Does your company need AI? How to tell

I love this question. It comes up in almost every early conversation I have with a new client, and my answer isn’t always yes.
I first wrote this in 2019, about data science. Today almost every company already uses AI somewhere: nearly nine in ten, according to McKinsey’s State of AI survey from August 2026. The real question is whether AI will change a metric that matters to your business. Only 37% of companies in that survey could point to any effect on profit.
Signs AI will pay off
Our ears perk up when we hear things like:
- “We have a complicated process for this, and people do it by hand every week.”
- “Our product isn’t as smart as it should be.”
- “We don’t know much about our customers.”
- “Does this really need to be done this way?”
Another common sign is a set of rules that keeps growing. Many products run on a tree of “if this, then that” logic that someone updates every time there’s a new idea. Eventually it gets too complex to maintain. A model can replace that tree with something fitted to your actual customers, and it can update itself as new data comes in.
AI also helps you understand why things happen. Say some customers come back for a second purchase and some don’t. Is it their age, where they came from, or which offer they clicked? With a handful of customers you could work that out by hand. With millions of customers and dozens of fields, you need a model.
Signs it’s too early
AI works on top of a good product. It’s rarely a product on its own. I’ve lost count of the times someone asked me to “make our product ten times better with all that data.”
Sometimes a small product change does more than any model. If you sell online and don’t offer an add-on at checkout, adding one will help far more than tuning which add-on to show. Once the basics are in place, AI makes them better.
It may also be too early if:
- You aren’t yet collecting the data you’d need. You can’t predict car color from shoe size.
- You can’t calculate the metric you care about from the data you have.
- Nobody on your team has time to give an outside team access and answer its questions.
Where you are matters
Early companies that know they’ll need AI later should talk to someone experienced now. A short conversation can make sure you collect the right data from the start, and that you know how long to collect it before a model is worth building.
Growing companies with real products and real data, still running on simple rules, usually have quick wins available.
Mature companies with no AI team, or one that’s struggling, can see the biggest results. These projects take longer, because established teams can be wary of new tools and of outsiders. I cover how to handle that in how to choose an AI consulting firm.
If you’re not ready, start simple
Say you sell 1,000 products to 100,000 customers and want to recommend products. Before any machine learning, try this:
- Count how many of each product sold and recommend the most popular ones. You may be surprised how well this works.
- Count only recent sales, so last season’s favorites don’t crowd out this season’s.
- Count separately for each group of customers, such as your United States and European customers.
These simple approaches have a big hidden benefit. They force you to build the pipeline that moves data from your systems to the model and back to your customers. Once that pipeline exists, swapping in a smarter model is a small job. They’re also easy to understand and easy to fix. I go further into this in how popularity can kill personalization.
How to find out for sure
The fastest way to know is to measure. Pick one metric, check that you can calculate it from your data, and estimate what improving it would be worth. That’s what our Discovery does in ten business days, for a fixed fee, before anyone builds anything. You can read the steps in how we work with clients.
If you decide to go ahead, the five reasons AI projects fail to pay off are worth reading first.
Talk to us
If you’re still not sure, the first call with one of our senior engineers is free and takes 30 minutes. We’ll tell you plainly whether AI is worth it for you right now.