in different positions across different providers' documents and sometimes differ between a covering letter and the schedule pages. You face the highest transcription risk here because you are moving numbers from a PDF into your back office (Intelliflo, Plannr, Curo, or Xplan) while also reviewing the context around them.

LOA Pack Charges and Holdings

Providers often place ongoing charge figures (OCF), annual management charges, platform fees, and any exit charges in appendices or footnotes. Fund holdings, including ISIN codes, units held, and percentage allocations, may appear on separate schedule pages from the main transfer value. If you review manually, you must cross-reference both sections to build an accurate picture of what the client currently holds and what it costs them.

Quickly Identifying Guarantees and Nominees

Protected tax-free cash entitlements, guaranteed annuity rates (GARs), guaranteed minimum pensions (GMPs), and Safeguarded Benefits require specific identification because they affect the suitability of any transfer recommendation. These fields carry regulatory significance under FCA rules. You must also capture beneficiary nominations and expression of wishes forms accurately, because these affect the client's estate planning position.

Navigating Varied LOA Pack Designs

The LOA process across the industry shows considerable inconsistency. Beyond timing delays, the document formats themselves vary by provider and sometimes by product vintage within the same provider. Pension policies opened years apart from the same provider can arrive in different layouts. AI extraction can handle that variation systematically by reading for financial entity types rather than fixed field positions.

How AI Extracts Structured Data from LOA Documents

Emma combines Optical Character Recognition (OCR) and Natural Language Processing (NLP) to extract structured data from provider documents. Emma uses OCR to convert images into machine-readable text, which matters because many providers still send LOA packs as computer-generated scanned documents rather than native digital PDFs, making the underlying data inaccessible to text-based processing without it. Its NLP then classifies and extracts structured data from that text, identifying named financial entities such as transfer values, plan numbers, and charge figures rather than reading words in sequence.

Emma uses Retrieval-Augmented Generation (RAG), which means the model references your firm's specific document templates and relevant FCA rules before drafting the summary rather than generating output in isolation. Emma by AdvisoryAI is trained on thousands of sample reports by ex-advisers and paraplanners and configured against the firm's own templates, and Emma cites each extracted line back to its source document, whether that is the fact-find, illustration, LOA, or back-office note, so every field is traceable. Watch this demo of Emma summarising a pension LOA pack to see structured extraction in action across a real provider document.

AdvisoryAI's Standard LOA Pack Summary Template

One of Emma's key features is automatic flagging of missing information. When provider documents omit required fields, Emma marks them as "Not provided" rather than leaving blanks, so you know exactly which information to request from the provider without re-reading the entire pack.

Below is AdvisoryAI's standard pension LOA pack summary template showing this flagging in action.

Missing information tracker: Where provider documents omit required fields, Emma flags these gaps. You can use a template letter to request outstanding details:

Subject: Missing Pension Information for [Client Name], policy number [XXXXX]

Dear [Provider],

I am writing regarding [Client Name]'s pension (Plan Number: [XXXXX]). Following our recent information request, we require the following additional details to complete our analysis:

[List missing fields]

This information is essential for us to provide appropriate advice to our client. Please provide these details at your earliest convenience.

Minimising Errors in Scanned LOA Data

OCR accuracy can improve when machine learning and NLP work together because the model checks whether an extracted value makes contextual sense. A transfer value of £4 on a pension policy is implausible, whereas £400,000 is not.

Identifying Critical LOA Data Fields

The AI maps extracted text to a structured field set covering policy details, transfer values, charges, fund holdings, Safeguarded Benefits indicators, and beneficiary nominations. This field-level mapping produces a structured draft rather than a raw transcription.

AI Reading Handwritten LOA Notes

Some providers include handwritten annotations or marginalia on older policy schedules. Handwritten OCR accuracy can depend on handwriting clarity and image resolution. Rather than inserting a potentially incorrect value silently, the AI flags lower-confidence extractions from handwritten sections as items for manual verification in the review stage.

Fast LOA Pack Document Review

Generating ten LOA pack summaries can take under an hour with Emma. The reason is the shift from sequential reading to parallel verification: you review a completed draft rather than building one from scratch.

How AI Masters Diverse LOA Pack Formats

One of the most common objections to AI LOA extraction is that provider formats are too variable for any automated tool to handle reliably. The practical answer is that AI can extract by financial entity type and contextual position, not by fixed page layout.

This means it adapts to format variation rather than depending on a rigid document that breaks when a provider updates their design. The Emma document generation demo shows extraction across varied document types, and the AdvisoryAI platform walkthrough shows how Emma fits within the full advice workflow.

Aviva: Parsing Pension and Bond Types

Aviva's pension and investment bond documents differ significantly by product type and vintage, with pension policies carrying Safeguarded Benefits flags and scheme pension details that investment bonds do not. The AI identifies the product type first, then applies the relevant field set for that product category, so a pension extraction does not attempt to locate fields that only exist in a bond document.

Aegon and Standard Life: Platform vs Legacy Formats

Aegon's platform products may present structured data more consistently than legacy Aegon or Scottish Equitable policies from pre-platform years, which often arrive as scanned physical schedules with less consistent field placement. The AI uses contextual markers, including headers, section titles, and table structures, to locate relevant data rather than relying on fixed page positions, so it handles both product generations without separate configuration.

Standard Life documents include internal fund codes alongside fund names. The AI can extract both identifiers, allowing your to display whichever your firm prefers without requiring manual cross-referencing to an internal fund list.

Royal London: Guaranteed Annuity Rates

Royal London plans, particularly older with-profits and unit-linked pensions, may be a source of GARs and other Safeguarded Benefits. The AI can flag any GAR or Safeguarded Benefits indicator as a high-priority review field, directing your attention to these items before you accept the summary, in line with the suitability requirements that apply to transfer recommendations.

Scottish Widows and Prudential: Complex Legacy Structures

Scottish Widows with-profits plans include terminal bonus rates, market value adjusters (MVAs), and guaranteed surrender values. Emma extracts these fields with source citations, so you confirm against the original document in a single click rather than searching the pack again.

Prudential's legacy plans, particularly PruFund and older unit-linked structures, often contain benefit components across multiple sections within LOA documentation. Emma can stitch related fields from different document sections into a single coherent summary and flags where information came from more than one source, so you know exactly where to verify.

AI Flags for Review: Ensuring LOA Accuracy

AI extraction is not infallible, and the firms we've seen use it most effectively treat it as a first-pass draft rather than a final document. The value is in the fields that extract correctly, which free you to focus on the items that need verification rather than the full document.

AI Alerts for Missing LOA Fields

Emma and Colin work together at the end of the extraction stage. Colin breaks compliance into clear categories showing exactly what passes and what needs attention. Where a required field, such as a Safeguarded Benefits declaration, is absent from the extracted output, Colin flags it as a missing item with a specific remediation prompt rather than leaving a blank field for you to discover during review.

Verifying AI's Low-Confidence Fields

Where OCR confidence on a specific value falls below the system threshold, for example on a degraded scan or a handwritten annotation, the field may appear with a low-confidence marker in the draft. Colin's built-in highlighting can take you directly to the problem area in the source document, so the verification step is targeted rather than requiring a full re-read of the entire pack.

Handling Unfamiliar LOA Formats

When the AI encounters a document structure it has not processed before, such as a niche provider or an unusually formatted product schedule, it can extract the fields it can identify with confidence and flag the remainder for manual input. You complete those fields directly in the draft, which may feed back into the model's learning for future documents of the same type.

Pension Transfer Value Discrepancies

If the transfer value in the main schedule differs from a figure quoted in the covering letter, providers sometimes send interim and final valuations in the same pack. In these cases, you confirm which value applies and document the reason, maintaining a clean audit trail.

Human-in-the-Loop Checkpoints for FCA Compliance

AI extraction does not remove professional judgment from the LOA review process. It restructures where that judgment applies, moving it from data entry to data verification. This is the author-to-editor shift applied to document review: you confirm and approve rather than write from scratch.

Reviewing AI LOA Pack Summaries

Every Emma-generated LOA summary reaches you as a draft requiring review and approval before it populates your fact-find fields, including personal information, investment details, and employment details, or feeds into the suitability report. Report review shifts from painstaking line-by-line work to targeted spot-checks, where clicking a citation shows the source document and confirms accuracy. The professional sign-off remains your responsibility throughout.

Verifying AI LOA Data Fields

The structured field format can make verification faster because you are confirming named values against a source document rather than searching for them. Transfer value confirmed, GAR flag confirmed, charges confirmed: the verification step runs through known fields in sequence rather than requiring an open-ended read of the original pack.

Validating AI LOA Summaries for Audit

FCA supervision requires a clear audit trail of the client relationship and the decision-making process that led to an advice recommendation. Emma's citation of every extracted field back to its source document means the audit trail is built into the summary itself, not reconstructed at file review. Colin can store a time-stamped record of each compliance check, providing documentation for any future FCA supervision visit or s166 examination.

Consumer Duty Evidence Requirements

Consumer Duty evidence requirements focus on demonstrating good customer outcomes, not just documented processes. Accurate LOA pack summaries can contribute directly to that evidence base by ensuring the suitability assessment rests on confirmed policy data rather than manually transcribed figures that may contain errors introduced under time pressure. For a practical look at how Colin checks suitability documents against Consumer Duty and COBS requirements, read the introducing Colin post on the AdvisoryAI blog.

Reduce LOA Review Time: The AI Workflow Impact

Finsource Partners reported an 80% reduction in LOA pack review time using Emma, with the remaining time concentrated in the human verification step rather than extraction itself.

Before AI: Manual LOA Pack Processing

The manual process follows a predictable sequence. You open the provider PDF, read through to locate the transfer value, note the policy number, search for any guarantees, cross-reference the charges table, list the fund holdings, record the beneficiary nominations, and then manually key all of those fields into the back office and the draft summary . Each step takes time and each is a potential source of transcription error. Missed fields typically surface only when the suitability report enters review, requiring a return to the source document and a correction cycle that adds further delay.

AI LOA Summaries

With AI extraction, you upload the provider PDF once. Emma processes the provider document, extracts named fields including transfer values, charges, fund holdings, and Safeguarded Benefits indicators, and populates your firm's summary as a draft ready for your review, with each field cited back to its source and any missing or low-confidence items flagged for your attention. That extracted data then feeds directly into the suitability report alongside meeting notes from Evie, fact-find data from your back office, cashflow modelling outputs, and risk profile assessments, so LOA extraction is one step in a connected workflow rather than a standalone task.

Manual LOA Data Extraction

Manual extraction remains the default for firms without AI tooling, and the cumulative cost is substantial. Across a multi-adviser firm, LOA packs accumulate quickly, representing a substantial paraplanner workload that compounds across every active case.

Table 1: AI LOA Tools Compared

Table 2: Levels of LOA Processing in UK Financial Advice

This framework illustrates the practical progression from manual extraction to conversational query, helping firms identify where they currently sit and what the next step looks like.

Must-Knows for AI LOA Pack Summaries

These are the questions that matter most for UK FCA-regulated firms evaluating AI extraction tooling before committing to any platform.

Verifying AI LOA Summary Accuracy

When evaluating any extraction tool, we recommend running it against your actual provider documents, not a curated demo set. Upload three or four real LOA packs from your most common providers and verify the extracted output field by field against the source documents. Specifically check:

• Transfer values match source documents exactly

• All Safeguarded Benefits flags are captured

• Fund holdings include correct ISINs and allocations

• Charges breakdown reflects all fees listed in the pack

Inconsistent LOA Data Input

The quality of extracted output can depend on the quality of the source document. Consider these practical steps before extraction:

• Request digital PDFs directly from providers where possible, as digital files generally produce better OCR results than low-resolution scans

• Flag providers that consistently send poor-quality documents so those cases receive additional verification attention

• Where a scan is degraded, Emma flags the affected fields rather than silently inserting a potentially incorrect value

Ensuring FCA Audit Trail for LOA

Every extracted field in Emma's output is cited back to its source document, creating the documentation trail required under Consumer Duty and COBS. Colin stores compliance records of each check, providing the audit evidence a file reviewer or FCA supervision team would expect to find in a well-maintained client file.

Back Office Integration for LOA Pack Summaries?

AdvisoryAI connects directly with Intelliflo, Plannr, Curo, and Xplan, pushing structured LOA outputs directly into fact-find fields including personal information, investment details, and employment details without manual re-entry.

Once LOA data is extracted and filed alongside meeting transcripts, suitability reports, and fact-find data, Atlas, AdvisoryAI's conversational interface, connects all of it into a single queryable dataset. Rather than returning to the LOA summary to check a transfer value or a GAR flag, you ask Atlas directly and receive a cited answer drawn from the relevant source document. Across a client book, that means you can query multiple LOA packs and prior meeting transcripts simultaneously, identifying patterns in charges, transfer values, or benefit structures across several clients without opening each file individually.

How Client LOA Data Is Secured

AdvisoryAI secures client LOA data through multiple layers:

• Client data stored with UK data residency confirmed

• Encryption applied to client data in transit and at rest (firms should request current security documentation from AdvisoryAI)

• Security certifications include Cyber Essentials certification and ISO 27001 (in progress), additional certifications are available on request from the AdvisoryAI team

• Firm-specific configuration, your client data is protected within your firm's environment

• Client data is stored on UK-based AWS servers and is not used to train models

• Anonymised data used for tone and training stays within your firm's configuration

• Changes made within the platform remain within your firm's environment

Firms with enterprise procurement requirements should request the current certification status directly from the AdvisoryAI team before proceeding. For a worked example of how Emma handles suitability letter generation using LOA-derived data alongside meeting notes, fact-finds, ceding information, cashflow modelling outputs, risk profile assessments, and illustrations, visit the AdvisoryAI website to request a demonstration.

Start a 14-day free trial of Emma, no credit card required, and run it against your firm's actual LOA packs from your highest-volume providers. Emma is priced at £299 per user per month on a monthly rolling agreement with no lock-in. A 30-day money-back guarantee applies, and annual plans carry a 10% discount. To see how Emma processes a complex LOA pack from a specific provider, request a demo.

FAQs

How Does Emma Integrate LOA Data into the Full Advice Workflow?

Emma extracts LOA data and pushes it to your back office (Intelliflo, Plannr, Curo, Xplan) where it populates fact-find fields including personal information, investment details, and employment details. That extracted data then feeds into suitability reports alongside meeting notes, cashflow modelling, and risk assessments.

Can Emma Work with Our Firm's Existing Suitability Report Templates?

Yes. Emma generates reports from your firm's own templates, not a standardised vendor format. The platform captures your advice style and tonality per firm, and off-the-shelf templates are fully customisable.

What AI Model Does Emma Use and How Was It Trained?

Emma was trained on thousands of sample reports by ex-advisers and paraplanners. The CTO holds an MIT Masters in AI/ML.

How Does AdvisoryAI Support Deployment Across Large Advice Networks or Consolidators?

AdvisoryAI serves UK financial advice firms, networks, consolidators, and investment management firms, including the majority of the UK consolidation market and several top-five IFAs. The platform supports bespoke templates per firm within a network deployment, allowing each acquired firm to retain its own document structure while sharing the same compliance infrastructure.

How Does Atlas Support Pre-Meeting Preparation and Investment Opportunity Identification?

Atlas is AdvisoryAI's conversational interface, operating across the full dataset your firm has built: meeting transcripts from Evie, suitability reports and LOA summaries from Emma, uploaded documents, and client data from your back office. Because LOA packs feed into that connected dataset, Atlas can surface patterns that would otherwise require manual cross-referencing across multiple files. Before a review meeting, Atlas can pull a client's vulnerability history, prior meeting context, and the key figures from their most recent LOA pack into a structured pre-meeting brief, so the adviser arrives informed rather than relying on memory or a manual file search. Across the client database, Atlas may identify clients whose current policy charges or transfer values indicate a conversation is warranted, supporting investment opportunity identification without requiring the adviser to run separate reports. No comparable tool currently offers this conversational query layer across the full advice documentation set.

Key Terms Glossary

Letter of Authority (LOA): A document signed by the client authorising their financial adviser to request policy information from a ceding provider. In UK practice, the "LOA pack" refers to the full document bundle the provider returns in response to that authority.

Safeguarded Benefits: Benefits in a defined benefit or legacy pension scheme that carry guarantees such as guaranteed annuity rates (GARs) or guaranteed minimum pensions (GMPs). Advisers must specifically identify and document these before recommending a transfer away from the arrangement.

Ceding scheme: The existing pension or investment arrangement a client is considering transferring away from. The ceding provider is the organisation currently holding the plan and returning the LOA pack in response to the authority.

Consumer Duty: The FCA's conduct framework requiring firms to evidence that clients receive good outcomes and fair value across the product or service lifecycle. For LOA packs, this means documentation must demonstrate that the full policy picture, including charges, guarantees, and transfer values, was accurately captured and used in the suitability assessment.