How a Leading Financial Services Firm Transformed Knowledge Access with InsightBot

AWS Gen AI Competency

Amazon Connect SDP

Amazon Bedrock

InsightBot RAG

Financial Services

500 Employees

10,000 Documents

Multi-Corpus Search

Role-Based Access

Results at a Glance

Industry: Financial Services · Employees: 500 · Documents indexed: 10,000+ · Data sources: Multiple (5 systems) · Platform: InsightBot on AWS Bedrock

Executive Summary

A leading financial services firm with 500 employees was struggling with a problem that is almost universal in enterprise financial operations: critical knowledge locked across five disconnected document repositories, no way to search across them simultaneously, and analysts spending hours every day manually gathering information before they could begin any meaningful analysis. Decision-making was slow not because people lacked insight — but because finding the right information at the right moment was an exhausting, manual process. P3Fusion deployed InsightBot, its advanced enterprise RAG platform built on Amazon Bedrock, in 10 weeks — indexing 10,000 documents across all five source systems into a unified, intelligent knowledge layer. Employees can now ask a plain-language question and instantly receive a cited, accurate answer that draws from every relevant document in the firm's corpus — regardless of which system it lives in. Search time dropped by 80%. Decision-making that previously required assembling information from multiple sources over hours now happens in seconds. The firm's analysts describe it as the single biggest change to their daily workflow in a decade.

About the Customer

A 500-Person Financial Services Firm with a Knowledge Problem

The customer is a well-established financial services organisation operating across investment analysis, portfolio management, compliance, and client advisory functions. With 500 professionals spread across multiple departments, the firm manages a significant body of internal research, regulatory documentation, client records, investment policies, and operational procedures — accumulated over more than two decades of business.

Like most firms of this scale, knowledge was never centralised. Documents lived in SharePoint, shared drives, an internal wiki, a compliance document management system, and email archives. Each department had its own filing conventions. There was no single place to search. And with 10,000 documents in total — and more added every week — navigating that landscape manually had become genuinely unsustainable.

The Challenge

Information Was Everywhere. Answers Were Nowhere.

The firm's leadership had recognised for some time that their knowledge access problem was not a storage problem — it was a retrieval and connection problem. Documents existed. The right information was somewhere in those 10,000 files. But getting to it required knowing which system to search, how that system's search worked, and then manually reading through results to piece together an answer — often by asking colleagues who had previously worked with similar documents.

For analysts preparing investment reports or client briefings, this process was costing an average of 3 to 4 hours per working day. For compliance teams checking regulatory alignment across policies, it meant hours of manual cross-referencing. For new joiners, it meant weeks of onboarding before they could navigate the firm's knowledge effectively. And critically — for time-sensitive decisions, it meant delays that had real commercial consequences.

🔍

Fragmented search across 5 systems

Employees had to search SharePoint, the compliance DMS, shared drives, the intranet wiki, and email separately — with no unified view. Cross-system queries were impossible without manual effort.

⏱

3–4 hours lost per analyst per day

Knowledge gathering — not analysis — was consuming the majority of analyst time. Decisions that required multi-source insight were delayed by hours while information was assembled manually.

🔗

No connected analysis across documents

Answers existed across multiple documents simultaneously. But no tool could read across all of them, connect the relevant pieces, and surface a synthesised response. Analysts had to do this manually.

🔒

Access control complexity

Not all documents were appropriate for all roles. Compliance documents, client records, and investment theses required different access levels — making a unified search tool feel impossible without a sophisticated permissions layer.

Our analysts are some of the sharpest people in the business. But they were spending half their day being librarians — searching, chasing, and compiling — before the actual thinking could even begin. That was the problem we needed to solve.

— Chief Operating Officer, Financial Services Firm (name withheld)

Why InsightBot

A Platform Built for Exactly This Problem

The firm evaluated several options before selecting P3Fusion's InsightBot. Generic enterprise search tools could index documents — but they returned lists of results, not answers. Off-the-shelf AI chatbots could generate responses — but they hallucinated freely and had no access to proprietary internal documents. What the firm needed was something that combined both: a platform that knew the firm's actual documents, retrieved precisely from them, and delivered cited, trustworthy answers — with enterprise-grade access controls built in from day one.

InsightBot met every requirement. Built on Amazon Bedrock with pgvector for vector storage, it could ingest documents from all five of the firm's existing systems, enforce role-based access at the retrieval layer — not just the application layer — and deliver answers that cited the exact document and section they came from. Every answer was grounded. Every answer was auditable. And the access control model meant that a junior analyst asking a question would only ever receive information their role permitted them to access — automatically, without any configuration per query.

P3Fusion also demonstrated InsightBot on a sample of the firm's actual documents within the first two weeks of evaluation — returning accurate, cited answers from the firm's real compliance policies and investment research. That proof-of-concept on real data was the deciding factor.

The Solution

From Five Siloed Systems to One Intelligent Knowledge Layer — in 10 Weeks

P3Fusion deployed InsightBot across the firm's complete document estate in a structured 10-week engagement. The deployment was phased to avoid disruption to live operations — at no point during the rollout did employees lose access to any existing system.

Document Discovery & Access Mapping (Weeks 1–2)

P3Fusion catalogued all 10,000 documents across the five source systems, mapped existing access permissions to InsightBot's role-based retrieval layer, and ran a 1,000-document proof-of-concept on live firm data achieving 95%+ extraction accuracy. Sign-off from the COO and compliance team within Week 2.

✓ PoC approved by leadership

Multi-Corpus Ingestion Pipeline (Weeks 3–5)

InsightBot's connectors ingested documents from SharePoint, the compliance DMS, shared drives, the intranet wiki, and email archives — normalising formats (PDF, DOCX, XLSX, HTML) through a multi-modal parsing pipeline. Tables in financial reports were parsed as structured data. Charts and diagrams were captioned by a visual language model and made searchable. All 10,000 documents were indexed into pgvector with HNSW indexing in under 72 hours.

✓ Full corpus indexed, zero disruption

Access Control & Compliance Layer (Weeks 5–7)

Role-based access controls were implemented at the vector retrieval layer using row-level security on pgvector — ensuring that each user's query only surfaces documents their role permits. Four access tiers were configured: General Staff, Senior Analysts, Compliance Officers, and Executive Leadership. Tested across 200 user scenarios with zero access violations before go-live.

✓ Compliance team sign-off received

Hybrid Search, Guardrails & Connected Analysis (Weeks 6–8)

InsightBot's hybrid search — combining Amazon Bedrock semantic vector search with BM25 keyword matching — was configured for financial document retrieval. Amazon Bedrock Guardrails were tuned to the firm's compliance requirements: blocking out-of-scope regulatory topics, detecting and anonymising sensitive client PII, and flagging answers where retrieval confidence was below threshold. The connected analysis feature — drawing simultaneously from multiple documents to synthesise a single cohesive answer — was validated on investment research queries that previously required manual assembly from 3–5 documents.

✓ Guardrails approved by compliance

Phased Rollout & Training (Weeks 8–10)

InsightBot was released to pilot users in Week 8 (50 employees across three departments), achieving a 4.6/5 satisfaction score within the first week. Full rollout to all 500 employees completed in Week 10. A 90-minute training programme was sufficient — the natural-language interface required no technical knowledge. A dedicated InsightBot feedback channel collected user input that feeds the monthly quality improvement cycle.

✓ 500 users live · Week 10

The technical heart of InsightBot's advantage in this deployment was the connected analysis capability. When an analyst asks a question that touches on investment policy, a client's historical position, and the relevant regulatory guidance simultaneously, InsightBot retrieves from all three document types, synthesises a single grounded answer, and cites every source — in one response, in under three seconds. That capability — drawing across a multi-corpus knowledge base and presenting a connected insight rather than a list of documents — is what the firm's analysts describe as genuinely transformative.

Amazon BedrockTitan Embeddings V2Bedrock Guardrailspgvector · PostgreSQLAmazon ECSAmazon CognitoAmazon S3Amazon CloudWatch

Before & After

The Measurable Difference

Before InsightBot

3–4 hrs

Average time per analyst per day spent searching for information across 5 systems

~60%

Keyword search precision — 4 in 10 top results were irrelevant, requiring manual review

Hours

Time to assemble a connected analysis drawing from multiple document sources

Weeks

Onboarding time before new hires could navigate firm knowledge independently

→

After InsightBot

<30 min

Average time per analyst per day on knowledge search — 80% reduction

95%+

Answer accuracy across the full 10,000-document corpus — every answer cited and verifiable

<3 sec

End-to-end connected analysis across all five source systems simultaneously

2 days

Time for new hires to become independently productive with firm knowledge

Results

Decisions at the Speed of Business

80%

Reduction in information search time

3–4 hrs/day → under 30 min

95%+

Answer accuracy across 10,000 documents

Up from ~60% with keyword search

<3s

Connected analysis across all five systems

Previously took hours manually

Decision Speed

Same day

Investment and compliance decisions that previously required overnight information gathering now happen within the same working session — with full source citations for audit trail.

New Hire Onboarding

2 days → productive

New analysts reach independent knowledge proficiency in 2 days instead of several weeks. InsightBot effectively gives every new joiner instant access to the firm's accumulated institutional knowledge.

Compliance Confidence

100% cited answers

Every answer InsightBot delivers cites the exact document and section it came from. The compliance team can audit any decision's information trail in seconds — a capability that did not exist before.

User Adoption

4.6 / 5 satisfaction

Pilot satisfaction score in week one. Full 500-user rollout completed without resistance — the first enterprise-wide technology adoption in recent memory that required zero follow-up change management.

The outcome that the firm's leadership highlights most is not a single metric — it is the change in how decisions get made. Senior managers describe a shift from "let me come back to you on that" to genuine real-time discussion of investment positions, compliance questions, and operational queries — because the information to answer those questions is now available in the room, immediately, in every meeting.

InsightBot didn't just save us time searching. It changed what we can discuss in a meeting — because the answers are available the moment the question is asked. That's a fundamentally different way of working.

— Head of Investment Analysis, Financial Services Firm (name withheld)

Next Steps

Phase 2: Real-Time Market Intelligence & Automated Report Drafting

Building on the success of the initial deployment, the firm is now working with P3Fusion on Phase 2 of the InsightBot rollout. Phase 2 will extend the platform's knowledge base to include real-time market data feeds — allowing analysts to ask questions that combine the firm's internal research with live market context simultaneously. A second workstream will use Amazon Bedrock's generation capabilities to produce first-draft investment summaries and client briefings, reducing report preparation time from hours to minutes.

The firm's COO has described InsightBot as the foundational infrastructure for every AI initiative the firm plans to pursue over the next three years — the trusted knowledge layer that all future intelligence tools will build on top of.

P3Fusion

AWS Generative AI Competency Partner and Amazon Connect Service Delivery Partner. InsightBot is P3Fusion's flagship enterprise RAG platform — deployed across financial services, energy, and construction on Amazon Bedrock.

Gen AI Competency

Connect SDP

Bedrock

RAG

pgvector

ECS

Ready to transform how your organisation accesses knowledge? We run a proof-of-concept on your real documents in 2 weeks.

Construction · InsightBot

Global construction firm cuts project knowledge retrieval time by 75%

Energy · InsightBot · Amazon Connect

Energy firm gives offshore teams instant access to operational procedures via InsightBot

All studies