Before AI writes your quotes: drop 4 kinds of past jobs and log why
Before AI drafts quotes from past jobs, split reference jobs from excluded ones and log why and when to restore each.
Key points
- Pick reference jobs: AI won't vet prices
- Drop 4 kinds of jobs with skewed prices
- Log why in one line, with a return rule
- Check the jobs AI used before approving
Contents
Pick reference jobs before AI quotes
A&A perspective
When you hand quote drafts to AI, the first thing to settle is not how accurate the AI is but which past jobs it should copy. A reference job here means a past job the AI looks at to ask what similar work cost before it prices a new one.
From the sources
Anthropic, a US company that develops AI, published a post on September 15, 2026 introducing features for small businesses. It says that of more than 1,000 owners it spoke with in the spring, about a third wanted help growing the business: finding leads, answering inquiries and writing proposals.
From the sources
The post describes the proposal flow like this. You walk a job site and record a two-minute voice memo on the way back. By the time you sit down, a proposal priced from your past jobs is waiting for your edits. Once you approve it, it goes out to the client for signature.
From the sources
KANE, a creative and growth agency in the US, used to spend weeks per deal on scoping, pricing, proposals and contracts. Its co-founder built a pipeline that scores fit from meeting notes, prices from the agency's own past projects and drafts the proposal.
From the sources
A process that used to take weeks now finishes in almost a few minutes, and even clients notice, one of the co-founders says.
A&A perspective
Automate & Augment (A&A), the company writing this article, cares about what the post leaves out. What the AI reads is what you charged for similar work, not whether that price was right. If discounted jobs are in the pile, their low prices become a model for the next quote too.
A&A perspective
When you wrote quotes by hand, you remembered that a job was priced low as a favour, and you corrected for it. That correction lived in your memory, not on the quote. Hand the old quotes to AI and the correction is the one thing that drops out.
A&A perspective
The post is a product announcement, and the time saved is KANE's own account. It says nothing about how to check whether a price was sound, or how to choose the jobs to copy. Choosing them is work you do yourself before handing quotes to AI.
Drop 4 kinds of jobs whose price is off
A&A perspective
Exclude four kinds of jobs: loss-making jobs, discounted jobs, jobs whose scope grew, and jobs whose conditions have changed. In each, the recorded price does not show what the work really costs.
A&A perspective
A loss-making job is one where the actual days and materials, recalculated at today's rates, come to more than you were paid. Its price is low by the work you could not see when you quoted.
A&A perspective
A discounted job is one priced below list because it was a first job, a referral or a slow season. The work itself often went well, so it looks like a good job on the list. But once a price cut for a one-off reason becomes a model, it becomes your usual price.
A&A perspective
Exclude jobs whose scope grew, even if extra billing recovered the cost.
From the sources
Paul Graham, who has advised founders, wrote in 2013 about a product company giving one customer consultant-like attention. While you are not paid for it, the customer is grateful. Once they start paying for that attention by the hour, he writes, they expect you to do everything.
A&A perspective
A&A applies this to quoting. On jobs where extra work started being billed by the hour, the work asked for keeps growing past the original promise. So the price of a job whose scope grew is closer to the price of the hours spent than to the price of the work first promised.
A&A perspective
A job whose conditions have changed is one where the conditions of the time no longer exist. That includes jobs where material or subcontractor prices have since moved, and jobs where the client then had someone who could settle the specification. This kind grows as time passes.
| Kind | Use as reference? | Reason to log |
|---|---|---|
| Loss-making job | Exclude | Main reason for the overrun |
| Discounted job | Exclude | Reason for the discount and the list price then |
| Job whose scope grew | Exclude | Kinds of work that grew |
| Job whose conditions changed | Exclude | The condition that no longer holds |
| None of the above | Use | No reason to exclude; note the pricing assumptions |
Log why in one line so you can undo it later
A&A perspective
For each excluded job, write a one-line reason, not just a mark. Without it, when you revisit pricing six months later you won't know why the job was excluded, and you will make the same call again.
A&A perspective
The line holds two things: why you excluded the job and the condition for bringing it back. What to write as the reason is in the right-hand column of the table in the previous section. For example: "Discount. List price was 1.2 times the quote. Quote list price if the same request comes again." With the condition written down, the list of excluded jobs becomes a draft of your pricing policy.
A&A perspective
Don't have AI write this line. Details such as "we cut it as a favour" or "the client's contact knew the work, so it took less effort" are in no record. What you give the AI is only the result: the "use as reference?" column.
Hypothetical example
From here on, every number is a made-up example. Say a 20-person company that fits out shops and offices adds two columns to its list of 40 past jobs: "use as reference?" and the one-line reason. It excludes 23: 5 loss-making, 8 discounted, 4 with grown scope and 6 with changed conditions.
Hypothetical example
Of the 17 jobs left as references, 12 are shop fit-outs and 5 are office fit-outs. Five is too few to hand office quotes to AI. So the company lets AI draft only shop quotes, and keeps quoting offices by hand while recording costs by stage.
A&A perspective
Choosing reference jobs looks like work to improve pricing accuracy. In practice, it decides which work you can hand to AI.
Check the jobs AI used before the price
A&A perspective
When you approve a draft, check the past jobs the AI used before you look at the price.
From the sources
According to the post, by default the AI in this feature waits for the user's approval before anything sends, posts or pays. Every workflow starts in approval mode, it says, and once you are comfortable you can let each one run on its own.
From the sources
The post also describes use without connections to outside tools such as accounting software or e-signature services. Upload the memo, photos and a past proposal as a template, and the AI builds and prices the proposal from those. Sending it to the client for signature is up to the user.
A&A perspective
A&A thinks approval just before sending comes too late to stop a pricing error. Faced with a price already laid out as a proposal, you see a polished package, not the jobs behind the price. Looking only at the price without seeing the references, approval tends to become a formality.
A&A perspective
So have the AI list the numbers of the past jobs it used alongside the draft. Before the price, the approver checks that each listed job is marked "use as reference" on the list. If an excluded job slipped in, fix the references before fixing the price.
A&A perspective
This check should take less time than rethinking whether each price is right. That is A&A's expectation, not a measured result. The question changes from whether this price is right to whether this job may be a reference, which the list answers.
A&A perspective
Neither the post nor Graham's essay shows that choosing references leads to correct pricing. Choosing alone does not guarantee that the remaining jobs' prices are right.
Summary
A&A perspective
AI does not check whether past prices were right, so before handing it quotes, choose the past jobs it should use as references. Exclude four kinds: loss-making, discounted, scope-grown and changed-condition jobs.
A&A perspective
For each excluded job, keep one line with the reason and the condition for bringing it back. When you approve a draft, check the reference jobs before the price.
AI that drafts quotes from past jobs does not check whether those prices were right. List your recent jobs, mark which ones to use and why others are out, then decide whether to hand AI the drafts for the kinds of work with enough reference jobs.
FAQ
Is this needed if I only have a few past jobs?
With only a few jobs, recording each job's costs by stage helps more than choosing references. Choose references once you have enough comparable jobs. If your business has a single list price, the problem of choosing references does not arise.
How many jobs should I start with?
Adding two columns, "use as reference?" and "reason and return rule," to a list of your 10 most recent jobs is enough to start. How many of the 10 remain as references tells you whether you can hand quotes to AI now. If few remain, tidy your job records before handing quotes to AI.
Do I delete excluded jobs from the list?
No. Keep them with the reason and return rule, and bring them back as references when the condition is met. If you delete them, six months later you cannot check why they were excluded.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- Claude for Small Business launches new workflows, integrations, and training programs
Anthropic · 2026-09-15
Accessed 2026-10-24 - Do Things that Don't Scale
Paul Graham · 2013-07
Accessed 2026-10-24