The hype and the honest evidence
Two things are true at once. Large language models are the most significant productivity tool the consulting industry has adopted in a generation. And most of what firms have been told about AI in bid writing is marketing, not measurement.
The MIT Sloan Management Review has been tracking generative AI adoption in professional services since 2023. The pattern is consistent across their case work: adoption is high, measurement is thin, and the biggest gains are showing up in narrow, structured tasks, not in end-to-end drafting. That should sound very familiar to any bid manager who has watched a "one-click proposal" demo.
Where AI is genuinely lifting productivity
1. Compliance matrix extraction. Reading a Terms of Reference PDF and producing a first-pass compliance matrix used to be a two-day analyst task. With a well-prompted LLM constrained to structured output, that becomes minutes. The gain is real because the task is deterministic and verifiable: every requirement either exists in the ToR or it does not.
2. CV tailoring. Rewriting a consultant's CV to align with the evaluation criteria of a specific procurement is a task humans do slowly and unevenly. LLMs do it in seconds. Evaluators do not care whether the CV was rewritten by a human or a machine. They care whether the CV addresses the evaluation criteria.
3. First-pass reviewer feedback. Using an LLM to flag weak win themes, missing evidence, and compliance gaps before a human reviewer picks up the draft consistently reduces the number of full review cycles per bid. This is not magic; it is a checklist run at machine speed.
Where AI is quietly hurting bids
1. Freeform methodology sections without retrieval grounding. LLMs hallucinate references. On donor and government procurements, a bid that cites a policy that does not exist is often flagged and remembered.
2. Answering donor-specific questions from base model knowledge. Small factual errors on donor policy are disproportionately damaging. Grounding every answer in a firm-controlled knowledge library, using retrieval-augmented generation, is not optional.
3. Tone regression. Bids drafted end-to-end by LLMs read as bids drafted end-to-end by LLMs. Evaluators notice. Human voice, opinion, and specificity remain competitive differentiators.
The three-line policy every bid team needs
- Use AI for structured extraction and first drafts of repeatable sections. Do not use it as a substitute for expert judgement.
- Every AI-generated statement must cite a source in the firm's knowledge base. No exceptions.
- Track productivity per stage, not per bid. Whole-bid productivity metrics obscure where the gains actually live.
The honest closing
AI is not going to replace bid managers. Bid managers who use AI are going to replace bid managers who do not. The World Economic Forum's Future of Jobs Report 2025 makes this same point across white-collar work more broadly, and consultancy is no exception.
The firms measuring their AI use are pulling away. The firms treating it as a marketing story are quietly losing ground.