Sales sales department: where to start
Automating sales is usually pitched as an all-at-once move. In practice, the opposite works: hand the machine a few spots where people waste time, and leave the rest alone. Here's a look at what those spots are.
What needs automating isn't sales itself, but what gets in the way of selling
A sale is a conversation where a person decides about money. You can't automate it entirely, and attempts to do so all end the same way: the client realises they're talking to a machine, and leaves.
But around the conversation there's work that isn't conversation at all. Replying at night with "hello, we'll sort this out." Finding out someone's sizes and deadlines. Recognising that this client will buy for three million, and that one wandered in by mistake. Writing a week later to someone who said they'd think about it. This is exactly what's worth handing to a machine.
The owner of a China delivery company described her situation like this: "The girls just sit on the phone all day — how many kilograms, what material, please send the spec." This isn't sales. It's work that people do because there's no one else to do it.
First reply
Speed of first response is the one part of a sale where you gain an advantage without any skill at all. A client writes at eleven at night, gets a clear answer within a minute, and keeps talking to you instead of whoever answers in the morning.
The owner of a gate manufacturing company puts it plainly: "Doesn't matter if it's night, day, morning, everything's on fire around — it still answers 'good afternoon.' The faster we respond to clients' enquiries, the better our chances of getting richer."
It's important not to confuse speed with a brush-off. A reply like "your enquiry has been received, a manager will contact you" is worse than silence: it consumes the client's attention and gives nothing back. What works is different — answering the actual question and asking the next one.
Handling inbound: the ones who will never buy
Any incoming flow includes people who've come to the wrong place. Not through any fault of their own — ads are broad, search is imprecise, a friend gave a recommendation. A sales rep has to talk to them politely and figure out what doesn't fit.
The owner of a gate manufacturing company put it more precisely than any textbook: "Wanting to buy a Lada Largus and submitting an enquiry at a Porsche dealership." He also noted the cost of it: "Five minutes with one client, five with another, five with a third — that's already an hour gone. And in that hour you could get plenty of other important things done."
An hour a day is twenty hours a month from a single rep. Spent on conversations that were never going to lead anywhere.
The agent does this sorting on its own, without any irritation: it clarifies the request, politely explains if it's not something we handle, and doesn't drain anyone's mood in the process.
Qualification: who to call first
Filtering out the wrong leads isn't enough. Among the rest, you need to know who matters more — and know it before a manager spends a day working through them in the order they came in.
There's usually one signal: the size of the order to come. And there's no need to ask about budget — most clients won't answer that anyway. What they say on their own is enough. "I have a house in a gated community, I need panoramic gates" — that's one category. "I have an old garage co-op, want to replace the old swing gates" — another. "I'm an architect, a project is coming up, who should I talk to" — back to the first.
The agent sorts enquiries by these markers and tags them in the CRM. From there, a simple rule applies: those with budget and timeline come first, the rest follow. Without this sorting, that order is impossible to keep — enquiries just get handled in the order they arrived.
The second, third, and fifth touch
Everyone handles the first touch. What comes next is where the money gets lost: a person asked for time to think, promised to come back, went to talk it over with their spouse — and vanished. Not because they changed their mind, but because no one followed up.
The owner of a gate manufacturing company called this out as a pain point of its own: "the pain of the second, third, fourth, and fifth touch." The owner of a delivery company put it differently: "They submit an enquiry, then we call — they don't pick up. Or we send a proposal — no answer."
Here the machine has an advantage a person doesn't: it doesn't forget, and it doesn't feel awkward. A sales rep feels uncomfortable writing for the fifth time, so they put it off and never come back to it. The agent writes calmly, on time, and to the point — and half of the "vanished" leads respond.
What happens to an enquiry inside the CRM
Sales department automation is not a messenger bot. The bot replies, but the work lives in the CRM, and if nothing reaches it, there is no automation.
Done right, it looks like this: an enquiry from any channel — the website, WhatsApp, Telegram — becomes one card, not three. The card shows what the person asked, what the agent answered, which category it falls into, and what to do next. The manager opens the card and picks up the conversation where it left off, instead of starting over from scratch.
If the CRM is still empty after the bot launches, and the conversations live somewhere else entirely — what was built isn't automation, it's an answering machine.
A plan for the day instead of a task list
The next level starts when the system looks not just at a new enquiry, but at the entire database. Any company with some history has hundreds of deals sitting in the CRM that never got followed up: someone was thinking it over, someone was waiting for a quote, someone just got lost.
Going through this backlog by hand is impossible — it's bigger than a working day. A machine can do it: every morning it goes through the database and tells you who to call today and why. Not a list of a thousand deals, but two dozen with a reason for each: here an invoice was sent and never paid, here there's been eighteen days of silence after a measurement, here a client asked something and never got an answer.
The sales rep gets a plan for the day built not from memory, but from where the money actually is.
What not to automate
The exact price. A machine can calculate it from dimensions and parameters, but quoting a final figure to a client without a measurement is a good way to end up in an argument during installation. A ballpark figure is fine — the exact number comes after the measurement.
Deadlines and discounts. These are promises the company keeps, not the agent. It doesn't negotiate them.
A conversation where the person is hesitating. When a client is unsure, they need a human — that's not a weakness of the technology, it's simply how selling expensive things works.
One more thing: don't try to pass the agent off as a person. It gets found out, and unpleasantly so. People are fine with a program answering them — what irritates them isn't the bot, it's the silence.
Where to start
Not with buying a platform, and not with choosing a model. With a single question: where in your sales department is time spent on work that isn't sales.
The answer usually turns up within ten minutes, and it looks similar for everyone: first reply, sorting incoming enquiries, follow-up touches. That's the first stage.
Next — launch the simplest version and watch it on real conversations. The first working version takes days, not months, to build, and its job isn't to impress but to show where your assumptions were wrong. Adding complexity makes sense after that, not before.
Frequently asked questions.
Where does automating a sales department begin?
With identifying where sales reps spend time on things other than selling. Usually that's the first response, handling the inbound flow, and follow-up touches. Starting by buying a platform means solving a problem you haven't defined yet.
Will the agent replace sales reps?
No. It takes on work that isn't actually selling: replying, clarifying, sorting, following up. The conversation where a person decides about money stays with a person.
How long does a rollout take?
The first working version takes days. It answers questions and qualifies the enquiry. After that, you add what real conversations show you.
Do you need to change CRM?
No, the agent is built into the one you already have. If there's no CRM at all, that's a separate job — bolting an agent onto nothing makes no sense.
What happens if the agent answers incorrectly?
It answers based on your materials and hands off everything else to a person. In the first weeks, the conversations are read through in full — that's how you find the spots where it misread a question.
How is this different from a scripted chatbot?
A scripted bot is given questions to ask; an agent is given a goal. A detailed breakdown is available in a separate article: how an AI agent differs from a chatbot.