Insights · The AI Department · Sales

What an AI Sales Director Actually Does All Day

An AI sales director does not sell. It drafts the follow-up, the proposal and the pipeline notes from approved sources, and the salesperson still runs every conversation.

A woman in her late thirties, owner-seller of a small HVAC company, in a plain work polo at the front-office desk, phone just set down, eyes closed for one seco

When people hear "AI sales director," they picture a machine that sells. It sends the outreach, works the pipeline, closes the deal, and the salesperson becomes optional.

That is not what the role is, and it is not what we would want it to be. Inside a 10 to 50 person services business, the person who sells is usually the owner or one senior seller with the relationships. Nobody is replacing that. What is replaceable is the pile of drafting, chasing and tidying that sits behind every conversation and never gets done because the seller is on the phone.

So here is what Mercury, the sales specialist in the Nexus department, actually does all day. It is less glamorous than the pitch, and considerably more useful.

Where the selling hours actually go

Picture a twelve-person HVAC company. One person quotes and sells. Each week that person runs site visits, sends estimates, and then, in theory, follows up on every estimate that has gone quiet. In practice the follow-up happens for the two or three deals that feel hot, and the rest drift until the customer either calls back or does not.

Ask that person where the time goes and the answer is rarely "talking to customers." It is finding the last thread, remembering what was promised, re-reading the estimate, drafting a message that does not sound like nagging, and updating the CRM afterward, if there is time, which there is not.

That is the work an AI sales director takes. Not the conversation. The preparation around it.

Morning: the follow-up queue, drafted

The day starts with the deals that have gone quiet. Mercury reads what it is allowed to read: the CRM notes, the estimate that went out, the last email thread, and whatever call notes the seller has approved for the system to see. From that, Mercury drafts a follow-up for each one.

The draft says something specific and true. It refers to the actual scope, the actual objection from the last call, the actual next step that was floated and never confirmed. It is not a template with a name field.

Then the seller reads it. This is the part that matters. Every message is edited or discarded by a person, and every message is sent by that person, from their own account, in their own voice. Mercury drafts, you send. The system is designed so nothing reaches a prospect any other way, and outreach is never scheduled into the hours when a recipient would rather be asleep.

Where Mercury has inferred something, the draft says so. A note like "your CRM shows the estimate sent on Tuesday; the guess about budget is mine" keeps your data and the system's reasoning from blending into one confident paragraph. The seller can accept the fact and reject the guess.

Midday: the proposal that sounds like you

When a discovery call ends and the customer wants something in writing, the proposal usually waits until evening, because the seller has three more calls. Mercury drafts it from the call notes and the approved offer, so the first version is waiting when the seller sits down.

The rule that makes this safe is that Mercury works from approved claims only. If your company has three results you are willing to put in front of a buyer, those three appear. If a proof point is not on the approved list, it does not get invented, however persuasive it would be. No made-up testimonials, no rounded-up outcomes, no borrowed case study from the internet.

We wrote about the mechanics of this in turning meeting notes into a draft proposal. The short version: the draft is only as good as the source material, which is exactly why the source material is governed.

Afternoon: pipeline hygiene, which is really pipeline honesty

Every services business has a pipeline where deals sit in the same stage for months. Nobody moves them because moving them means admitting they are dead, and nobody works them because they feel dead.

Mercury's afternoon job is to look at that pipeline and name what is happening in it. Which stage do deals stall in. What the stalled ones have in common. Which ones show a clear next conversation and which ones show none. The output is a short map, not a dashboard, with re-engagement language for the deals worth one more try and a plain recommendation to close the file on the ones that are not.

Again, every observation is labeled. "This is what your data shows" and "this is my read of it" are two different sentences, and Mercury is built to keep them apart.

What Mercury will not do

Some of this bears saying plainly, because the industry's default pitch is the opposite.

  • Mercury does not send. Not an email, not a text, not a reply in a shared channel with a customer in it.
  • Mercury does not invent proof, testimonials or results. Only claims you have approved.
  • Mercury does not use pressure tactics or false urgency. Persuasion stays honest, which is the only kind that survives a second deal with the same customer.
  • Mercury does not spend, discount or commit. A price is your decision.
Mercury drafts. You send. There is no third option in the design.

Mercury also stays in one lane. When you bring a problem that is really about positioning, Mercury will say so and, if you ask, the work moves to Apollo. When an offer's economics need checking before anyone sells them, that is Midas. When a stalled deal is really a relationship question, that is Venus. When the real issue is who you are selling to at all, Athena. Those handoffs happen when a person asks for them. Nothing gets rerouted behind your back.

All of this runs inside Nexus, the governed system installed around the company: identity, permissions, approved sources, logging, and the connections into the CRM and inbox the team already uses. Mercury is one of the ten specialists that system can bring in when the work calls for it. If you have read why an AI department is a reporting line, this is that idea applied to sales.

The person at the end

At the end of every one of those workflows is the salesperson. They read the follow-up and decide whether to send it. They rewrite the proposal paragraph that does not sound like them. They look at the pipeline map and make the call on which deals to work and which to let go. They run every conversation, and they close.

That is the whole point. The specialist takes the drafting and the chasing so that the person with the relationships can spend their day in the conversations only they can have. For a seller who wants to sharpen those conversations, OnCue by Advisy is the practice layer for that, separate from the department itself.

If you want to know which of your deals should be closing but is not, and where the follow-up quietly dies in your business, that is what our two-day audit finds. Tell us how your business runs today and we will come back with one recommended first project, its deliverable, and what it would need from you.

Who is running your AI department?

We start with a two-day audit of how work actually moves through your business, before anyone talks about building anything.