AI follow-up copilot for quotes and proposals
The follow-up copilot
The cheapest revenue an operating business loses is from people who already asked to buy. Follow-up dies because it depends on someone's memory in a week that's already full. This is what a follow-up copilot looks like as a shipped system — and what it can't do.
01
What it is
The leak has a shape. A quote goes out Tuesday. The customer has two questions Thursday. Nobody answers until the following Wednesday, because the estimator is on site and the inbox doesn't escalate. Multiply by every open quote and the pattern appears: the pipeline isn't short of demand, it's short of attention — revenue already earned the hard way (site visit done, quote built, trust established) quietly expiring.
The copilot is not "an AI salesperson." It watches the quote pipeline, knows which items are aging past your response norms, drafts the follow-up in your voice with the quote attached, and routes judgment calls — discount requests, scope changes — to the owner instead of answering them itself. A human approves every send for the first month; after adoption settles, routine nudges go out on schedule and exceptions still stop at a person.
02
When it's the right wedge — and when it isn't
Three things have to be true first:
- Quotes live somewhere queryable — an estimating tool, a spreadsheet with discipline, even a folder with consistent naming.
- One named owner for the pipeline. The system reports to a person, not to "the team."
- A response norm you actually believe: "every open quote gets touched every 5 business days" is enforceable; "we stay on top of things" is not.
If quotes live in text-message photos, we fix that first — that's a workflow redesign, not an AI problem. If follow-up isn't your biggest leak, don't start here; most operating businesses have four to six leaks running at once and this is only sometimes the top one. And if the pipeline is short of demand rather than attention, follow-up polish won't create demand — that's a different problem and we'll name it.
03
How it's built
The four layers of the spec:
- Entities — quote, contact, touch, response norm, exception. The pipeline's memory made explicit.
- Workflow — watch → detect aging → draft → approve → send → log, with the approve step held by the pipeline's named owner.
- Agents — drafting against the quote's actual content in your voice; escalation agents that recognize a judgment call and route it up instead of improvising.
- Surface — an approval queue, often one lane of an operator console, where a send is one keystroke and a skip is a logged decision.
It ships with an owner and an SOP, and the measure stage puts hours returned and quote-response time in the ledger — reviewed against what the wedge was scored on, not against enthusiasm.
04
Evidence boundary
This remains a build pattern rather than a published case study. Cab‑O‑Matic's verified catalog breadth shows the kind of vertical context a quoting follow-up system may need, but it does not by itself prove a deployed follow-up workflow or outcome.
05
What a v1 proves in weeks
One pipeline, human-approved sends, real quotes. The v1 proves that aging quotes surface reliably instead of depending on memory, that drafts are good enough that the owner edits less each week, and that the response norm actually holds for the first time.
The math worth running before building: open quotes per week × follow-ups currently missed × close-rate delta × average quote value. That's illustrative math, not a promise — but if the range doesn't clear the cost of the build, the wedge score says so and we don't build it.
FAQ
Questions this pattern has to answer.
Will customers know it's AI?
The drafts are in your voice and a person approves them until you trust the system. What customers notice is that they hear back the same day.
We use an industry CRM — does this work with it?
Usually. The harder question is whether your quotes are actually in it consistently — that's part of the week-one inspection, and if the answer is no, fixing that comes before the copilot.
What does this cost to run?
Pennies per follow-up in model costs. The real cost is the build and the adoption work, which is why the wedge has to clear scoring before anything gets built.
What if follow-up isn't our biggest leak?
Then don't start here. Start with the Product Wedge Review — the point of scoring is to find the workflow where a v1 proves value fastest, and follow-up is only sometimes it.
Next step
Score the wedge before you build.
Bring the workflow, owner, data, and proof line. The review turns that into a build, fix-first, or don’t-build call.