Five AI applications that actually pay off in sales

Nine in ten sales teams work with AI agents or expect to within the next two years. Fifty-four per cent use them today. Eighty-eight per cent say AI makes them more productive, ninety-one per cent that it helps with sales planning.

Those are impressive numbers, and they say almost nothing.

Because they do not answer the question that matters day to day: where does it actually pay off, and where do you burn time on something that looks good and changes nothing.

I have worked through this over the past two years. Some of it worked, some did not. What follows are five applications that stayed, and the reason why these of all things.

Why the discussion is usually held in the wrong place

When I talk to sales people about AI, we almost always end up at the conversation itself. How does the technology help in the meeting, what objection handling does it suggest, how does it improve the close rate.

That is the most interesting part of the work. It is also the part where AI helps least. I have written about why elsewhere: the judgement about whether to raise a topic or better not forms in the first two minutes, out of tone, hesitation and what somebody does not say. There is no model for that.

The return lies somewhere else. It lies in what happens before and after the conversation.

Studies put the share of time salespeople spend writing, sending and answering emails at up to thirty per cent. Another piece of research finds that handling the inbox alone eats up to a quarter of the working week. That is a considerable part of a sales week in which nothing is sold.

The five applications that follow sit exactly there. They are not sorted by technology but by the path an enquiry takes through the organisation: from intake through approach and quote to processing and the installed base.

---

I. Intake: lead rating and routing

Effect: no lead is left lying around because it was unclear who owned it.

Enquiries come in from every direction. Through the website, through vendors, through events and webinars, through marketing campaigns. In a mid-sized organisation that quickly adds up to a four-digit number per month.

Between arrival and handling there is a manual step. Somebody has to look at who this is, how big the company is and who would be responsible. That sounds like two minutes. At a thousand enquiries a month it is a full-time job.

That can be automated, in three steps.

First, the company’s headcount is determined for every lead. In the licence business that is the decisive figure, because more employees mean more devices and therefore more licences. From that an A/B/C rating emerges. Alternatively you can check whether the expected enquiry volume exceeds a certain threshold, which helps above all with smaller companies that have large requirements.

Then the postcode determines the region. And finally the lead lands automatically with the role responsible for it.

The real gain is not the time saved. It is that nothing is left lying around. The classic lost lead is rarely the one somebody handled badly. It is the one nobody handled, because it was unclear who owned it.

What you need for it: a clean definition of what an A lead is. Without that you only automate your own vagueness, and at speed.

---

II. Approach: briefings before the meeting

Effect: the half hour of research before every meeting disappears.

Before every larger meeting there is the same half hour. Look up the company, skim the annual report, who is in the room, what do they do, what has the company said publicly of late. Is there a tender, an announcement, an occasion. Not demanding work, but work you cannot skip without sitting there unprepared.

This preparation can largely be handed over. From a well-kept knowledge base come briefings containing four things: the structure of the buying centre, the previous history with the customer, cross- and upsell potential, and the competitive situation.

The difference to doing it by hand is not only time. It is completeness. A person under time pressure looks at whatever comes to mind. A briefing looks at what has been defined.

The limit is clear all the same. The AI supplies the information that a company is currently building a plant in Eastern Europe. It does not supply the question of whether the person opposite is proud of it or whether it is a political issue. Whether the project carries their career or costs them sleep. That judgement stays with the human, and automating the rest does not make it less important but more.

What you need for it: a knowledge base that is maintained. A briefing from a neglected CRM is a well-phrased false statement, and that is more dangerous than no preparation at all.

---

III. Quote: producing the offer

Effect: the effort per quote drops to a fraction.

A quote consists largely of repetition. Line items, licence metrics, price tiers, standard text, payment terms, legal notes. The individual share is considerably smaller than the time spent on it suggests.

Anyone who lets the document be assembled from the existing data cuts the effort per quote to a fraction. It is the application with the clearest and fastest visible return of all five.

And it is at the same time the one I would watch most closely. A quote that comes into being in thirty seconds also goes out in thirty seconds. If the time gained flows only into more quotes, you have a volume effect and nothing else. At least part of it has to flow into checking.

The second effect is often overlooked: quotes become more consistent. If five salespeople offer the same product in five different formulations, the customer’s procurement team notices at the latest when they put two of them side by side.

What you need for it: clean pricing logic in the system. And somebody who actually reads the result before it is sent.

---

IV. Processing: orders and returns

Effect: work nobody misses once it is gone.

The most invisible item in sales. Order processing, order confirmations, queries about delivery status, corrections. None of it appears in a target agreement, nobody is praised for it, and yet across an entire inside sales team it adds up to several working days a week.

Plus a small thing with surprising effect: voicemail messages can be transcribed automatically and arrive as text in the inbox. That sounds like nothing. In practice it means nobody listens to a message three times any more to write down a number, and a callback does not get left because it was hidden in a mailbox.

This application makes no headline. It gives back hours that nobody had booked as a cost before, because they appear as a line item in no calculation. Which is exactly why it is usually skipped when companies think about AI.

What you need for it: the willingness to look at processes nobody finds exciting.

---

V. Installed base: customer communication

Effect: more time for the conversations where it really counts.

At a five-digit number of email interactions a year, the quality of standard communication decides a considerable part of the customer experience. At the same time it is the work that most reliably gets left undone, because nobody enjoys it and because it only becomes noticeable once it is missing.

Standard texts that adapt to the specific case instead of coming out of a rigid block take away a large part of that load. The answer becomes faster and more personal at the same time, which sounds contradictory at first and works in practice.

The real lever, though, is following up. Most existing customers are not lost because somebody made a mistake, but because nobody got in touch for months. When the system recognises where there is movement and where there is silence, following up becomes a process instead of a question of memory.

What you need for it: clear guard rails for the tone. Otherwise every salesperson writes in a different voice, and the brand frays.

---

The pattern behind all five

When I lay the five side by side, one thing stands out. None of them sells.

They cover the path from intake to the existing customer, but not one of them reaches into the conversation where the close happens. All five clear away work that sits before and after.

And all five share the same three properties. They are repeatable. They have a clear structure. And their result can be checked. That is exactly the profile where this technology is strong.

The conversation itself has none of these traits. It is one-off, unstructured, and its result is a judgement. That is why AI works so poorly there, and why I consider the discussion about it the least productive in the whole field.

The more preparation is automated, the more the close is decided by the part that cannot be automated.

What I underestimated

Two things are harder than they sound.

The first is the data foundation. Every one of these applications is only as good as what sits in the system. Automate an unclean CRM and you get faster mistakes. A considerable part of the work lies before the actual AI rollout, and that part appears in no product demo.

The second is acceptance. The technology is the easier part. Convincing people that a tool does not take their work away but the least popular part of it takes considerably longer. And it only works if you explain why you are doing it, instead of rolling it out and hoping for enthusiasm.

There is also a point that is rarely mentioned. You have to teach people to distrust the results. AI-generated text always looks good. It is fluent, well structured and sometimes simply wrong. Pass it on unchecked and you look assured while being off the mark, and that is a worse combination than unprepared and honest.

Where I would start

If somebody began from scratch tomorrow, I would not start with the most exciting use case but with the dullest.

Take the process everybody has complained about for years and nobody takes care of. It usually has exactly the three properties above: frequent, uniform, checkable result. That is the best entry point there is, and it has the pleasant side effect that nobody defends it.

The biggest mistake is the broad approach. Declare a company-wide AI strategy and you produce pilots that never reach production. Take one concrete, named problem and you actually make progress.

In the end it is less a technology question than an old sales question: where does the time go, and what would you rather have spent it on.

Which process at your company would nobody miss if it ran by itself tomorrow?