Skip to content
All insights

Industry operations

Most proposal work is assembly, not writing.

A proposal contains a small amount of original thought and a large amount of material that already exists somewhere in the firm. The ratio is what makes this workflow worth examining.

Published September 8, 20265 min read

What this covers

  • The expensive part of proposal work is finding and assembling existing material, not deciding what to offer.
  • Client requirements are buried in email threads and attachments, and extracting them is a mechanical step.
  • What is offered and at what price stays entirely with your team, and a deployment should say so plainly.
  • Confidentiality is a design constraint here more than in any other sector, which shapes where the agent runs.

A professional services firm sells the time of people who are expensive. Which makes it worth asking, honestly, what those people spend their time on between the work that is actually billed.

The answer, in most firms, is documents. Assembling proposals from previous proposals. Rebuilding a methodology section that exists in four earlier versions. Extracting what a client actually asked for from a thread with six replies and three attachments. Producing the same project documentation for the fifth client this quarter.

The proposal, broken into its parts

PartWhere it comes fromWho should do it
What the client asked forAn email thread, a brief, a tender documentExtraction: mechanical
Our understanding of the problemThe consultant's reading of itThe consultant
Approach and methodologyPrevious proposals, adaptedAssembly, then the consultant's edit
Team and credentialsThe firm's own recordsAssembly
Relevant experiencePrevious engagements, filtered for relevanceAssembly, then selection
Scope, price and termsA commercial decisionThe partner, always

Three of those six rows are assembly. They are also, in most firms, the rows that consume the most elapsed time, because the material is scattered across a document store, previous proposals and individual people's memories of which engagement was similar.

The consultant should be editing a draft that already has the client's requirements extracted and the relevant precedents gathered.

Requirement extraction, which is genuinely tedious

Client requirements rarely arrive as a list. They arrive as a first email, a follow-up that adds two constraints, an attachment with a scope of work, a call summary, and a later message narrowing the timeline. Reconstructing the actual ask from all of that is a careful reading exercise that takes an hour and produces nothing billable.

An agent reads the whole thread, including the attachments, and produces the structured version: what is being asked for, the constraints, the dates, what was added or changed later, and what was asked but never answered. The last of those is frequently the most useful output, because the unanswered question is the one that causes a scoping problem later.

The line that does not move

It is worth being blunt here, because the sector is understandably wary. An agent does not decide what your firm offers, what it costs, or what you are willing to commit to. Those are the judgments clients are paying for, and automating them would be selling something other than professional services.

What it does is arrive at the point of judgment with everything prepared: the requirements extracted, the precedents gathered, the credentials assembled and the standard sections drafted from your own approved material. The partner edits and decides. Where the agent stops covers how that line is set explicitly rather than assumed.

The back office behind the billable work

Proposals get the attention, but the steadier drain is the operational administration that surrounds the delivery.

  • Routine client requests: status, documents, previous deliverables, invoice queries.
  • Document requests that involve finding something somebody produced eighteen months ago.
  • Supplier and subcontractor invoices, matched against engagements and approved.
  • Purchasing services and materials for a project, including the chasing.

None of these are intellectually demanding and all of them currently interrupt people who were doing something else. Customer operations automation and invoice processing cover those two patterns in general form.

Confidentiality as a design constraint

This sector has a constraint the others have in weaker form: client documents are frequently covered by confidentiality undertakings that restrict where they may be processed and who may see them.

That has a concrete implication for deployment. The agent runs inside your own infrastructure and works directly in your systems, so client documents are not copied into a shared service to be processed. Where model inference runs, and the data terms that apply to it, are agreed before deployment and named explicitly rather than implied. In a sector where you may have to answer that question in writing to a client, that distinction is not a technicality.

Security sets out the deployment boundary, and AI agents for B2B services covers the workflows.

Questions

Questions we get about this.

Will it write our proposals?

It prepares them. The client's requirements extracted from the thread and attachments, relevant precedents gathered, credentials assembled and standard sections drafted from your own approved material. What is offered, at what price and on what terms stays with your team.

How are confidential client documents handled?

The agent runs inside your own infrastructure and works directly in your systems, so client documents are not copied into a shared service to be processed. Where model inference runs, and the data terms that apply, are agreed with you before deployment and stated explicitly.

Does it work with our CRM?

Yes. The CRM platforms common in professional services are among the systems it works with, and records are updated as part of the workflow. Per-system implementation is scoped during the review.

What does it do with requirements a client never clarified?

It surfaces them. Reconstructing the ask from a thread usually reveals questions that were raised and never answered, and listing those is often the most useful part of the output because they are what causes a scoping dispute later.

Is the first workflow always proposals?

Not necessarily. Where proposal volume is low but client requests and supplier invoices are constant, the back-office workflows return more time. The test is the same as anywhere: volume, stated rules, reachable data and a defined outcome.

See what this would look like in your operation.

A Workflow Review traces one of your real workflows end to end and names the first step an agent could take over. No platform to evaluate first.

Start with one workflow

Fixed scope for the first workflow, agreed before we start. Expansion is your decision once it proves value.

How it works

Security, stated plainly

Where the agents run, what they can and cannot do, and exactly what we do and do not claim.

Security