7 Key Steps to AI
The increased interest in AI from the business world motivated by success cases derived directly from AI applications, improvements of AI algorithms and the exponential data availability, has not yet been reflected in the massive adoption of AI in business. According to a report issued by The Boston Consulting Group and MIT Sloan Management Review, only about 20% of companies have incorporated AI in some offerings or processes, and only 5% of companies have extensively done so. Moreover, about 60% of all companies don’t have an AI strategy at all in place.
For a successful implementation of AI in business, the following 7 steps are of paramount importance. Each one contributes to the mitigation of foreseen problems on the AI journey.
1. Identify Problems to Solve
Identify where AI can solve problems in your company or extend capabilities to your products and services and translate each one to a specific use case with clear and verifiable business value.
The focus should be on use cases, keeping the connection tight between the problem and the solution, provided by AI technologies. A few of many possible examples are: improve customer experience with a chatbot or a recommendation system; automate straightforward and daily tasks with image classification algorithms; avoid fraudulent use of your services or products with fraud detection; automatically identification of defects in the production process.
2. Prioritize Concrete Value
Probably several use cases for AI implementation were identified in the previous step, to the point that it is easy to get lost with all the AI potential to solve problems. But it is crucial to keep the focus on the business value and for that, the final decision should be taken from a priority list of use cases with concrete business value.
One methodology to build that priority list is to score the possibilities with two dimensions, business potential or financial value, and feasibility. The business owners are best positioned to evaluate the first dimension, the feasibility should be assessed jointly with AI experts. This gets us to the next step.
3. Assess Current Capabilities and Future Needs
For rolling out AI in a business, a proper assessment of the existing capabilities and the ones necessary is fundamental. The required technologies for AI are quite recent and demanding, especially in the development phase where many iterations of all processes are necessary. Addressing the capability gaps across all phases of the data pipeline and evaluate the ability to achieve the needed capabilities within a given time frame is critical for the planning of the project.
Some of the technological aspects that can be relevant are: interoperability of open source software; fast and optimized data storage; networking and processing power.
4. Ensure Data Quantity and Quality
An essential step to the AI journey based on machine learning algorithms is to assess the data quantity and data quality, both are important. Quantity is important for the more performant algorithms that need huge quantities of data for the learning process. And scarce quality results in "garbage in, garbage out" scenario. Data may be “the oil for AI”, but its quality determines its accuracy.
Usually, data in organizations are spread out in multiple data silos of different legacy systems belonging to different departments with different visions about data. For obtaining quality data inside organizations, it may be necessary to form a cross-business taskforce to integrate the several existing datasets, solve inconsistencies and assess its accuracy.
All this data has to be properly integrated, cleaned and transformed, processes usually called data wrangling which constitutes one of the phases with more friction Typically data scientists spend around 80% of their time on this part of the process.
For all the reasons pointed out in this step, ensure that all data sources necessary for the use case are available and keep in mind that you may have to form a taskforce to integrate your data.
5. Set Up a Pilot Project
We at EAI know that Rome was not built in a day and for that reason, we believe that it is better to start small and accomplish a first pilot project. This pilot project should be well focused on straightforward goals, start only when all conditions are fulfilled and it is usually done with a time range of 3-6 months.
This principle of starting simple is transversal to the E.AI process culture and we believe most of the successful AI projects. The incremental nature of the AI processes allows us to prove value, get feedback, re-assess and develop accordingly. This approach can be materialized, for example, with a deployed model trained with only a fraction of the data and using less complex algorithms. Leaving the development of the final full-fledged model for the next iterations which will be done over an already functioning pipeline.
6. Build with Agility
Building an AI system is a combination of a technological and research project. To be successful it is necessary to iterate fast and stay focus on the use cases that deliver value to the business. To achieve the best performance of the AI systems algorithms should be trained iteratively with large volumes of data. For that to happen, the organization should implement agility in all AI development steps, moving from rigid and risk-averse to agile, experimental, and adaptable, embracing the test-and-learn mentality that’s critical for creating a solution in a short range of time. This new mentality provides also the flexibility to address the inevitable future challenges.
7. Give Power to Workers, Not Concerns
The insights and automation provided by AI should be presented to employees as a way to augment their capacity to deal with problems and accomplish tasks. For that to be a reality, companies must be transparent to workers on how AI can help them in a workflow, avoiding the concerns that it might eliminate the worker’s role in the company.
The preparation of the employees with the needed skills to use and take advantage of the change that AI is bringing is essential. Otherwise, the AI implementation risks being ineffective. This should be done in due time, starting small and scaling up to allow learn lessons, build trust and confidence.
Implementing AI in a business requires a strong commitment to innovation, and since it is an emergent field and only now brought to business on a large scale, many challenges are posed on the organization side. AI represents the next wave of digital disruption, but only with a strong strategic focus and openness to new ways to approach the business can this journey produce value from AI, and enjoy a critical advantage over organizations that don’t employ AI.
A great part of the challenges posed in these steps are addressed consistently on the Data Strategy Document, a document that defines and materialize the strategic approach for a data-driven business. The AI implementation benefits largely by the fundamental work made in this document. Our strategic AI implementation is always based on this document.