How to make an AI Strategy
When asked, most leaders say that AI will have a big strategic impact on their business. I hope you agree, because that is just how it is. AI has already snuck into almost all businesses. Sometimes unknowingly to the management but be sure that AI is somewhere in your business. So now that it’s here, and it’s going to make an impact, it’s about time for most leaders to face this new technology with consciousness and lay out a strategy.
To read this guide..
To read and use this guide to make an AI Strategy you don’t have to be a developer, a data scientist or another kind of techie. This is meant for the commercial people. The C-level, the product managers or whoever contributes to top level strategy in your organization. If that’s you, well then it’s your responsibility to make sure you have an AI strategy.
But don’t worry - making an AI strategy is not a difficult thing. Your strategy might also end up being very explicit about NOT being the frontrunner in your market. Making a strategy doesn’t imply that you have to do a lot of new investments and initiatives. It just means that you have made a conscious decision on whatever you choose to do or not to do.
In my opinion that’s all you have to do. Be explicit in your organization about whether or not you will start investing in AI and the reasoning behind the decision. If you choose to invest you should make a plan for it.
Who should have an AI strategy?
So, not all organizations have to have an AI-strategy of course. If you’re an army of one or a very small business there’s no reason to spend time on an AI strategy. That being said there’s no set amount of employees or yearly revenue that tells you it’s right to have an AI strategy.
Often the complexity of your business is usually the best tell. If your business is an ice cream shop you might not see the business case in AI going profitable. But even the ice cream shop could use AI for demand prediction. Weather, nearby events, marketing efforts, season and so on, makes the problem complex enough for AI to probably be superior to human demand prediction.
Another important point here is that the AI availability is changing and it’s changing with an amazing speed. Smaller and small businesses can make a profit on AI-investment. But why is that? There are two main reasons:
1) AI is becoming a commodity. More and more AI solutions are made by tech startups ready to use out of the box. You can get chat bots, sales support AI’s, invoice data capture and so on that already works. You just have to connect it to your existing business applications. If you choose to invest in developing your own AI, the costs here are also going down. AI development has reached a face where most of the hard work is already done and you can train your own AI on top of big general models.
2) The other main reason is the infrastructure available and as a result the data. There’s no AI without data to train on and an infrastructure to be applied on. Smaller businesses now have all the technical infrastructure needed to apply AI and the data readily available. Let’s take a look at the ice cream shop again. Nearly all ice cream shops have a point of sales system with an API that can deliver sales data, that can be joined with online weather data and so on. The ice cream shop probably also has an online accounting system that can be updated with the AI predictions. This availability of data and infrastructure used to be reserved for large corporations that invested heavily in IT.
So to conclude if AI is for your company, look at the complexity and available infrastructure if you are a part of a small business. For medium and large businesses an AI-strategy is definitely a must. If you don’t have the infrastructure and data in place, then it might be about time.
The strategy
So now for the actual strategy work.
First and foremost - An AI strategy should be a strategy to support the overall company strategy. This is probably the most important point here. I have seen AI for AI’s sake and AI being put on a completely separate track from the rest of the business. Would you do that with any other part of your business? Would you accept marketing to put all their efforts on something that wasn’t supporting the overall objectives? No one at their right mind would allow that but it is happening with AI. The reason is that AI has opportunism as a classical pitfall. With AI everything seems possible and the concept for many managers are a bit elusive. As a result the approach is often “let’s hire some AI people and see what happens”. That is the most expensive and least value-generating strategy of them all. Please stay away from that.
That being said, this guide is supposed to be as concrete as possible. I’ll try to give you as practical and ready-to-execute advice as possible. Feel free to write me feedback if you feel something is missing or should be explained better.
General features of the AI strategy
AI is special and not special at the same time. By that I mean that AI comes with a unique set of challenges and features but the solutions are usually well known concepts and disciplines.
The common challenges
I have written about common AI challenges before but in a nutshell they are here:
No AI without data: You have to have data in order to work with AI. Even if you buy the AI as a service you most likely still have to have a dataflow that is organized and available. We’ll get back to data a short while.
404 - Business case not found: Usually an important part of strategy work is to make sure the objectives can be met with a positive business case. So if you want to make a new product or target a new customer segment you must make sure there is a profit in sight. You do that by estimating costs and turnovers. With AI it’s impossible to know what it will cost to make and what impact it will make. I’ll get back to how we combat that.
Communication and people: We all know how much more press it gets when a Tesla crashes on autopilot than when people crash their cars. People see AI solutions differently than they see people. As a result you will have to communicate differently and carefully handle cultural change when implementing AI.
The Features
A very interesting feature of AI in a strategic perspective is that when looking at successful cases most focus is actually on the top line. Generating more revenue and creating more value generally have more success than those looking for cost cutting with AI. This doesn’t mean you shouldn’t be open to cost cutting but prioritizing AI for value makes sense since the payoff can be much higher.
As mentioned above the data is an important necessity for AI and here’s the secret - the data is a bigger strategic advantage than the AI development itself. I have written about it here. Basically the company that can get the highest quality data at the lowest price will be the most competitive. AI development and implementation is easy for the company with the right data. So much of your strategy must be about securing the data flow.
As mentioned the business case is impossible. You can’t plan out all AI initiatives up front and then execute. You should never decide on a specific product that works in a specific way with a given quality. What you should do instead is to set your objectives(see below) and create hypotheses on how to achieve them with AI. When you start executing these projects you should always milestone fund them so you don’t end up investing all your budget in a dead end. You can read about AI milestone funding here.
The objectives
So as mentioned you should aim to support your overall business objectives. To do that you have to set some AI-objectives. I’ll try to show this with an example.
Let’s say we have the ice cream shop. They are no longer just a shop, they have now grown into a change of about 25 ice cream shops. Let’s call it SundAI. Get it? :)
Luckily for us, SundAI has just made a new overall strategy for the company. The ambition is to grow into a large national chain of ice cream shops within 3 years. So now we can make an AI strategy supporting this.
First - let’s list the top 5 strategic objectives for SundAI:
In 3 years SundAI should have 100 shops
SundAI should be the highest customer retention(min. 5 visits per customer per year)
SundAI should be known as the chain with the most unique flavors. 10 new flavors each season
All customers at SundAI get’s same experience every time they visit. Nomatter what store
SundAI must have at least a 40% gross margin on ice cream including the cost of waste
This makes sense right? SundAI is on a mission to scale and improve or at least hold on to the quality of their product. It’s a very common mission but a hard one to pull off. Scaling fast without losing product and customer focus is difficult and many businesses fail here. Let’s see if we can come up for some ideas for AI that could help achieve this:
Image quality control: The first idea is to make an image quality control. Every time an employee is ready with an order of ice cream and puts it on a tray, a small camera takes a picture and approves the visual quality of the ice cream. The AI can’t control the inside but at least it will be able to see if it looks like it should. If the ice cream isn’t looking good a screen that only the employee can see gives feedback.
This achieves goal 2 and 4. The customers always get the same good experience. Goal number 1 is also getting some help here since new employees get more feedback.
Marketing prediction: Where I live in Copenhagen, not everyday is ice cream day. In fact there’s a few days a year where the weather is nice and everybody goes for ice cream. Ice cream vendors must be on top of their game these few days a year. So why not have an AI predict the days with the most ice sales a few days in advance so we can scale up marketing efforts right before the customers need a reminder about SundAI? We can do this with historical Point of Sales data, weather data and data from the ERP on what ice cream we have on stock. This will all help goal 1, 2 and 5 by helping profits and growth.
Production data: The last AI objective I will add here is to streamline data operations for production data. Data about what we sell when and how the costs are connected. By doing this we can later buy or develop AI’s that can help us produce cheaper and faster and produce the right ice cream. This objective is almost a given in any AI strategy. Notice how this helps goal 3 and 5 and now all the overall business goals have been helped by AI.
This process of course requires some knowledge about AI and what the possibilities and limits are. If you don’t have any insights here then you might need to get some help. Both to be inspired but also to be able to cross off the worst ideas early on.
SWOT it out
I initially wanted to do a whole SWOT for the SundAI case, but this post is already getting long. What I will say though is that doing the classic SWOT for your AI strategy is a really good idea. As AI is unknown to many and you already have so many blind sides here, stick with frameworks and approaches that you know whenever possible. SWOT is a great example of this.
Once you have been through a SWOT process it would be natural to reevaluate your objectives from this new insight.
The roadblocks
Once we know AI objectives we identify the roadblock or the challenges we have to overcome to achieve the goals.
A very common challenge in AI is change management. Introducing AI is not just difficult for management but also for all the employees whose daily lives it affects. AI can be a bit hard to grasp. It is often black boxed, so you can’t just explain how it makes decisions and why it sometimes makes obvious mistakes.
For SundAI it might be a real problem. A camera that quality controls ice cream can seem intimidating and you cannot argue with it like you can with a colleague if you disagree. The solution here could be to talk about the AI as a decision support system instead of a control system. You can also choose to make a feedback button with “I disagree” on the screen. This can both give you data but also give the user a sense of control.
The marketing prediction AI can have the same problem. If the AI predicts an important date for selling ice cream but no one buys ice cream that day , then who is to blame? And if the solution is black boxed we can’t explain or understand the mistake.
Another common challenge is getting data. I’ll get back to that in a minute.
The ressources
So what resources do you need to achieve your AI goals?
First of all it depends on whether you expect to develop AI in-house, get it developed by consultants or buy off the shelf software. All options are possible but it’s important to make a strategic decision here.
Developing AI in-house requires a lot of experts. You need data operations, data scientists, machine learning engineers and a good product manager. For SundAI that might not be worth it.
What I would do in SundAI’s case would be to hire a data operations manager to make having the right data all the time a capability of the organization. This will be an advantage for almost all possible AI projects in the future, whatever they are about. The actual AI development would be something I would try to buy either off-the-shelf or custom from outside depending on the case.
I have to underline that data operations here is a really important part of the AI strategy. No data. No AI.
Measuring success
How we measure success is of course different for each AI project, but in my experience it’s important to be clear on what success really is early in the process.
I would advise to always consider making the measurement about how much the AI really contributes to the overall business objectives.
If you need inspiration I can also recommend reading about recall, precision and accuracy here.
Ethics
In the end here I just want to mention ethics. If you ask around about AI you will probably be advised to make a lot of your AI strategy about ethics. I’m not against ethics but please don’t. Being a n ethical business should be a goal all the time, not just when you introduce AI. if more interested about this argument I read about it here: https://www.danrose.ai/blog/etik-er-ikke-relevant-for-ai
I think that was the quickndirty guide. Again let me know if you have feedback.
