The data science, machine learning and artificial intelligence (AI) market keep on growing – no change in that. Companies are investing more and more in these capabilities to create value, either through growing existing or new businesses, increasing resource efficiency, or growing customer value.
Simultaneously, these turbulent times set an increasing demand for resilience also in data science and AI solutions. As there are major shifts taking place in certain industries, the importance of fast innovation and exploration – understanding causality instead of mere correlation and phenomena instead of mere behaviour – is highlighted. But how to do this and how to ensure that investments remain value-creating and benefits scale-up also in the future?
Always focus on solving problems worth solving
Besides being technically and methodically competent, data science and AI teams should have a strong focus on understanding the customer and business needs they are trying to solve. The solution’s desirability for the customers and viability for the business must be analysed. The same goes for the technical- and data-related feasibility of the solution. However, in my opinion, feasibility is too often focused on more than desirability and viability. Customer centricity and the ability to understand the business needs are required throughout the solution lifecycle as needs evolve and the world changes. Therefore, to be adaptable and resilient, data science and AI teams need to be cross-functional and they need to include enough business and service design competencies. This ensures that the largest efforts are put into those problems that are worth solving and needs that need fulfilling.
When, for instance, developing Machine Learning based product recommendation systems for a web store, it is quite easy to identify a large pool of potential product recommendation functionalities which could be built. However, when enough time is put to understand both the customer needs and business realities, it is often quite clear that specific functionalities both add the biggest customer value and sales uplift.
Ensuring continuous customer centricity and business understanding helps data science and AI teams usually also to formulate smaller and simpler tasks to their development backlog. The potentially big leap to introduce a new AI model to tackle a new use case might change to a few smaller steps where the use case is first solved with a quite simple solution creating immediate value.
Understand the business value you create and measure it rigorously
When developing a data science and AI solution, or any solution for that matter, it is fundamental to have ways to analyse and measure the actual tangible value the solution creates. It is also almost as important to monitor how that value changes when time passes or when modifications are done.
Sometimes measuring value is hard, but it should be done even through indirect indicators. You should always aim to embed the data science and AI solutions or their outputs concretely into actual business processes and specific tasks in them. If a tight integration is not in place, there is a real danger that the solutions are underused, create less value than they could, and that the created value is hard to measure.
When, for example, developing data science tools for creating data-driven and purchase potential based target groups, there is a big difference whether the tool is integrated into the digital marketing team’s daily work or not. If the tool is specified and trained to be part of the digital marketing process and if the target groups created with the tool flow automatically to marketing automation systems, you are going to create far greater value than without these.
Measuring value gives justification for the investments done and supports potential further development by giving a benchmark against which to improve. Furthermore, measuring value and A/B testing capabilities enable a constant growth hacking type of mentality which in my opinion is needed when developing data science and AI solutions – continuously striving for improved results is a common denominator in top-performing data science and AI teams.
In addition, data science and AI solutions are tricky in a way that without intervention the value they create tends to decay as a function of time. Properly measuring business value also allows organisations to notice these changes and act accordingly. Focus on this enables resilience in data science and AI solutions.
Invest in automation, development practices, and scalability to keep the focus on innovation and new business value
Along the “AI journey” every company is in danger to end up in a maintenance trap. This trap is a situation, where all the time the data scientists go to maintaining already existing data science and AI solutions. This is a result of inefficient development and deployment practices, unclear ways of working, and a lack of automated workflows and monitoring. If not dealt with, this insufficient focus on creating production-ready solutions will result in declining management support as the promised benefits are not achieved.
The realisation of business value around the data science and AI capabilities is strongly dependent on the ability to free up time by automating a lot of the maintenance and development work in an efficient and scalable way. Why is this important when building resilience? It is relevant because the freed-up time can be used to keep focus in building additional value through innovation and development, and in those larger changes in the world and behaviours, which require manual fine-tuning in the existing solutions.
When, say, developing an AI-based smart labelling and semi-automated case handling system for a public organisation’s officers, it is likely feasible to create many dozens of different AI models for different topic domains the cases – for example, applications – relate to. While doing this you want to be in a position where a lot of the maintenance and development related activities are automated, or you will end up doing nothing but maintenance before you have even created all the necessary models.
There are also large differences in how organisations can create additional value out of new data science and AI applications. These differences partly originate from different approaches in overall data science and AI platform architecture design. A strong focus on fast business benefits is mandatory, but at the same time, sufficient time and vision need to be put into creating scalable and resilient solutions architecture. Based on my experience, you can take steps in the right direction by focusing on modularity and microservices. When done well, this supports scalability by enabling far greater reuse of components between different use cases and resilience by making it easy to change or modify obsolete components without the need to make large changes to the overall solution.
Minimise person dependencies
One additional component in being resilient is related to the competencies and skills an organisation has accumulated in its data science & AI teams. There often is a lot of tacit knowledge regarding existing AI solutions and, unfortunately, this knowledge might even be person dependent – deep understanding regarding the solution’s core embedded in an individual professional. At the same time, the job market is not exactly stable and turnover does exist. Thus, to ensure continuity and velocity, organisations with data science and AI investments should focus on minimising personal dependencies. This is especially important as – rightly – data science and AI solutions are increasingly more business-critical.
When, for instance, maintaining a large and complex machine learning-based predictive maintenance system which is also the basis for a large part of your company’s workforce optimisation, you must prevent situations where the departure of a single key professional could compromise the whole in-production system.
Foster common development practices, solution pipeline and project templates, as well as ways of working.
How to increase business value and enhance change resilience?
To sum it up, here are my key take-aways on ensuring higher business value out of data science and AI investments and on building resilience into them:
- Always focus on solving problems worth solving. Take time to understand the customer needs and the phenomena behind the behaviour.
- Understand the business value you create and measure it rigorously. Try to embed the data science and AI solutions in business processes and the actual tasks in them to ensure business value.
- Invest in automation, efficient development practices, and scalability to keep the focus on innovation and the creation of new business value. Free up time through modular architecture and by automating repeated steps in data science and AI solution development, deployment, and operations.
- Minimise person dependencies. Enable continuity and velocity in development through common development approaches and through working in teams.
This blog series focuses on key areas for building resilience. Read the other posts:
