By Urban Gavelin, , AI & Sales Leadership
Add AI to Chaos and You Get More Chaos
There is a quiet hope in almost every AI project I see right now: if AI does not help us, at least it will not do any harm. Unfortunately, that is not how it works.
The Comforting Myth
If the foundation is missing and you pile on more technology, more process and more AI, the result does not stay the same. It gets worse. You are not just automating what works. You are also automating the mess.
Most leaders assume AI is a neutral layer that sits on top of whatever exists underneath. It is not. It reacts to what it finds, and if what it finds is chaos, it turns that chaos into something that looks organised.
AI Actually Gives You an Answer
An AI model that reads half-filled data does not produce nothing. It produces something that looks like an answer. That is where the danger lies. An empty report is fairly easy to ignore. A polished report with conclusions, charts and three decimal points looks far more credible, even when the input was garbage.
I have seen forecasts get worse after automation. Not because the system calculated incorrectly, but because it calculated quickly on numbers that no one scrutinised properly anymore. Previously, an experienced sales manager could look at a forecast and say, that does not feel right. When the same forecast comes from a system with three decimal points, we suddenly start trusting the table more than reality.
The same happens with AI meeting summaries. If no one asked the important question, or the customer never explained why the change matters, you simply get a well-written summary of a fairly poor meeting. Then it gets filed in the CRM as if it were solid ground.
The Order Is Not Negotiable
Preparation before the meeting. Process before enrichment. Data before AI. That is not three pieces of good advice. It is a sequence where every step depends on the one before it.
I learned this long before anyone talked about generative AI. In the late nineties I sat in an important meeting with a large Danish retail chain. We were selling digital cameras and I really wanted to win the deal. I knew the products, the range and the arguments. The problem was that I had not researched the customer's purchasing strategy well enough.
When I finally sat across from the buyer, I could not connect my range to what they actually needed. The buyer was tough and I went home empty-handed. It became a useful lesson: preparation is not something that comes before the work. Preparation is part of the work. I could have had twice as many cameras in the range and it still would not have saved that meeting, because the foundation was missing. An AI tool on top of a CRM that no one fills in properly is essentially the same meeting. Just in system form.
Leverage Works Both Ways
Open your ten most recently won deals in the CRM and read them as if you started at the company today. Try to understand why the customer said yes, what tipped the decision, and who actually made the call. How many of the ten can you understand based purely on what is written in the system?
If the answer is fewer than seven, I would be cautious about adding more AI on top. You might have a CRM and a nice dashboard, but you do not necessarily have a foundation. The good news is that this is fixable, and often far cheaper than the next AI pilot. I would rather have three important fields filled in every single time than ten brilliant fields filled in only sometimes.
AI is genuinely excellent at amplifying. The only question is what you are amplifying. Good data, good processes and good behaviours give you leverage in the right direction. Disorder, bad habits and half-filled information give you leverage too. Just in the wrong direction. So bring one question to your next leadership meeting: if you switched off all your AI tools tomorrow, which decisions would actually get worse? The answer probably says more about your AI maturity than how many AI tools you have bought.
Frequently asked questions
What is the biggest risk of adding AI on top of messy CRM data?
AI does not stay neutral when the underlying data is poor, it produces confident-looking answers regardless of whether the input is solid or not. That false confidence is often more dangerous than an obviously broken report, because people start trusting the polished output over their own judgement.
How can leaders quickly test whether their CRM data is ready for AI?
Pick the ten most recently won deals and read them as if you were new to the company, trying to understand why the customer bought and who made the decision. If you cannot reconstruct that story from at least seven of the ten records, the data foundation needs work before adding more AI.