The proposal written at midnight
The thinking took an hour; the document takes a week of evenings, and the client has already spoken to two other firms.
See the survey pictureEste ejemplo se está actualizando. El texto de abajo sigue en inglés.
Example · Professional services and consultingProposal and onboarding agent
The proposal that takes a week to write
An agent that assembles a proposal from your own approved material, collects the client facts it still needs, and never invents a case study, a credential or a result.
Illustrative scenario — not a real client result
Proposal desk · 09:15
- Answered
3 drafts assembled
09:16 · From approved blocks: scope, team, fees
- Answered
Missing facts requested
09:19 · Budget range, decision date, signatory
- Person
“Can you guarantee 20% savings?”
09:22 · Commercial promise · with a partner
No case study, number or credential is generated
Proposal and onboarding agent · Professional services and consulting
- Sourced facts
- A partner signs
- Approved material only
01 — A typical day
Three small failures that add up.
What professional services and consulting owners tell us goes wrong in an ordinary week.
A case study that is not quite true
A drafted example is close enough to sound right, and it goes out under the firm's name with a number nobody checked.
See the documented caseThe facts you needed and never asked for
Budget, decision date and who signs are discovered after the proposal, so the scope is wrong from the start.
See the intake questions01 — A typical day · the evidence
What the numbers actually say.
Each figure opens its source, year, sample and limits. Click a badge to check it.
01
Source
Thomson Reuters, AI and professional time survey
2024
Method and sample are recorded in the source registry.
Read the source for its scope and limits.
Open the source4 hrs
a week professionals expected AI to save in the coming year
Thomson Reuters survey of 2,200+ professionals, April–May 2024. A forecast, not a measured saving.
Thomson Reuters · 202402
Source
Thomson Reuters, AI and professional time survey
2024
Method and sample are recorded in the source registry.
Read the source for its scope and limits.
Open the source12 hrs
a week they expected to save within five years
The same forecast, for legal, tax and risk professionals. Explicitly not a result.
Thomson Reuters · 202403
Source
Intuit QuickBooks, accountant technology survey 2024
2024
Method and sample are recorded in the source registry.
Read the source for its scope and limits.
Open the source98%
of surveyed accounting professionals had used AI with clients
Intuit survey of 707 US accounting professionals, March 2024. Vendor-commissioned, self-reported, and accounting-specific.
Intuit QuickBooks · 202402 — What the helper does
From your records to a decision you own.
The agent takes the brief, asks the questions that decide scope — budget range, decision date, signatory, constraints — and assembles a draft from blocks your firm has approved.
Every case study, credential and figure in the draft comes from your library, with its source attached. If the library has nothing suitable, the draft says so rather than filling the gap.
The partner edits, prices and signs. Commercial promises, discounts and any statement about results go to a person.
6 steps · 2 decisions you own
Reads from
- Document store
- CRM
- Calendar
Text version of the flow
- 1. Take the brief — Problem, scope, constraints.
- 2. Ask the deciding questions — Budget, date, signatory.
- 3. Assemble from approved blocks — Scope, team, fees.
- 4. Cite or say nothing — No invented evidence. Human checkpoint: Every claim has a source.
- 5. Stop at promises — Results, discounts, guarantees. Human checkpoint: A partner owns the promise.
- 6. Onboard on signature — The facts you will need.
02 — What the helper does · a simulated example
One simulated example, step by step.
Use the buttons or the left and right arrow keys. Nothing here is a real customer.
Simulated example
Aldworth Partners, an eight-person consultancy
Email · 09:15
A strategy consultancy responding to an inbound brief. The agent assembles a draft for a partner to finish. Every message is simulated; no real client data is used.
Aldworth Partners' assistant. I can gather the details for a proposal and prepare a draft for a partner to review. Anything about fees or outcomes comes from them.
03 — What you decide
The helper prepares. You decide.
The agent assembles. A partner commits.
Approved blocks only
Scope, team and fee text come from the library with a version and an owner attached.
Cite or flag
A claim without a source is removed or marked as a gap; the agent cannot write new evidence.
A partner signs
Pricing, discounts, guarantees and anything about results are a partner's decision.
You can pause it
Pause for a service line, a partner or the whole firm; briefs fall back to the partner's inbox.
Without your approval · No invented case study, credential, statistic or result — and no fee, discount or guarantee leaves the firm without a person.
04 — What changes when a safeguard is missing
Documented cases, not scare stories.
Three documented cases with their sources. Two concern citations that did not exist; one concerns records that were reconstructed and not labelled.
Nonexistent references in a paid report
A government-commissioned report on welfare-system IT contained nonexistent academic references and a fabricated court quote. Deloitte disclosed generative-AI use in the revised report and agreed to repay the final instalment; reporting describes the errors as apparently AI-generated rather than proving AI caused each one.
What the helper would do differently
Every citation in a draft is checked against the original before it leaves, and a human signs the source list. A model's output is never treated as a citation.
Refund
final contract instalment repaid
Six cases that were never reported
A filing relied on six nonexistent cases produced with ChatGPT. When challenged, the lawyers stood by the citations rather than checking them, and the court imposed a $5,000 penalty jointly and severally.
What the helper would do differently
The draft marks each claim with its source. A partner who cannot find the source in the library removes the claim — standing by an unverified assertion is the failure mode.
$5,000
sanction on the lawyers and firm
Reconstructed records filed as originals
In a litigation-support engagement, an accountant asked the client to recreate 26 missing estimate sheets and placed the reconstructions in the project files without marking them. The opposing party alleged fabricated records; the account describes an appeal and remand, not a final finding of liability.
What the helper would do differently
Anything the agent assembles is labelled with its origin and date. A reconstruction is marked as one, and client-facing evidence is approved by a person.
Litigation
malpractice claim over unlabelled reconstructions
04 — What changes · the evidence
The rest of the evidence.
Every figure resolves to a source with its sample and limits. No figure without one.
2,200
professionals in the Thomson Reuters AI survey
Legal, tax and risk professionals, April–May 2024. Vendor-run; about 44% based in the US.
Source
Thomson Reuters, AI and professional time survey
2024
Method and sample are recorded in the source registry.
Read the source for its scope and limits.
Open the source12 hrs
a week expected to be saved within five years
A forecast from that survey, not a measured saving. The page labels it as one.
Source
Thomson Reuters, AI and professional time survey
2024
Method and sample are recorded in the source registry.
Read the source for its scope and limits.
Open the source707
accounting professionals in the Intuit survey
March 2024, paid panel with post-stratification. Vendor-commissioned.
Source
Intuit QuickBooks, accountant technology survey 2024
2024
Method and sample are recorded in the source registry.
Read the source for its scope and limits.
Open the source
04 — What changes · your week
The same week, two ways.
Drag the handle, or use the left and right arrow keys when it is focused.
Typical month, beforeTypical month, after
Week 1
A brief arrives; the partner replies three days later with questions.
Week 2
The draft is written from the last proposal, with the old client's details left in.
Week 3
A case study is described from memory and a number is roughly right.
Week 4
Budget and signatory are discovered after the proposal is sent.
Week 1
Brief answered the same day; scope questions asked up front.
Week 2
Draft assembled from versioned blocks; no stale client details.
Week 3
Every case study links to the library; gaps are flagged, not filled.
Week 4
Partner prices and signs; onboarding starts from a fixed checklist.
Illustrative example · not a measured result
05 — Your version
Four questions about your own week.
Keep your answers. They are the brief for the first version we would build for you.
How many proposals go out in a month, and what does each one reuse?
Which case studies, credentials and results are approved to quote?
What do you still need from a client before the first call?
Who signs off a proposal, and what do they read first?
Next
Put your own figures into the calculator below. It runs in your browser, and nothing is sent or stored.
Put your week into the calculatorBring the four answers to the free one-hour call. We will use them to sketch the first version, and you keep the written plan either way.
Talk about your business05 — Your version · your numbers
Put your own week into it.
An estimate, not a promise. The model and its assumptions are printed below.
Your inputs
Inbound and outbound.
count
Drafting, formatting, chasing facts.
hours
Scope, team, standard terms.
%
Consultant or partner time.
per hour
Your own figure, last year.
%
Example defaults are typical for a small business. Replace them with yours. Nothing is stored, sent or shared.
Estimated result
IllustrativeConsultant hours back each month
0–0
- Time released at your rate
- £0–£0
- Briefs answered the same day
- 0–0%
- Time to a first draft
- Already immediate
How we calculated this
Routine work handled by the helper: 70% of the share you marked routine.
Time saved per item: 80% of the minutes you entered.
Monthly volume: weekly figures × 4.33; monthly figures used as entered.
Cost: hours saved × your staff cost per hour, in pounds as an illustration.
Reply time: routine items move within the hour while the helper runs. Complex work still waits for a person.
Questions
The things people ask next.
No. It uses entries from your approved library with their source and version. If nothing fits, it says so and flags the gap to the partner.
No. It states the standard schedule and passes any pricing, discount or guarantee question to a partner.
Not on its own. It assembles the draft, checks that every claim has a source, and sends it to a partner to edit and sign.
The library marks which entries may be reused and whether they are anonymised. A client-specific detail is never carried into another draft.
It has no ability to add a figure that is not in the library. A gap is reported as a gap and must be filled by a person.
After signature it collects the onboarding facts from a fixed checklist and books the kick-off. Scope changes go to the partner.
Three weeks: library curation, block versions, approval rules and a parallel run on live briefs.
Start here
Show us three proposals.
One hour with two proposals you won and one you lost. We will show you what the agent would assemble, and where it would have to stop.