Case studies / Professional services
Case studyA knowledge assistant for proposals
A consultancy with thousands of past proposals built a permission-aware knowledge assistant over approved content, drafting proposals faster with consistent content and visible sources.
Overview
An engineering and environmental consultancy in the United Kingdom bidding for infrastructure, energy and public sector work, with specialist teams across several offices.
Scale. About 12,000 documents: past proposals, CVs, case studies, method statements and policies, spread across shared drives and the document system.
The challenge
When a tender arrived, bid teams went looking for the last proposal that resembled it. They searched shared drives, asked colleagues in other offices and copied sections from documents of uncertain age, then spent hours checking whether the content was still true.
Consultants searched old proposals for content. The archive held thousands of documents, many of them near-duplicates with different edits, and file names gave little clue which version had been submitted or won.
Content drifted. Descriptions of the firm's methods, accreditations and project experience varied between proposals, and out-of-date statements occasionally reached clients.
Some material was sensitive. Pricing, client-confidential details and personal information in CVs could not be shared freely across the firm, which made a simple search over everything inappropriate.
The firm had trialled a general-purpose chat tool. It wrote fluent text but could not say where its content came from, and bid managers would not use material they could not verify.
Experienced bid writers knew where the best material was, but that knowledge sat with a few people. When they were busy or on leave, newer consultants spent days hunting for content or wrote it again from scratch.
The solution
Approved content is indexed from the document system, including proposals, case studies, method statements, CVs and policies. Content owners are responsible for keeping their material current, and documents past their review date are flagged.
Access controls follow the document system's permissions and the user's identity. Pricing and client-confidential material are only used in answers for people who already have access to them.
Consultants ask questions or request drafts in plain language, for example a summary of the firm's experience with flood defence schemes, and receive a draft with citations linking to each source document and section.
Feedback buttons let users mark answers as helpful or wrong and flag outdated sources, which go to content owners for review.
The assistant is available in the firm's chat tool and alongside the CRM, so bid teams can pull in relevant case studies and CVs for the client and sector they are bidding to.
Usage analytics show which questions are asked most, which content is used and where gaps exist, helping the bid team decide what new material to write.
The engagement
- Five months
- Four specialists
- Selection and clean-up of approved content, design and build of a knowledge assistant with access controls, source citations, feedback and usage analytics, integration with the document system, identity, CRM, chat and analytics, a one-team pilot and extension to other teams.
Integrations
- Document system
- Identity
- CRM
- Chat
- Analytics
How we delivered it
- 01
Select approved content
Five weeks
With the bid team we selected the content the assistant would use: submitted and winning proposals, current case studies, approved method statements, CVs and policies. Duplicates and outdated material were removed or archived, and each document was tagged with an owner and review date.
- 02
Build the assistant with permissions
Seven weeks
We built the assistant over the approved content, carrying each document's permissions from the document system so that users only see answers drawn from content they are allowed to read. Every answer cites its sources.
- 03
Pilot with one team
Four weeks
The infrastructure bid team used the assistant on live tenders. Their feedback on answers and sources improved the content set as much as the assistant itself.
- 04
Extend to other teams
Four weeks
The assistant was extended to the energy and public sector teams, each adding and cleaning its own content before going live.
Results
- Faster proposal drafting
- Consistent, current content across bids
- Sources cited in every answer
- A smaller, owned content library
Proposals are drafted faster. Consultants start from a cited draft built from approved content rather than a blank page or an old proposal, and spend their time tailoring it to the client.
Content stays consistent. Method statements, accreditations and experience descriptions now come from one approved source, and outdated statements are flagged rather than repeated.
Sources are visible. Every answer shows where it came from, so bid managers can verify content quickly and trust what goes into a submission.
Content hygiene mattered more than the model. The biggest improvements in answer quality came from removing duplicates, retiring outdated documents and assigning owners, not from changing the underlying model.
The bid team now maintains a smaller, better content library, and gaps identified through usage analytics have led to new case studies and method statements.
Newer consultants have become productive on bids sooner. They can find the firm's best material without knowing who wrote it or where it was saved, and experienced bid writers are interrupted less often.
We spent more time deleting old proposals than configuring the assistant, and that was the right way round.
What made the difference
Content hygiene mattered more than the model.
