Episode I: AI Service Unit — building the baseline for scaling AI
Deploying AI at scale can boost sales, transform operations, engage customers, and generate new insights. Too often, companies have done their AI experiments, built multiple minimum viable products (MVPs) and prototypes, but merely any one of them could be adopted by the business.
One insurance firm made a big investment in building an AI based app for next best action recommendations for its sales representatives, aiming at increasing sales. Yet months after the successful piloting, not even half of its salesforce have installed the app.
One of our recent Capgemini researches shows that only one in three companies implementing AI is doing it at scale — meaning beyond small pilots and test projects, and adopting AI applications across business units, functions or geographies. To ensure AI’s full potential could be leveraged, a clear governance framework such as an AI Service Unit is essential.
Our analysis also shows a central governing body for AI implementation drives greater benefits (see figure below). But only about 37% of organizations implementing AI have a dedicated team that decides which AI initiatives will be implemented. Building up an AI Service Unit provides an environment that allows quick upscaling of promising use cases to close the gap between small-scale AI prototypes and a scaled AI in operative use.
Firstly, AI capabilities of an organization must be aligned coherently within three dimensions: Organization, Processes and Technology. A high maturity in one dimension cannot compensate for a low one in another dimension. At the next stage, the capabilities should be developed and embedded in the company’s individual value-drivers.
Organization
Implementing AI is a strategic decision. Defining a clear vision and mission for an AI service unit which aligns to strategy of a company is thus a key success factor. In many cases, the AI service unit is the competence center generating business value by creating and providing innovation, while also being a challenger for business.
If a healthcare company e.g. is going to introduce a new cyber-physical system application in the coming year, it should set-up or adjust the AI service unit mission accordingly.
Also, the service portfolio of the unit must be set corresponding to the vision and mission. Predictive modeling and visualization, data quality management, data integration and migration, agile project management etc. could all be part of the portfolio. Furthermore, it is important to define from the beginning necessary roles and responsibilities including detailed drafts of role profiles and make clear decision on which roles are mandatory and which ones can be sourced externally. Our experience shows if there’s one role that can do the most to start unlocking value, it is the business analyst, which bridges the gulf between business needs and AI solutions. The target operating model defines how the unit is embedded in the organization to interact with other units and what hierarchy it is subordinated to. Hybrid organizational models often prove to be successful in order to minimize silos and functional barriers.
Processes
A standardized AI process should embody activities as well as decision gates and involve business units to build up a use case management from use case definition by business units, granting data access and prototyping to data-driven decision-making. It is often necessary to backtrack to previous tasks and repeat certain actions in practice — exactly in this manner, the use case management ensures the integration of business users with agile iterations. Aiming at scaling AI, selecting the right use cases to scale is key. Our research shows that many organizations are jumping straight to some of the most challenging use cases, while only small minorities are focusing on low-hanging fruits in the sense of high benefit but low complexity. Neglecting those low-hanging AI initiatives would be a missed opportunity. Therefore, it is only essential to define a prioritization approach and abortion criteria of use cases at the beginning.
In many cases, the change-management challenges of incorporating AI into people and decision-making processes far outweigh technical AI implementation challenges. It is critical to enforce a step-by-step enablement that allows for constant re-skilling of the workforce. The shift of business focus caused by deploying AI will further require leaders to create a culture of continuous improvement and learning. This also includes the new way of working fueled by agile principles and co-creation with business users. The use case management reflects agile principles. Digital tools such as JIRA, Confluence etc. could be applied in order to accelerate this process. Co-creation of business users relies on ambassadors, who are located within each business unit and will lead the initialization of new use cases, as well as providing feedback in their evaluations.
That was the story at a large conglomerate. The AI service unit of the company implemented a handful of use cases without measuring their business values or feasibility during the implementation, which only led to large-scale waste. Later, the company introduced prioritization criteria of use cases as well as agile way of working, so that low profit use cases could be abandoned in time.
Technology
No AI can be built without tools. The AI service unit must be equipped with a technology stack with suitable tools. An agnostic tool selection process embodied with a proof of concept based on real data ensures suitable choices based on the service portfolio. The functionalities and values of tools can only be leveraged fully when the selection has been made coherently with business needs. Consequently, the selected tools can be integrated into an AI platform, which is a foundation that is engineered to implement AI initiatives in any computing environment. The AI platform supports every phase of the AI life cycle — from data sourcing, to discovery, to deployment. It reduces data preparation and cleansing effort, provides more trusting insights and thus leads to smarter and more confident decision-making.
For businesses that are consumer-facing and tend to see the highest adoption of new AI technologies, we have seen frequent assessments and constant improvements of their technology stack.
A full-fledged data security concept of an AI platform enhances privacy by design, which also builds compliance measures and AI ethics into the design, operation and management of the AI process. With the ever-increasing relevance of consumer trust, it is only essential to enforce privacy by design in order to be able to deploy AI at scale.
AI Capabilities of these three dimensions should be developed in alignment with each other. However, it is not always necessary that all three dimensions must achieve the highest maturity. One of our clients uses an automated API Yellow pages to complement their client information. That might not sound like rocket science, but it helps to derive viable customer clusters in order to conduct target marketing.
The AI service unit enables the company to create and deliver high potential AI use cases, as well as transfer them into real value.
The insurance company mentioned at the beginning of this article realized the failure of the next best action recommendation app was caused by low involvement of the first line salesforce in the run-up phase. Important features were overlooked in the AI model and thus it delivered insufficient results. The company setup a central AI service team, embodied necessary capabilities and introduced new ways of working. Sales numbers increased in the end, after its sales representatives started using the recommendations given by the algorithm.
Building up an AI service unit is an essential step in realizing a company’s AI strategy. The unit should be in continuous development while enforcing legal compliance as well as the company’s ethical AI code. Its organizational, procedural and technological capabilities will take AI practices from hype into reality and create a long-term, sustainable approach to generate genuine business value from data.
Sources: Capgemini Digital Transformation Institute, Turning AI into concrete value: the successful implementer’s toolkit, 2018. All figures mentioned in this article are from this study.
Stay tuned and follow this account for another article on this matter.
About the author:
Hongbin Xiang is management consultant with a special focus on data analytics in Capgemini Invent’s Frankfurt office.
Read also: #valuefromdata Episode II & #valuefromdata Episode III &
Our Success Story: Establishing a Data Analytics Competency Center for a German Federal Office
