AI should multiply intelligence, not activity

There are two ways to use a lever. You can use it to lift something heavier, or you can use it to lift the same thing faster and more often.

Almost every business I look at has chosen the second one, and most of them do not realise they made a choice at all.

The pattern is easy to spot. AI arrives, and within a month the output goes up. More emails sent. More LinkedIn messages. More blog posts, more variations, more sequences, more touchpoints. The dashboards look tremendous. Activity is up forty percent and the cost per unit of activity has collapsed.

Then you look at revenue and nothing has moved.

That is not an AI failure. That is what happens when you multiply activity in a system that was already converting badly. You have not fixed anything. You have made the same mistake at a much higher speed.

AI should multiply intelligence, not simply multiply activity.

Volume was never the constraint

Here is the thing worth sitting with. If more emails were the answer, the answer was available to you before AI existed. You could have hired three people to send them. You did not, because you knew, somewhere, that it would not work.

AI did not change that judgement. It just made the bad option cheap enough to stop arguing with.

And the market has adjusted faster than anyone expected. Your prospect's inbox is now being hit by the same tooling you are using, by every competitor you have, with roughly the same phrasing and the same fake personalisation about their recent post. The volume play was a real edge for about eighteen months. It has since become the baseline, which is another way of saying it has become worthless.

What has not become worthless is being genuinely good at the moments that matter. And that is where the same tools have barely been pointed.

Three places to point it instead

Response. Not more responses. Better and faster ones.

When an enquiry comes in, the difference between a reply in four minutes and a reply the next morning is not a percentage point, it is most of the deal. AI is extremely good at this and it is not being used for it. Watching an inbox, recognising intent, drafting a specific and relevant first reply within seconds, routing it to the right person with the context already assembled. That is the highest-return use of the technology in most businesses and it is almost never the first thing anyone builds.

Follow-up. This is the one that pays for everything else.

People stop following up for human reasons. It feels like pestering. They forget. They have a bad week. They decide, without evidence, that the prospect has gone quiet because the answer is no. A system does not have those feelings. It remembers what was discussed on the third call, notices that the renewal date they mentioned in passing is now six weeks away, and puts a genuinely relevant reason to make contact in front of a human being.

Note the phrase: in front of a human being. The machine should be doing the remembering and the drafting. The judgement about whether to send it stays with you.

Decisions. The quietest one, and the one with the longest tail.

Most founders run their business on a mixture of the numbers they look at and the impressions they have formed. The impressions are usually wrong in a specific direction, because they are built from the deals that were memorable rather than the deals that were typical. You remember the client who haggled. You do not remember the eleven who did not.

Put your last two years of closed and lost deals in front of a decent model and ask it what actually predicts a win. Deal size. Source. Time to first response. Number of stakeholders. Which industry. Whether the first meeting was a demo or a diagnosis. The answers are frequently uncomfortable and almost always more useful than the forecast.

The rule I keep coming back to

If AI is producing something that a human then has to check, you have automated the easy half and kept the expensive half.

If AI is producing something that makes a human better at a decision, you have done it the right way round.

More emails need checking. A ranked list of which twelve prospects in your dormant database are most likely to buy this quarter does not need checking, it needs acting on. One creates work. The other removes it.

A warning worth taking seriously

Be careful about what you are about to amplify.

I said earlier that you can multiply activity in a leaking system and end up worse. I want to be more specific about why, because it is not just wasted spend.

Reputation moves in one direction. If you put a high-volume, low-relevance sequence in front of your market, you do not get a neutral result. You get a market that has now categorised you. Those people are not neutral prospects afterwards. They are prospects who have already decided what you are, and you will not get the chance to correct it.

The businesses that will be in a strong position in two years are not the ones who sent the most. They are the ones who were noticeably better to deal with at the four or five moments where a buyer was actually paying attention, and who used the technology to be reliably good at those moments rather than occasionally good at all of them.

More was never the point. Better was always the point.

AI has just made better a lot cheaper than it used to be, and most people are spending the savings on noise.