Socket can now draft a proposal for you from a call transcript or a few rough notes. Paste in what you have and the AI matches what it finds against your pricing menu and existing clients, then builds a draft ready for review. What used to be fifteen to thirty minutes of admin after a good call is now an upload and a review step.
Socket AI only ever creates drafts. Nothing is sent to a client until you've reviewed and sent it yourself, exactly as with any other proposal.
Where to find it
Go to the Proposals page and click Generate with AI, next to the New button. The wizard walks you through five steps: Source, Context, Agent run, Review plan, and Draft ready.
Step 1: Tell it where the discovery is from
This is where you give the AI its raw material. A full call transcript works brilliantly, but so does a handful of rough notes typed straight into the box.
Upload a file or paste directly, and don't tidy anything up first. Even a few loose lines, such as where you met the prospect, their business name, and the services they are likely to need, is enough for a well structured draft.
Then choose who the proposal is for:
New Prospect: Socket creates the client record from your notes.
Existing Client: links the proposal to a client you already have, for renewals or scope changes. See Proposal Renewals or Adjustment for more on the renewal flow.
Let AI Decide: searches your client list and matches automatically. Pick this if you're not sure.
Step 2: Add a bit more context (optional)
Add any background your notes don't cover: a referral source, how urgently the client wants to move, or related clients.
There are two toggles here. Use my pricing templates applies your saved bundles where they fit, and Use my design templates lets the AI pick the most suitable presentation profile. We'd recommend leaving both on.
Once you're happy, click Start agent.
Step 3: The agent runs
A live timeline shows the agent reading your notes, identifying the client and services, and checking everything against your pricing menu. The AI can't invent services that don't exist in your practice.
You don't need to sit and watch. Leave the tab and a draft plan will be waiting when the run finishes.
Step 4: Review the plan
Before anything is created, the AI shows you its full plan to approve or override, with a short summary at the top explaining what it's understood.
If anything essential is missing, you'll see a callout. Some fields, like a contact name and email, block the draft until they're filled, and you can add them right there. Below the callout you might also see missing data points that aren't essential but would improve the proposal.
The Clients section shows who the proposal is for and whether the AI matched an existing client. If there's no match, create a new record or link one manually. You'll also see the pain points the AI picked up from your notes, carried through as companion notes on the proposal.
The Services section lists every line the AI wants to include, each with a confidence indicator. A medium or low confidence chip means the AI is less certain about that line, so it is worth checking before you keep it. Anything tagged Extra is speculative rather than core. Remove any line with a click; prices, drivers, and individual lines are editable once the draft opens.
Work that doesn't map to your pricing menu, like a one off bookkeeping migration, is drafted as an ad-hoc project with a written scope. Check these carefully.
Once you're happy with the plan, click Generate draft.
Step 5: Your draft is ready
The proposal lands in Socket as a draft on the proposal dashboard under the AI proposal dashboard.
Open it, adjust pricing and wording, check the presentation, and send it when you're ready.
The AI never contacts a client or sends anything on your behalf.
Getting the best results
The biggest lever is your Socket setup. The AI builds from your pricing menu, pricing templates and design templates, so tidy templates and clear service names make a real difference.
On the input side, the single best thing you can do is give it the full transcript. The more you give it, the more it has to work with, so a full call transcript will usually map far more of the services you discussed than a few quick lines would. The richer the input, the closer the draft is to finished.
Short notes still work well when you want speed and often need barely a change. Just bear in mind that a thin note gives the AI less to go on, so it will sometimes fill the gaps with services that are standard for a practice rather than ones the client actually asked for. Give the Services list a quick read on the review step and remove anything that does not fit. The lower confidence chips are a good guide to what to check first.
One thing transcripts rarely capture is the admin detail: the contact name and email, the company registration number, and the year end. The AI flags these as missing at the review step, and the essential ones need filling before the draft is created. The quickest fix is to add them in the box at the Context or Review step, or drop in a short note or screenshot of who the client is so the AI can build the record properly.
The Socket AI can't look anything up externally. It works only from what you give it plus your pricing menu and design templates, so if a detail matters, make sure it's in your notes or add it at the context step.
Using it from Claude or ChatGPT
The same capability is available through Socket's MCP connection. Ask Claude or ChatGPT to draft a proposal in plain English and it will build and validate it in exactly the same way, landing in Socket as a draft for your review. See Connecting AI Assistants to Socket (MCP) to set that up.
Questions or feedback?
This is an open beta and your feedback shapes where it goes next. If something doesn't work the way you expected, or you've got an idea that would make it more useful, message us via the chat icon in Socket.