A recent study shows that only 2 out of 10 AI projects are successful. It is important for business leaders to understand that the transformation that comes with AI is not only technological, but will mainly impact different levels of the organization and projects, mainly in teams, tools and their execution. In this post, discover some recommendations to accelerate AI projects, make it mature and successful.
Adoption of AI by organizations
The adoption of AI by organizations implies a transformation of working methods, leadership models and the corporate culture itself. This creates resistance to change in teams and a high level of uncertainty about how this technology may threaten or reshape their jobs.
Without this organizational transformation, 87% of AI and Big Data projects fail and are never implemented, according to several recently published studies.
In fact, a Harvard Business Review article published at the end of 2023 points out that 8 out of 10 AI adoption projects fail due to leadership issues, lack of certainty about the expected results, or the implementation of models that do not apply to the local needs of the business.
The size of the global AI market is expected to grow at a compound annual growth rate (CAGR) of 36.8% over the forecast period, reaching $1,345 billion by 2030. Whether these numbers are realized will depend not only on investments, but also on the experience that executives can gain in implementing and accelerating them.
Knowing how to effectively implement one or more AI projects in the organization is part of its maturity, development, and ultimately, leveraging it to grow.
In a volatile and uncertain context in which many companies want to develop AI projects, the risk of not getting out of the Proof of Concept (POC), of spending more budget than planned, or of not using the appropriate methodologies are common cases that generate obstacles to the generalized process of AI adoption.
The Three T’s: Keys to Accelerate AI Projects
If we want to accelerate or streamline AI software project processes, this acceleration mainly revolves around three axes or three T’s as a continuous improvement framework.
- Team
- Team Maturity: It takes time, training, and a lot of human effort to have a mature or solid team to run the business. The ideal is to be able to build a high performance team within the company, but if this is not possible, an alternative is to outsource it. On the other hand, sometimes it is better to have a team of more junior people with some years of experience, well motivated and managed, than to look for a more senior team that is difficult to pay and manage.
- Motivation is key: having a happy, integrated team that gets along well, having a work culture that is open to change and innovation is one of the pillars of this topic, even more important than the necessary budget or material compensation to make the team work.
- Using similar techniques for all AI models: is very important and also depends on the predominant or most specialized skills in the team. For example, if several people come from the field of computer vision or deep learning, it is probably a good idea to do several projects together grouped by these skills rather than by others. Obviously, specialization in technique and modeling helps to speed things up.
- Stakeholders are also part of the team: Undoubtedly, stakeholders are also part of the team, and their speed in definition and acceptance is key to the adoption and acceleration of AI projects.
- Tools
- Mature cloud platform available: Having a mature platform helps the process. The full stack from Google, Microsoft or Amazon means that nothing has to be done by hand.
- End-to-end platform: These platforms allow you to manage and control all stages of the process. It is not just about being able to help people, but also about being able to standardize processes, which helps speed up projects.
- MLOps: MLOps tools solve the most critical or time-consuming problems in the organization. These include model deployment, model storage, versioning, training trial logging, infrastructure access, and SCM.
- Well-governed data lake: Having curated, organized and structured data for each project is one of the keys to deploying AI and standardizing it at different levels and throughout its lifecycle. This allows data to be used effectively for decision making. 3) Tasks/Practices:
- Tasks/Practices:
- Create base models and replicate them in the organization: this is the case of the typical model that applies to one sector of the company, but then we copy, paste and adapt it to another sector regionally. When these base models are developed in the organization, they are easier to implement than creating a global model for the entire organization.
- Raw data engineering separates from the rest of the model: from a manufacturing and mass production point of view, we have people who prepare the pizza dough and people who stretch the dough, put the sauce and cheese on it, and put it in the oven. And these people need to have completely different goals, they do not need to be so close because that creates a certain dependency that interferes with the development of the project. And at the same time you need to have the data sources collected and available and the feature engineering. Going back to the pizza metaphor, we need to think of prepared data as a key ingredient for models that can be used for many models, not just as a slice of the model. While it may seem that the dough itself is useless, that the only thing that is useful is the already prepared pizza, it helps a lot to have a team dedicated to getting all those pre-models ready.
- Much mature the methodology: it is necessary to adopt a standard methodology and be consistent in order to improve and optimize it. At the level of methodology, a lot can be done with common methodologies such as Knowledge Discovery in Data Bases (KDD); beyond what we use, the key is to achieve agility and maturity in the projects.
- Separate the incubation of ideas from the project itself: it is important that the team in charge of the proof of concept is separate from the one that delivers the model. Sometimes the POC can be a simple idea that stays in a notebook and does not go into production. But conceptually it has to be outside, it is not a model or a machine learning project. Because sometimes we ask too much of the POC and the key is to determine if it can really be turned into a model.
- Work on KPIs that can be moved with an AI model: it is important from the project and sponsorship side to start working on the KPIs. It can happen that we have a KPI and the model adapts to the KPI, or we have the model and the KPI adapts to the model. However, sometimes there are KPIs that are not of sufficient interest from the business side, it can be from the risk of incidents in hygiene and safety to the retention of talent in human resources.
Let’s say we have a system with an HR chatbot. What is the HR KPI we want to move? Improve retention.
Maybe it is a formula where the model only reduces complaints because ultimately with the chatbot they ask more questions and feel more supported. But there is no direct correlation between complaints and people leaving the company. And it is difficult to establish.
But clearly we have to start with the fact that the model reduces the number of complaints to HR, so we can understand how it works and look for correlations in those outputs of the model. The only thing that is obvious is that we are not working on these relationships between the variables of the KPIs of the models. So, if we want to speed up projects, we have to work deeply on this and not just leave it as a formula whose inputs and outputs we do not fully understand.
In the end, the dynamic and synergistic relationship between the team, the tools and the execution of the practices (tasks) can make the development and the result of an AI project in your company much more effective.
7Puentes: The Master Key to Accelerating AI Projects
With more than 15 years of experience and more than 100 projects in leading companies with high-level and high-performance teams, we have a deep understanding of the new challenges of accelerating AI projects.
Do you want to know how to adopt, mature and accelerate your AI projects? At 7Puentes, we can help you on all fronts: selecting and evaluating tools, building and developing a high-level team, adopting practices and standards to accelerate enterprise projects and align them with the business model. If you need the right advice, we have the expertise to help you succeed.
