Two questions get mixed up in most AI conversations. Where are we spending? And which number do we expect that to move?
Put both on one page and the gap shows. I use a simple grid for it, with two axes.
Down the side: who it changes
Individual: one person works faster. Team: a workflow gets better. Enterprise: your own data gives you an edge.
Across the top: how you get it
Optimize: use what you already have. Buy: purchase something specialised. Build: develop it on your own data.
What goes in the boxes
| Optimize | Buy | Build | Moves | |
|---|---|---|---|---|
| Individual | Turn on Copilot in Office 365 | GenAI models | – | Hours saved, output volume, task speed |
| Team | Enable the AI features in your CRM and ERP | Forecasting and revenue tools | – | Forecast accuracy, pipeline velocity, cycle time |
| Enterprise | Activate your vendors' ML features | Churn and pricing tools | Propensity and lead-scoring models | Win rate, churn, revenue, margin |
The tools are different in every row, and so are the outcomes. An hour saved by one person is real. It just does not show up as win rate, and unless the work around it changes, the hour tends to get absorbed by the next task.
Read it from the right
Start with the metric, not the tool. Which numbers have you been asked to move this year? Find each one in the last column. That tells you which row it lives in.
Then mark where your AI spend actually goes. I wrote “perceived” on the slide on purpose: it is where people think the investment is happening. A licence everyone has switched on is easy to count as a strategy.
Most conversations, and probably most spend, sit in the top row. Most of the numbers a leader is judged on sit in the two below it.
What the newer research says
Start in Norway. The NHO report from January 2026 found that 55% of Norwegian businesses used AI in 2025, up from 24% in 2023. About one in five are what the report calls frontrunners. They report bigger productivity gains and more often higher revenue.
Buyers are also changing what they count. In a survey of 830 IT decision-makers published in February 2026, Futurum Group found that revenue growth and profitability together are now the main measure of AI return for 21.7%. Time saved as the main measure fell from 23.8% to 18.0%. On the grid, that is a move from the top row towards the bottom.
What each way takes
I will not tell you whether to optimize, buy or build. That depends on your business. What I can do is be clear about what each one takes. The dashes in the grid are not a rule. They show where I seldom see anyone build, because something to turn on or buy usually exists.
| Why it is quick to start | What it takes to keep going | |
|---|---|---|
| Optimize | It is already paid for, in tools your people use every day. | Someone to switch it on, set it up and teach people to use it. You get what the vendor ships, when they ship it. |
| Buy | Someone has built it for exactly this job, and keeps improving it. | A subscription, and the work of connecting it to your other systems and checking it fits your data and your rules. You depend on the vendor. |
| Build | It fits your process and your data exactly, and you own it. | A team to build it, and then to keep it going: fixing what breaks, scaling it, adding what the business asks for, looking after every connection to other systems, and the documentation some rules ask for. |
Build can look like the smart move today. AI means a first version in days, and a subscription you can cancel. But the first version is the cheap part.
Take an applicant tracking system. You know your hiring process and what a great hire looks like at your company, so you build your own. At first it feels smart. Then you grow, and it has to scale. New needs appear and you build them. Something breaks and you fix it. The links to payroll, HR and onboarding need looking after. A rule like the EU AI Act arrives, and you have to document how it works and keep an audit trail.
Without choosing to, you have started a software company next to the one you actually run. There is a reason others build these tools, sell them and live off it. For them it is a full-time job, often for a whole team.
So the question is not whether you can build it. It is whether this is where you want your energy and your best people to go. For some, with data and a process nobody else has, the answer is yes. For many, it is software with AI in it, and the energy goes into the products, the processes and the customers the business is there for.
The grid in use
Here is a made-up example, so you can see how it reads. A sales leader with a team of 15. Her goal this year is a higher win rate. She has switched on Copilot for everyone, and the forecasting feature that came with the CRM.
| Where the spend is | The number it should move | What she sees | |
|---|---|---|---|
| Individual | Copilot for everyone | Hours saved | Spend and number match. It is not the win rate. |
| Team | CRM forecasting, switched on | Forecast accuracy, cycle time | Spend, but no one owns the forecast yet. |
| Enterprise | Nothing | Win rate | The number she is judged on has nothing under it. |
Two things show straight away. The number she is judged on has no spend under it, and most of what she has spent sits where it cannot move that number.
Then ask the people doing the work
She does not have to guess what belongs in the team row. The people doing the work know what steals their time: updating the CRM after every call, building the same quote by hand, chasing a handover that never arrives.
Those answers sort into four kinds of fix. What AI can take on. What to change in the process. What needs a decision from someone else. What needs more people. Each one points to a different next step, and a shorter cycle is something she can measure.
Try it on your own numbers
Write down the two or three metrics you are accountable for. Write down what you spend on AI and where it lands on the grid. Then look for two things: a metric with no spend under it, and spend with no metric above it. Then ask your team what steals their time, and let their answers decide what goes in the team row.
What each row needs underneath
The grid shows where to invest. It does not show whether you are ready to get the value. That changes a lot from one row to the next.
| Needs from your data | What changes for people | Plan for | |
|---|---|---|---|
| Individual | Little. It works on what the person types in. | Almost nothing. The same job, a bit faster. | Licences and a short how-to. |
| Team | Clean, shared records. A forecast is only as good as what went into the CRM. | How work is handed over. People go from entering data to checking it. | Licences, redesigning the workflow, training. |
| Enterprise | A lot. Data people trust, joined up across systems, with someone responsible for it. | Who decides. A pricing or churn model changes who has the final say. | Data work, ownership, governance and change support, as well as the tool. |
In the top row a bad result gets edited before it is sent, and the risk stays with one person. In the bottom row the output sets prices and priorities. If the people using it do not trust it, they will work around it.
A forecasting tool will not fix a forecast nobody owns. And expecting a bottom-row result while the budget and attention sit in the top row is a gap I see often.
So: the metric first, the investment second, and then a straight look at whether the people, the workflow and the data around it are set up to deliver.