Using past proposals with AI: what engineering and consulting firms should know

What can AI do with past proposals?#

AI can use past proposals to draft first versions of approach and project understanding sections, find the most relevant project descriptions for a new pursuit, adapt resumes to a scope and check a draft against the RFP's evaluation criteria. The firm's archive is the raw material; the tool is only as good as the proposals it draws from.

Proposal teams at engineering and consulting firms already reuse content by hand, usually by copying from the last similar submission. AI tools make that reuse faster and broader, which also means they spread whatever sits in the archive: an outdated fee, a former employee's resume, a teaming partner's text or a client detail that should never have left the original pursuit.

For consulting firms the archive often reaches beyond proposals to statements of qualifications, interview presentations and client capability decks. The same checks apply to each, and interview materials in particular tend to hold client insights that were shared in confidence.

Proposal components, AI uses and confidentiality checks#

Proposal components differ in how safely they can be reused. The table shows the typical AI use for each and the check to run before the content enters a library or a tool.

Building a proposal library AI can use#

A proposal library AI can use starts with fewer, better documents. Volume helps less than knowing which version was submitted and how the pursuit ended.

The outcome field is the step firms most often skip and the one that matters most. Without it, a tool cannot tell a winning approach from a losing one, and drafts drift toward whatever language appears most often in the archive.

• 1Collect the final submitted version of each proposal from the marketing share, the CRM or Deltek, not working drafts.
• 2Attach the outcome, such as won, lost, cancelled or no decision, with debrief notes where they exist.
• 3Tag market, service line, client type and contract vehicle so retrieval finds the right precedent.
• 4Move resumes, fees and teaming content into restricted folders.
• 5Retire stale content, such as discontinued services and departed staff.
• 6Set access controls that match how sensitive each folder is.

Questions to ask an AI proposal tool vendor#

Questions for an AI proposal tool vendor should be asked in writing before any proposal is uploaded. The answers belong in the contract, not only in a sales conversation.

If a vendor cannot answer these questions clearly, limit the pilot to content with no client details, such as firm boilerplate and service descriptions, until it can.

When proposals could also be licensed#

Proposals with outcomes can also be licensable records, because they show how experts frame a problem, structure a scope and argue for an approach, which is the kind of reasoning AI developers look for. That creates a second, separate question: whether some de-identified proposal records could be licensed.

Licensing calls for a stricter review than internal use. Client names, sites, fees, staff names and partner content are removed or excluded, client RFP text stays out, and proposals to public agencies are checked against public records rules and agency terms. The firm decides what is in scope, and each proposal set is reviewed on its own.

Keep the two decisions apart. Rolling out a drafting tool should not quietly settle whether the archive is licensed, and a licensing review should not hold up the firm's internal use of its own content.

Common mistakes with AI proposal libraries#

The most common mistakes with AI proposal libraries come from treating the archive as clean when it was built for a different purpose: getting one submission out the door.

• 1Loading every pursuit folder, drafts included, so the tool learns from language that was never submitted.
• 2Leaving fee spreadsheets in the same folders as narrative text.
• 3Indexing partner resumes and project sheets the firm has no right to reuse.
• 4Letting the tool cite projects a client asked the firm not to publicize.
• 5Skipping a human review of every AI-drafted section before submission.

Illustrative: an engineering and consulting firm cleans up its proposal archive#

Illustrative: a fictional water resources engineering and consulting firm keeps proposals in a marketing share, with pursuit tracking in its CRM. Each pursuit folder holds several drafts, the RFP, partner resumes and a fee spreadsheet, and nobody can say quickly which file was actually submitted.

The marketing lead pulls only final submitted proposals, matches each to the CRM outcome, moves fees and partner content into restricted folders and drops resumes of departed staff. The firm then pilots an AI drafting tool on the cleaned library after the vendor confirms in writing that it does not train on customer content.

Separately, the managing principal asks whether a de-identified set of won proposals and debriefs could be licensed, and sends that question through a rights review instead of deciding it inside the tool rollout.

How SourceX treats proposal archives#

SourceX treats a proposal archive as one record family among many in a firm's inventory. Under the SourceX Enterprise Data Value Framework, outcomes and debriefs add human-generated signal and domain expertise, while fees, names and partner content add privacy burden and preparation cost.

If a firm proceeds, the SourceX five-step transaction, Supply, Rights, Preparation, Approval and Delivery, keeps scope decisions with the firm at each step, and the SourceX Evidence Packet records which proposals were included, on what rights and with which details removed.