AI for sales in a small business pays off between inquiry and proposal, where the writing work sits. It can prepare five steps: researching the client before the call, the call notes with need, constraints and open points, the proposal draft from your service building blocks, the follow-up email and keeping the client record in your CRM up to date. What stays with you are the decisions with consequences: the price, a discount, any commitment to the client and the sending. And because client data runs through the AI tool, you sort out the data processing agreement first.
In my business, Anton, my AI employee for proposals and invoices, handles this path. He turns an initial conversation into a proposal PDF and, once the client says yes, creates the contact, offer and invoice as drafts. The rule behind it is short: Anton drafts, I approve. No offer and no invoice leaves the system without my click. How he works in detail is on Anton, who writes the proposals.
Two neighbouring topics stay short here. Reaching new contacts, and where the law draws limits, is covered in Automating Cold Outreach: Limits for AI. Answers to objections like “too expensive” from real calls are in Objection Handling From Real Sales Calls. This article starts when the inquiry is in and ends with the prepared proposal in the system.
AI for sales: five steps from call to proposal
The right-hand column is why the left-hand one works. An AI that writes proposals and also decides on the price is negotiating in your name.
Before the call: researching the client
The best preparation is one page, not a dossier. The AI reads the company's website, public information on its services and, if there is any, your previous exchange in your own system. From that it writes a short profile: what the business does, what the inquiry said, which questions are open. Every statement carries its source, so you can see what is backed up and what the AI is only guessing.
Two limits apply. You research the company and the reason for the inquiry, not the private life of the contact person. And an AI searching the web can mix sources up. How to check research against its sources is in Deep Research: Which AI Researches Best.
Call notes and summary
After the call, what matters is what follows for the proposal. Useful call notes have four blocks: the need in the client's own words, the constraints (deadline, budget signal, decision maker), your commitments and the open points.
A documented run with Anton shows what this looks like. He received an inquiry with a leadership topic and time pressure, “before October”, with an open budget. Instead of guessing a price, he suggested targeted follow-up questions. In the conversation review he identified the need for individual leadership coaching, a budget signal of “up to two and a half, I'd have to sleep on it”, the commitment “offer by the end of next week” and one open point, possible employer co-funding.
If you want to record the call, you need everyone's consent first; the rules are in Transcribing Calls: What German Law Says. How Teams, Zoom and Meet take notes is in Automatic Meeting Minutes with AI.
The proposal draft from your building blocks
An AI can derive line items from the call notes, write the texts in your voice and recalculate the totals. It cannot know a price you have not given it, so quality depends on your price list, not on the model.
For Anton, in the run above, that meant: he picked the individual coaching building block, stayed within the 2,500-euro signal and delivered the proposal PDF together with a reasoning memo. He recalculates net, VAT and gross himself and checks the finished PDF page by page as an image for line breaks and special characters. Two rules guard against the most common mistake. Without a price source, he sets no price. If he finds two different prices for the same service in the materials, he lists both with their source and asks, instead of averaging.
An AI employee like this is trained on five to ten old proposals rather than a long prompt, and feedback goes into an onboarding journal that is read back into every new proposal. The drafting steps in detail: Writing Quotes and Invoices with AI; the structure: How to Write a Quote That Convinces.
Following up without pushing
Most proposals do not fail because of a bad follow-up email, but because nobody writes one. An AI can watch the deadline and put the draft in front of you. The timing comes from the validity period in the proposal, the content from the call notes: the open point the client wanted to clarify is a better reason to write than “just checking in”.
In my business, Sonja writes business emails, usually handovers to a new client after a website redesign. For a campaign of 15 such emails she also prepared the follow-up round. She checks every fact against the source, writes in my voice and drops the text into my inbox as a draft. Nothing goes out until I say so explicitly. Following up on a proposal is the same process with a different source. When to follow up, how often and when to stop is covered in How to Follow Up on a Quote by Email.
CRM upkeep as a by-product
Client records get left undone most reliably, although everything needed comes up in the steps before: status, next step with a date, open points. Let the AI fill in these fields after each step as a suggestion, not as a silent change. What is a fact (“proposal sent on …”) you confirm; what is an assessment (“client seems price-sensitive”) stays marked as one or gets dropped.
Whether your CRM is a dedicated system, an accounting tool or a spreadsheet matters less than an interface through which the AI can create drafts without making anything final. Once the client says yes, Anton creates the contact, offer and invoice as drafts in Lexware Office; through a suitable adapter, sevdesk or another accounting system with an appropriate interface also works. Final approval and sending stay with me. The ways to connect it are in Lexoffice Integration: The Main Routes.
Where the human decides: price, discount, commitment
Four decisions you do not hand over, even if the AI could technically make them.
Price. The AI fills in prices from your list. Whether a job deserves a different price is your call.
Discount. A discount is a decision about your margin and about the signal you send the client. An AI may point out that a client asked for one, but not grant it.
Commitment. Deadlines, scope and guarantees written into a proposal bind you. What was not promised in the call does not belong in the proposal.
Sending and closing. Anton never finalizes on his own. Even if pushed, he first names the consequences, final invoice number and immutability, and wants a clear yes from me.
Data protection: client data and the DPA
Three questions set the frame. This isn't legal advice, just the operator's perspective.
Are you allowed to process the data? If a prospect contacts you, Art. 6(1)(b) GDPR lists as one possible basis that processing is necessary “in order to take steps at the request of the data subject prior to entering into a contract”.
Does the AI service work on your behalf? Then it is a processor, and Art. 28(3) GDPR requires a contract that sets out, among other things, the subject-matter, duration, nature and purpose of the processing and stipulates that the service processes the data “only on documented instructions”. Before the first client record, check whether the provider offers such an agreement for your plan. If it does not, client data does not belong there.
Do you record calls? In Germany, recording the privately spoken word of another person without authorisation is a criminal offence under section 201 of the Criminal Code. Consent comes before the microphone, not after.
What the AI gets to see at all is covered in AI and Privacy: What the AI Gets to See.
Frequently asked questions
Where does AI help most in a small business's sales?
With the proposal draft and the call notes, because that is where most of the writing sits. Research, follow-up and CRM upkeep come next, once these two steps run smoothly.
Can an AI send proposals on its own?
Technically yes, sensibly no: a proposal binds you to price, scope and deadline. For Anton this is a fixed rule: no offer leaves the system without my click.
What are examples of AI in sales?
A short profile before the first call, call notes with needs and open points, a proposal from service blocks with recalculated totals, a follow-up email at the end of the validity period and a CRM entry with the next step. In each case the AI delivers a draft and a human approves.
Can I enter client data into ChatGPT or Claude?
Only into a plan for which there is a data processing agreement under Art. 28 GDPR, and only as much as the step needs. This is not legal advice.
How to carry on
Take your last three proposals and put next to each one what was said in the call before it. Mark which line item came from which sentence and where every price came from. Wherever you find no source is the gap an AI would later fill by guessing; close it first, with a price list. After that, your first AI proposal draft is a comparison, not a guessing game.
Anton is available as a ready-made package in my AI Employee Framework community, with the building-block system and the onboarding journal. In the calls we work out together which steps in your business go to the AI first. Everything about it is under Community. What happens after the deal, when clients have questions, is covered in AI in customer service.
