Proposal drafting with AI and what to check before sending

5 days ago
3 min read

Updated: 4 days ago

Introduction

Writing proposals is the least enjoyable and most repeated task in professional services and contracting. Eighty per cent of the document is scope, method, terms, credentials and structure that barely changes between clients, and the remaining twenty per cent is the part that wins or loses the work. Drafting help is genuinely useful here.

It is also the application with the most expensive failure mode, because a proposal is not marketing copy. It is an offer that can be accepted. A scope line that overstates what you will do, a timescale you cannot meet, or an omitted exclusion becomes a contractual obligation the moment the client signs, and the review step is what stands between the two.

1. Proposal drafting with AI and what to check starts with the scope section

This is the clause that binds you.

Every deliverable listed is something you have now offered to provide. Generated text is inclined to be generous and comprehensive, because that reads well. Read the scope line by line and ask whether you would be content to be held to each item.

2. Check the exclusions have survived

The most common omission.

Assumptions, exclusions, dependencies on the client, and what happens if they are not met. These are the least interesting paragraphs in the document and the ones that decide any later dispute. Verify they are present and specific to this job.

3. Verify every number against your own quote

Do not trust restated figures.

Prices, quantities, rates, day counts, discount percentages and totals should be checked against the source. Numbers that have been reformatted or summarised are where transcription errors hide, and a wrong figure in a signed proposal is your problem.

4. Check dates and durations against your actual capacity

Timescales are commitments.

A generated programme will happily promise a start date and a duration that reads well. Confirm against your current workload and your team's availability, because a missed programme in a proposal is the fastest way to a bad first month.

5. Remove any claim you cannot evidence

Credentials attract invention.

Accreditations, years of experience, client names, case study figures, insurance levels and team qualifications must all be true and current. Client names in particular usually require permission, and a proposal is a document that circulates.

6. Make sure the client's actual problem is in there

The failure that loses the work.

Generic proposals lose to specific ones. The section that restates what the client said they need, in their words, is the part that cannot be generated from your template and the part that decides the outcome.

7. Read the terms attached at the end

They are not boilerplate.

Payment terms, retention, liability caps, termination, intellectual property, variation procedure. If these were assembled rather than chosen, check that they match what you actually intend to offer this client on this job.

8. Keep a reviewed template rather than regenerating each time

Repetition is a feature.

The strongest use is a template that has been checked once properly, with generated help confined to the client-specific sections. This gets you most of the time saving and almost none of the risk of regeneration.

9. Have someone other than the author read it

The last check.

The person who assembled a document cannot see what is missing from it. A second reader, given ten minutes and a short checklist, catches the wrong client name, the leftover paragraph from another proposal and the number that does not match.

Keep a record of what you sent, in the form you sent it. Disputes about scope turn on the document that was accepted, and reconstructing it later from a template is not the same thing.

Conclusion

Review the parts that create obligations, because a proposal is an offer capable of acceptance.

Read the scope line by line and be willing to be held to every item, confirm the assumptions and exclusions are present and job-specific, check every number against your own quote rather than the restated version, test the dates against real capacity, remove any credential or claim you cannot evidence, make sure the client's own problem appears in their words, read the attached terms as choices rather than boilerplate, keep a properly reviewed template instead of regenerating, and have a second person read it before it goes.