How to Actually Land an AI Strategy

When people talk about AI strategy, they usually start with models and tools. In my experience this is the wrong end to start from. AI initiatives rarely get stuck because the technology can’t do the job. Much more often nobody has a clear mandate, the work isn’t linked to a goal the company cares about, or after six months no one is able to show what all the work was worth.

Therefore the approach I use looks at the organization first and at the technology second. It consists of five steps. I list them in order here, but in practice they overlap: you learn something about a use case which changes your view of the foundation, and then you go back one or two steps. This is what the five steps look like:

1. Get Your Mandate

Before you do anything else, make sure you have an explicit mandate, and check if it covers enough. It should include organizational approaches and the way people collaborate, and not only the purchase of technology. If an AI initiative is only allowed to touch the tech stack, but not the way the company is set up to use it, you cannot change most of the things which decide whether the initiative becomes a success.

2. Identify the Goals

Link every AI initiative to a goal of the company. This is the most useful habit of the whole approach, as it keeps your work visible and funded. If every project or product can name the company goal it supports, the people who control the budget can easily see why your AI portfolio is relevant.

AI initiatives usually serve one of a small number of goals. I work with the following six:

Efficiency: you get the same result with fewer resources, or a better result with the same resources.
Turnover: you enable or grow revenue.
Quality: you get better outcomes and fewer errors.
Agility: you can respond to change faster.
Innovation: you can do new things which weren’t possible before.
Employer branding: you are a place where skilled people want to work.

Once you have named the goal a use case serves, the use case usually becomes much clearer, and measuring the impact later on (see step 5) becomes easier, too.

3. Build the Organizational and Technical Foundation

This step gets skipped quite often, although I think it’s the most important one.

On the organizational side you need the following things:

A cross-functional core team, which consists of data and AI specialists, developers, representatives of the business domains and Enterprise Architects who understand how the parts of the company, its data and its ideas are connected. In addition you need a strong sponsor.
A decision-making body, so that the initiative has the authority it needs and everyone knows how to get approval.
AI as part of the existing project and product processes of the company, which means the way budgets get approved and work gets prioritized and governed. If AI lives in a separate process, it will stay a side project.
Visible progress: invest in internal communication, so that employees and management understand what is happening and why.

On the technical side the same topic comes up again and again: data needs to be able to flow through the company and be available in time. This means you have to decide how data is provided (centrally, in a decentralized way, or as a mix of both), make existing data structures discoverable and keep the data quality high. And the teams need more than AI use cases. They also need the technical foundation to build their own.

4. Identify and Realize Use Cases

Now you can collect use cases and decide in which order you want to realize them. Identify a few quick wins and lighthouse projects early. They should deliver value, of course, but they also keep the support of your management high while the harder and slower work is still in progress.

One thing I would add from my own experience: choose lighthouse projects which create value your customers can see, and not only internal templates. If a lighthouse project only impresses the people who built it, management will lose interest quite soon.

Let’s say the sales people of a company create their quotes while the customer is sitting in front of them, and this means about twenty minutes of clicking through the ERP software for each quote. If AI generates the quote, the sales person only has to select the customer and briefly describe the task, which takes two to three minutes, while everything else is automated. This saves more than fifteen minutes per quote, which the sales people can spend on advising the customer or on the next customer. They see this value with every single quote, and it’s easy to link it to the goals from step 2: efficiency first of all, and quality or turnover, depending on how the saved time is used. A use case like this makes a good lighthouse project.

5. Measure the Impact and Make It Transparent

Finally, show the value you created, including the parts which are a bit uncomfortable.

A simple technique works well here. For each initiative, note how much AI contributed to its result:

It would not be possible without AI.
It would be noticeably less successful without AI.
The contribution of AI is irrelevant.

I find this more useful than a single headline number, as it shows where AI makes the difference and also admits where most of the value came from somewhere else. People will question your numbers anyway, and numbers which include the less flattering results are much easier to defend when you ask for next year’s budget.

Conclusion

If you look at the five steps again, you’ll see that technology is only one part of one of them. Mandate, goals, organization, the order of your use cases and the measurement of their impact make up most of the work.

Of course you still need to take the technology seriously. But I would organize data and AI around the value the business creates first, so that they don’t end up as a separate silo next to it.