The most operationally significant recent development. Brightwell replaced Capita following competitive tender, completing the transition in 11 months — an ambitious timeline for migrating 117,000 members. The transition included deploying IntelliPen, launching a new personalised online portal, and establishing qualified pension administrator call handling from Chesterfield.

Growth pipeline: CEO Morten Nilsson has been explicit that Brightwell is “not a volume game” — the firm targets a select group of large schemes. But each new scheme introduces distinct benefit structures, legacy data formats and member communication requirements. The operational cost of growth scales with complexity, not just member count.

Pension Schemes Act 2026: New surplus extraction rules enabling sponsors to access surplus. Value for money framework imposing structured governance and reporting obligations. Enhanced administration guidance from TPR. Each creates documentation, calculation and communication demands across every managed scheme.

DB run-on: Brightwell’s own research with mallowstreet found 70% of large DB schemes now target run-on rather than buy-out (collapsed to 4%). Over 40% of employers intend to share surplus with members. This creates longer administration horizons and new demands for predictive modelling of member behaviour — retirement timing, transfer propensity, mortality trends, cashflow forecasting.

Cloud infrastructure: c. £6k+ p.a. Baseline assumes standard business-hours operation in a single region.

Commercial LLM API costs: c. £6k+ p.a. Highly volume-dependent; private LLM reduces this significantly.

Private LLM hosting (if deployed): c. £12k+ p.a. Recommended for Project 3 at scale; eliminates API variability. Infrastructure will be right-sized after a period of beta testing and live running, driven by volume of emails and the complexity and number of attachments processed.

Platform licensing: Driven by the number of deployed instances, number of company affiliates, and number of active users. Scoped during Discovery.

Aivantor support retainer (optional): c. £18k+ p.a. New scheme onboarding, tuning and escalation support.

High / High — P5 pipeline not yet production-proven. The migration pipeline is currently at prototype stage. This project includes additional development to take it to production grade. PoC and Decision Gate are critical. Allocate 20% additional contingency. Do not commence until at least one earlier project is delivered.

Medium / High — Target admin system integration complexity not yet known. Assessed during each project’s Discovery. API abstraction layer reduces but does not eliminate risk.

Medium / Medium — Source document quality inflates P4 effort. Historical or poorly scanned documents require more tuning iterations. PoC tests against a mixed sample of modern and historical documents to quantify risk before development commitment.

Low / High — Supplier due diligence not satisfied, stalling engagement. Schedule commercial credentials session with Simon early and in parallel with first Discovery so neither workstream blocks the other.

1. Execute NDA and agree which project(s) to initiate first.

2. Schedule Discovery — Projects 1 and 2 recommended as the lowest-risk starting point. Can run concurrently.

3. Provision shared infrastructure in parallel with the first Discovery.

4. Confirm routing list and Outlook screenshots with James Pearson ahead of Project 3 Discovery.

5. Schedule supplier due diligence session with Simon covering Aivantor’s financials, client references and organisational credentials — in parallel, not blocking project initiation.

6. Agree rate card for incremental items: additional schemes, new mailboxes, future migrations — to govern ongoing work beyond initial project scope.

Brightwell’s challenges are not independent. Each new scheme win triggers correspondence volumes, a data migration exercise and a document intelligence requirement simultaneously. Solving these with separate tools creates integration overhead, duplicated security configuration and siloed knowledge.

The aivantor-X platform provides a single orchestration layer managing LLM routing, security policy enforcement, audit logging and knowledge persistence. Each service line — email triage, data migration, document intelligence — operates as a dedicated agent pipeline within shared infrastructure. Security policies, data governance rules and knowledge base content are defined once and applied consistently.

Two challenges converge here. Graham and James identified that Copilot operates at individual user level and cannot support operational, multi-user processing across shared mailboxes. Separately, there is legitimate concern about open-ended per-token LLM pricing for high-volume use cases.

Private LLM deployment resolves both: predictable, fixed costs with no uncapped per-token exposure; operational-scale processing across any number of mailboxes without per-user licensing; and complete data sovereignty with 382,000+ member records never leaving Brightwell’s controlled environment.

Graham and James raised inconsistent handling of sensitive categories — bereavements, complaints, retirement requests — where tone, priority and procedural accuracy all matter. Template-based responses exist but are applied inconsistently. On the data migration side, Wojciech described 886 candidate fields requiring expert judgement to map correctly.

Human-in-the-loop is the default across all service lines. No email response is sent, no field mapping confirmed, no calculation generated without explicit human approval during the initial phase. Brightwell can progressively enable automatic processing for high-confidence operations while reserving human approval for nuanced or ambiguous cases.

Brightwell owns a technology company. IntelliPen is mature and in production across BTPS, MPS and global software clients. Intelli-ACT launched in February 2026. These are genuine products with a Bristol-based development team. Any technology partner must respect the existing product roadmap.

But Intelli-ACT references “the powerful data infrastructure that underpins modern AI solutions” while the system itself is an actuarial calculation engine. No AI is deployed for member correspondence, document processing or knowledge management — the operational areas where Graham, James, Andy and Wojciech have identified the sharpest pain.

Simon rightly identified supplier due diligence as a prerequisite — financial standing, client references, organisational depth and delivery track record. And Brightwell needs to establish commercial viability before committing time and resource to further evaluation.

Aivantor’s engagement model is built around client independence. Implementation is a defined engagement with clear scope, milestones and handover. Involvement scales down as the platform is embedded and the team is trained. Data ownership remains entirely with Brightwell.

Wojciech identified the absence of a structured knowledge base from onboarding outputs. The intelligence extracted during migration — scheme rules, calculation logic, data structures — is not captured in a queryable form. Each new scheme repeats the same manual effort with no institutional memory.

Template learning is built across all service lines. Confirmed field mappings are saved as supplier templates — the model biases towards historically confirmed choices. FAQ categorisation reduces repetitive email processing. Extracted scheme rules and calculations load into Aiva, creating a growing, queryable knowledge base that supports ongoing administration, training and audit.

LLM Routing: configurable per deployment and service line. Claude, GPT, Gemini and private LLM fallback.

DLP Policy Engine: block, mask or warn. Policies configurable per data category and scheme.

Context Window Management: proprietary chunking for large documents exceeding standard LLM context limits.

Role-Based Access: per user, per scheme, per service line. Full audit logging.

M365 Connectors: Outlook, SharePoint, Teams. API abstraction for CRM and IntelliPen.

Mailbox Monitor — Monitors shared mailboxes at scale without per-user licensing. Full correspondence estate across BTPS, EEPS and MPS from a single deployment.

Sentiment & Classification — Category classification against Brightwell’s configurable taxonomy. Bereavement, retirement quote, complaint, transfer, address change and scheme-specific categories.

Attachment Processing — Death certificates, identity proofs, signed forms automatically extracted, classified and associated with the correct member record.

Response Drafting — Suggested responses from scheme-specific and org-wide templates. No response sent without approval during human-in-the-loop phase.

Agentic Workflow Engine — Available now. Configurable post-classification actions: record updates, workflow triggers, task creation, specialist escalation. Autonomy configurable per category and per scheme.

MI & Reporting — Volumes, response times, category distributions, sentiment trends surfaced through configurable dashboards.

Purpose: Validate assumptions, define scope, identify risks and produce a Discovery Report with an initial PoC scope — before any build commitment.

Initial Discovery is a fixed-fee engagement of £14,000 covering the first project(s) to commence. Each subsequent project includes its own Discovery stage within its project estimate.

Activities:

• Requirements workshops with named stakeholders
• Existing systems, data and integration assessment
• Document and data inventory (format, condition, volume)
• Risk identification and dependency mapping
• Output format and acceptance criteria definition
• PoC scope agreement

Outputs:

• Discovery Report
• Agreed PoC scope and success criteria
• Preliminary risk register
• Revised effort estimate for PoC stage

A working prototype using real (anonymised) Brightwell data, validating the technical approach and surfacing unknowns before full commitment to build.

A formal review milestone. Aivantor presents PoC findings, a revised development estimate, and a recommended approach. Brightwell decides: proceed, pause, or cancel. No development spend is committed until this gate is passed.

Full build, integration and testing against agreed requirements. Estimates shown are pre-gate and will be refined at the Decision Gate. Blended rate: £900/day.

Environment preparation, UAT completion, cutover planning, data migration (where applicable), and go-live support.

Two weeks of dedicated post-go-live support: proactive monitoring, rapid issue resolution, and stakeholder confidence building before handover to BAU.

Objective: Monitor Brightwell’s operational mailboxes; automatically classify inbound emails by type and sentiment; extract and classify attachments; draft suggested responses from approved templates; route to the correct team — with human approval required before any action is taken.

Stakeholders: James Pearson, Graham Coleman

Discovery (10–15 days): Mailbox inventory; email category taxonomy; routing rules definition (using James Pearson’s routing list and Outlook screenshots); template library scoping; volume profiling; Outlook/Exchange integration assessment.

Proof of Concept (8–12 days): Live monitoring on a single mailbox; classification of real inbound emails; attachment extraction demonstration; sample response drafting.

✓ Decision Gate: PoC review; revised development estimate; go/no-go

Development (30–48 days): Full Outlook/Exchange integration; classification and sentiment engine; attachment extraction and classification; response template library build; human review and approval workflow; routing rules configuration; volume and load testing.

Transition (5–8 days): UAT completion; routing rules sign-off; go-live cutover.

HyperCare (5–8 days): 2 weeks post go-live: monitoring, classification accuracy review, rapid routing adjustments.

Project Management (6–10 days): Planning, reporting, risk, stakeholder management.

Total: 64–101 days