[AI for Business 1] - A Roadmap for AI Implementation

Getting started with AI use cases

The AI space is booming - amidst a stream of tech news, legal grey zones, and a bunch of fears starting with job replacement and ending with doomsday scenarios, companies struggle to formulate and implement a realistic roadmap for value creation with AI. This newsletter will lay out a path that you can follow to implement AI in your business. It targets business practitioners who aim to generate value using AI, without any requirements on technical background. The content is inspired by learnings and best practices from AI beginners - companies that are just starting out on AI, as well as top performers - companies that already have implemented AI for a large part of their value creation.

In this first episode, I will outline a very simple, three-step process that you can follow to start implementing AI use cases in your company. Later episodes will elaborate on these steps, providing practical advice and enriching them with strategic considerations. Let’s dive straight into the process:

A three-step process for AI implementation

First, you identify AI use cases and formulate the basic dimensions along which you want to create value with AI. Then, you jump into the technical implementation of the solutions. Finally, you follow through and make sure that the AI solution gets adopted, used, and improved throughout the company.

Scoping value and identifying AI use cases

We will define use cases as existing business challenges that can be addressed with AI in a way that leads to measurable outcomes. For example, a use case in marketing is the application of the AI-driven generation of personalized emails. The outcomes are two-fold - beyond saving time (i.e., costs) on manual writing, the personalized character of the e-mails also improves engagement and conversion.

Before you start mapping out potential use cases, it is important to understand the various dimensions along which you want to benefit from AI. While the end game is about increasing profits and positive impact, a more granular understanding of the benefits of AI will help you focus the development and improve your communication with different stakeholders.

Let’s look at some of the main benefits of AI:

1. AI can be used to boost productivity and efficiency. This can happen on an individual level - thus, you probably already used ChatGPT to write your texts when you didn’t have enough time or patience. On a higher level, AI can make your business flow by removing friction and integrating different processes. For example, we often see a disconnect between marketing and sales departments - AI can provide the missing link by connecting the data and activities across both business functions.

2. AI can “augment” people by amplifying the tasks they can perform. Especially in the realm of creative tasks, AI helps us bridge the gap between craft and creativity. Thus, a marketer who specializes in writing can now enhance his texts with images generated by models like Midjourney. AI can also augment humans by processing huge data quantities and extracting insights to support decision-making and action.

3. Finally - and this is what we are particularly excited about - AI can help you innovate, create new revenue streams, and catapult your competitive advantage to unseen levels. That doesn’t mean that AI will take over your business - quite to the contrary: you are in the driving seat, using AI to leverage and amplify the unique and deeply human expertise you have at your company. Let’s look at pharma as an example - pharma companies like Insilico Medicine that master the art of AI-driven drug discovery can drastically shorten the ultra-slow cycles of traditional drug development (~10 years as of now). This enables them to address the long tail of thousands of rare diseases out there in the world, reducing suffering and multiplying their profitability.

So, whenever you consider a use case, try to formulate the benefits it will bring to different stakeholders. And just a hint—while increased efficiency and productivity are the low-hanging fruits many of us are after, they don’t really provide you with a competitive advantage. Top performers play their game at level 3, leaving their comfort zones and using AI to transform their businesses in the long term (cf. this survey by McKinsey).

Now, let’s look at what makes a good or bad use case for AI. Some of the good candidates are:

• Daily drudgery: Routine processes where many small decisions need to be made, such as fraud detection, customer service, and invoice processing. Automating these with AI will simply reduce human workload. Before you do this, be sure to have a plan to deal with the resources you are freeing up—otherwise, you might run into chaos and tons of resistance (cf. step 3—Integrate and adopt AI).

• Nice-to-haves: Tasks that you would like (someone) to do but don’t because you lack the time and resources—in short, tasks that are not exactly the highest on your priority list. For example, this can be a more regular routine of marketing content that you push out to your customers.

• Inspiration and research: Business environments, incl. competitors, regulations, and customers, are changing at high speed. To stay relevant over time, you need to get into the habit of adjusting and innovating on a continuous basis, even in situations of stress and uncertainty. AI can be your best friend on this journey. It can aggregate and summarize large quantities of information that you would not be able to process. Additionally, generative AI models like GPT-4 are creative folks - you can rely on them for a consistent stream of ideas, even in those stressful situations where your brain shouts fight-or-flight. Needless to say, you are still in charge when it comes to selecting and curating their suggestions.

• Personalization: Modern consumers are becoming more demanding. They expect products and services that adapt to them by addressing their individual preferences. Many B2C tech companies have already mastered this discipline—think of the personalized recommendations on YouTube, Netflix, or Amazon. For companies in other B2C sectors like healthcare, finance, and education, large-scale personalization is a promising area to generate long-term competitive advantage.

By contrast, here are some bad candidates for AI solutions:

• The One Big Decision: Don’t try to automate one-off or infrequent decisions - it is simply not worth it. Let’s say you are an energy company planning to acquire a cleantech startup - do your homework manually or delegate it to an M&A agency. By contrast, if you are a private equity investor and want to open up cleantech as a new investment arm, you can consider setting up an automated screening system since you will be flowing through the process over and over again.

• Full explainability required: In many of our decisions, there is room for gut feeling - whether it is coming from ourselves or a probabilistic AI model. However, some decisions need a clear and objective answer to the question “why?” - especially if these decisions can significantly impact people's livelihoods, as in the case of credit scoring and legal procedures. Advanced AI models cannot deliver this level of transparency. Simple, rule-based models should be preferred when you need full explainability.

Selecting and implementing AI solutions

So, you have identified several promising AI use cases in your company. Before implementing them, you need to scope your resources and evaluate their feasibility. Beyond budget, the two main resources to consider at this stage are your internal AI skills and the available data:

• Evaluate your internal technological acumen. This will help you choose between in-house development and outsourcing. Do you have a solid IT team that could take on AI tasks? Are there any specialized skills in AI or data science that are already available? In this case, you can consider developing your AI solutions in-house, which will also allow you to build up solid internal AI know-how and assets. No AI experts on your team? Don’t just write the in-house option off. If you can create the right working environment, good developers are always up for a new challenge, while AI technologies are becoming more and more accessible for engineers without specialized AI skills.

• Analyze your data situation - what data do you have to address the identified use cases? Is it already cooked up and ready to consume for AI, or is it spread all over the place? The latter situation is quite common - data practitioners spend 80% of their time finding, cleaning, and organizing data. Rather than carrying the additional burden of messy data throughout your projects and slowing down your pace, consider fixing it up-front via an internal effort or consulting project. Obviously, you will still need to make adjustments as you move forward, but these will be occasional disruptions rather than a constant drain on the time and energy of your developers.

With a clean data landscape and a lineup of skilled developers, you are ready to jump into the technical work of discovery and development. Of course, the developers will carry out the lion’s share of this work, but successful development requires skillful communication at the interface between business and technology. As anyone who has supervised or accompanied software projects of a certain complexity knows, development can turn into a rabbit hole of never-ending iterations and “improvements” that are not always aligned with business outcomes. To avoid these traps, we will be covering aspects of AI development step by step, introducing basic AI concepts, checkpoints, and practical hacks that you can use to manage the process and keep control (if you are impatient to learn more, jump directly to my book).

Following through with AI adoption

Fast-forward a couple of months - you’ve made it. You’ve implemented and deployed an AI system, maybe even the first AI system in your company. While you are sure to get some applause, recognition, and a surge of initial interest, the job is far from done. Now, you need to make sure that people in your company start using the system and feed it with data that can be used for continuous improvement. You need to understand and overcome their resistance, educate them on the right usage, and make sure that they follow through and keep using the AI. This stage is about people and emotions - managing fear, building trust, and getting people enthusiastic and optimistic about the ways in which AI can transform their work and their company.

Let’s look at the following poll by YouGov, which shows the major emotions people have about AI:

Emotions people experience in the face of AI innovations (source: YouGov research)

That’s a tough baseline to work with - most people are cautious, concerned, skeptical, and scared (and this is US data - let’s “cautiously” assume that our European colleagues are even more reserved in their AI optimism). Here are some guidelines that can help you shift the picture toward trust, hope, and excitement:

• The earlier you start, the better. If you involve stakeholders and users from the early stages on (use case identification and implementation of solutions), you give them a sense of ownership, making it easier to get buy-in and adoption. In future episodes of this newsletter, you will learn about the tools to do this, incl. user testing, discovery, and ongoing education.

• Fears around AI can turn into hidden showstoppers—identify and address them. One of the major fears in the workplace is job replacement/displacement. In many cases, this is unjustified since AI is augmenting humans rather than replacing them. In other cases, replacement is going to happen, and you are responsible for providing people with clarity about the future of their jobs—whether it involves re-skilling, a shift in responsibilities, etc.

• Build trust into the AI system. For many people, AI is the new kid on the block - they don’t know how it works and don’t have a successful track record of helpful and correct AI outcomes. As Stephen M. R. Covey puts it, “nothing is as fast as the speed of trust” (from The Speed of Trust) - once people overcome fear and start trusting your system, adoption follows. Trust is built by ensuring high quality and accuracy, but also by providing transparency into how the system works and what its current capabilities and limitations are.

• Manage expectations - educate people on the power, but (and especially) also the limitations of the AI system. Make them aware of what AI can’t do, and make sure they understand when they need to intervene by curating, editing, or rejecting its outputs. Wrong AI results that go unnoticed and end up in presentations, communications, or decisions are a major trust killer and will quickly scare users away from your system. As an example of failed expectation management, some of you might remember Microsoft’s early AI assistant Clippy. This cute paperclip widget would offer help in tasks clearly out of its domain of competence. After several attempts to establish Clippy, it was discontinued, and Microsoft switched back to a more understated approach to AI.

• Empower people to co-create - as users and consumers of AI, we have more power than we think. We are not only feeding AI systems with data that will be used for further fine-tuning and optimization. Rather, we can also shape the whole human-AI interaction by making it clear where we need AI help, where we want to be in charge, and what a helpful output from the AI looks like. When driving AI initiatives in your company, set up feedback processes so users can share this information and integrate it in the next iterations of the AI.

With that, we are at the end of our simple three-step process for AI implementation. By following this process, you can ensure that your AI initiatives are successfully implemented and adopted in your company, ultimately fulfilling their value promise. In future episodes of the newsletter, we will drill down into each of the steps, illustrate them with examples and case studies, and provide you with the tools to address the related challenges.

I hope you enjoyed the read. If you have feedback to share, topics that you would like to cover, or interesting AI stories with lessons learned, get in touch!

Best wishes

Janna

---

If you would like to learn more about my work:

• For my friends in Paris, I will be giving a presentation at ProductTank Paris on May 14th about the user experience of AI products.

• Check out The Art of AI Product Development, my upcoming book about value creation with AI.

• Read Building AI Products with a Holistic Mental Model for a preview of the different components to be considered when designing an AI system.