In a previous Unit8 Talks webinar, we outlined our AI project selection methodology based on more than 60 successful AI and machine learning projects that we have delivered for our customers. Today, we would like to approach the problem from a different stance and look at how companies adopt AI at different maturity levels.
We believe that with such a strategic overview, firms can anticipate future steps while learning from best practices to avoid common pitfalls. Whether you are simply thinking about getting started with AI or you are feeling that your efforts are plateauing, the insights presented here will likely resonate with you and support your growth.
At Unit8, we advise companies to view their AI and data journey in five stages, from gaining awareness to leveraging the full potential of AI and lock in its associated value.
Adopting AI across your organization and creating value from it is no easy task: industry research firm Gartner estimates that about 80% of AI projects never make it to deployment and hence fail at making any impact. Based on our experience at Unit8, we have outlined a typical adoption path that we have seen numerous times in practice. We will share with you the key activities associated with each of the five stages, as well as some best practices. Yet, it does not mean that every organization necessarily follows exactly these five stages: this framework can easily be adapted to your specific starting points or requirements.
The very first step into your AI journey is to expand your knowledge on the topic. It is a crucial step as the field is quite noisy and constantly evolving. It is then easy to either feel overwhelmed by information or fall into the AI hype trap. Here, we recommend starting with what you know: you understand your business more than anyone else, so focus on how AI is used within your domain by reading whitepapers or attending industry events. With that initial knowledge in mind, you can start discussing it internally with your peers, and establish a first long list of use cases in which you see potential for AI-driven value creation. Finally, make sure to conduct a thorough assessment of the resources at your disposal (i.e. talent, infrastructure, budget) before even beginning to draft a basic data & AI roadmap.
We will elaborate on how to draw your AI roadmap in a future article, but in a nutshell imagine it as a series of guiding principles and actionable steps you should take in order to leverage AI at scale in your organization. The roadmap is set to evolve over time as companies get more mature with AI, yet it usually encompasses 5 key areas: aligned business, data and AI strategies; people & capabilities; AI governance and ethics; value realization at scale; and continuous engineering. These 5 areas aim for the same goals, namely ensuring that AI gets past the proof-of-concept stage, becomes trusted across your organization and delivers on critical elements of your business strategy.
“Always start with what you know best: avoid the noise and focus on concrete AI applications within your industry”
We recommend a few best practices to increase your chance to succeed. First, building knowledgeable internal AI champions early on in your organization will create excitement and momentum and at the same time limit any dangerous misconceptions on what AI can do or cannot do. Second, AI initiatives will not last long without proper leadership commitment, so you need to ensure C-level buy-in before moving to more complex phases. Finally, considering concrete actions to increase data literacy early on within your organization will prove very valuable in the long-run.
Once you have gained a better understanding of AI use cases within your industry and assessed your internal capabilities, it is time to experiment. But hold your horses before jumping straight into tons of projects! Do not lose sight of your goal at this stage and concentrate your efforts on building momentum with small projects that have a measurable business impact. The key here is again to focus: take your original list of ideas, and narrow it down to a maximum of 2-3 projects for which you envisaged the highest potential in terms of both business impact and technological feasibility (see methodology outlined in one of our previous articles). Then, give these pilots to a small team to move forward with them, or consider external partners.
“Be disciplined and prioritize small projects that have a measurable financial impact.”
Excitement will likely be high at this stage as you start solving problems and optimizing processes – do not let it drop because of lengthy projects with low chances of success. Instead, focus on low-hanging fruits in an agile manner, and do not be afraid of failing fast if needed. Another tip that will pay off in the long run as projects complexify: pay attention to data quality and availability, and start improving accordingly if needed.
By now, you should have a few pilots on your hands that have successfully demonstrated added value, but their scope is likely still limited. The next logical stage is to challenge all these initial quick-wins and assess their impact on a larger scale. This shift from pilot to production is usually difficult, and not only for technical reasons: Budgets will likely increase, models might lack explainability and acceptance by non-technical end-users will not be as easy as expected. To facilitate the transition and gain broader acceptance, it is crucial to secure a direct executive sponsor (and budget) while investing time into training end-users with a focus on explainability to avoid the typical black-box effect.
“Scaling AI projects from pilots to production is no easy task, yet efforts will quickly pay off as impactful solutions get adopted.”
You will quickly see whether your solution is attractive if you monitor its usage – doing so will also help you identify gaps for scaling further, and build better solutions in the future. In the same vein, motivation to move forward will likely grow as you will start witnessing the financial impact on your top & bottom line.
At this stage of maturity, you must have successfully proven that you can develop impactful AI solutions and deploy them to a larger scale. Yet, as your number of use cases increases and your solutions improve, you will likely lack storage capacity, computing power, or other advanced features. It might be a good time to finalize your AI strategy and develop your data & AI infrastructure. Concretely, this step requires you to strengthen your data foundations by moving to the cloud, investing in on-prem servers, or opting for hybrid options.
“No AI strategy will be sustainable without solid data foundations. Do not gloss over them and ensure that you have strong internal capabilities in place.”
Additionally, do not lose momentum and keep pushing for democratization of data usage across your organization: as you become more mature, make sure to gather feedback from your internal clients (both creators and consumers of AI use cases). There, “translators” can play a great role as an interface between tech and business people to add features or adapt your solutions. On a side note, we recommend paying particular attention to the “mushrooming” of AI activities. By this we mean pilot projects and little AI teams that pop-up anywhere in the organization without central oversight or governance. This will create inefficiencies and hard-to-break silos in the long-run.
AI benefits have no more secrets for you at this stage. Your data teams are growing and many use cases are running at a large scale across your organization. It is now time to consolidate these gains and lock in the value created by your solutions. Setting up an AI Center of Excellence (CoE) could be an important next milestone, primarily to bundle know-how and improve governance. This AI CoE will be responsible for maintaining infrastructure and platform operations, implementing data and model governance standards to mitigate risk, and finally improving model monitoring and reusability with MLOps practices.
“Implementing AI governance and MLOps processes is the final step to lock-in value and make AI easily accessible across your organization.”
To reach these goals, we first recommend setting clear data ownership guidelines, as well as precise roles and responsibilities for the entire model lifecycle. On a similar note, implementing a model registration & validation process will improve model reusability and explainability and also mitigate risks that could arise from your AI activities such as sub-optimal business decisions and legal and reputational damage. A successfully established CoE can become a powerful sparring partner and quality manager for all AI activities in your organization.
The business value created by AI is now such that it can no longer be considered a ‘nice to have’ by organizations. Yet, the path to capturing that value is more of a bumpy road than a sunny-day scenario. Therefore, this article aimed at giving you an overview of a typical AI journey, alongside actionable best practices.
For instance, we encourage you to be pragmatic and, at first, focus on projects that can be delivered quickly, have high chances of success and that can bring real business value to your organization. Similarly, at any stage of your journey, you will maximize impact when your domain experts work hand in hand with AI specialists.
However, nothing can replace practical experience. We would hence always recommend you to dream big and start small at first by focusing on quick-wins, and then get support from experienced partners when roadblocks are limiting your progress.
Leave us your contact details, so we can discuss your challenges and show you how to apply our Data & AI methodology to your use cases.