Enterprise Knowledge Assistant Giving Consultants Their Time Back
An 800-person strategy consulting firm was losing 30% of billable time to knowledge search — consultants manually trawling SharePoint, email, and Notion for past deliverables and methodologies. We built a RAG-powered knowledge assistant connected to all internal repositories, enabling natural language search with cited, source-linked answers. Average search time fell from 47 minutes to 9 minutes.
The firm advises Fortune 500 companies on strategy, operations, and digital transformation. Its differentiated value lies in institutional knowledge: 6 years of client deliverables, proprietary frameworks, competitive analyses, and methodology documentation. That knowledge was practically inaccessible — spread across 12,000 SharePoint folders, 3 million emails, and three generations of Notion workspaces.
Knowledge retrieval had become a hidden tax on productivity. Partners estimated that junior consultants spent 2–3 hours per day searching for past work. Senior consultants spent 45 minutes per day answering questions that had been answered before. A KM team of two people maintained a manual index that was always out of date. The firm had tried three previous KM tools; none had been adopted.
Consultants spending 30% of billable time on knowledge search — $3.2M in annual revenue impact
12,000 SharePoint folders, 3M+ emails, and three Notion workspaces with no unified search
KM team of 2 maintaining a manual index that was 6 months out of date
Three previous KM tool failures — adoption was the central challenge, not technology
No citation or attribution in search results — consultants couldn't verify the source or context of retrieved content
We built a hybrid RAG system using Claude 3 as the generation layer, Pinecone for vector storage, and a custom ingestion pipeline that processed SharePoint, email (with permission controls), and Notion into a unified knowledge base. The system returns answers with mandatory source citations — every response links to the originating document, slide, or email thread.
Knowledge audit: catalogued all 12,000 SharePoint folders, prioritised by access frequency and recency — reduced ingestion scope to highest-value 40%
Ingestion pipeline built for SharePoint, Notion, and email with metadata preservation and permission inheritance
Hybrid retrieval: dense embedding (text-embedding-3-large) + BM25 sparse retrieval + cross-encoder reranker
Claude 3 Opus as generation model — chosen for citation accuracy and instruction-following on source attribution
Permission-aware retrieval — consultants can only see documents their role entitles them to
Slack integration — consultants query the assistant in the tool they already use
Six months post-launch, 87% of the firm's 800 consultants were active users. Average knowledge search time fell from 47 minutes to 9 minutes. Partners reported that junior consultants were delivering higher-quality first drafts — because they were building on institutional knowledge rather than starting from scratch.
Knowledge retrieval time: 47 min → 9 min (81% reduction)
87% consultant adoption within 6 months
Estimated $2.8M annual revenue benefit from reclaimed billable time
Senior consultant interruptions for knowledge questions: down 73%
First-draft quality scores (internal peer review): up 24 percentile points
“It's like having the institutional knowledge of our entire firm available on demand. Junior consultants now produce work that used to take twice as long. The ROI was immediate and obvious.”
Caroline Hughes
Managing Partner, Strategy Consulting Firm
Mandatory citation in every response was the single feature that drove adoption — consultants trusted a system that showed its work
Permission-aware retrieval was non-negotiable for an organisation handling client-confidential data — it had to be correct from day one
Slack integration outperformed a standalone web app by 3:1 for daily active usage — meet users where they already are
Key Results
Services Engaged
Technology Stack
Facing a Similar Challenge?
Let's discuss your specific context and what results are realistic for your organisation.
Get in touch