Defining a Successful AI Strategy for 2018: Key Thoughts from a Data Scientist

As the hype around AI continues, building and executing on an AI strategy that supports market competitiveness will be top of mind for executives. The AI pilots are complete, yet executives are still grappling with what AI means for their organisations.

As the use cases develop and capabilities emerge, businesses will look to defining an AI strategy for the enterprise to maximise the benefit and impact. Core to this strategy will be the understanding of how data is accessed and integrated, as well as the plan for talent and skills development, infrastructure evolution, auditability requirements and governance requirements.

These strategies will ensure organisations are building the capabilities needed to succeed with AI in the long term and transform operational business models.

Deploying Competitive Algorithms at Scale

Simply put, automation is required to harness the multitude of algorithms coming into play, enabled through accessibility of data, technology, tools and frameworks. With the focus on AI, organisations have access to a whole new dimension of analytics, providing algorithms that can operate on complex data types to create a plethora of insight.

As companies work to keep up in a marketplace of change, there will be continuous AI innovation as analytic teams develop new models using a myriad of languages, libraries and frameworks. The potential of these algorithmic solutions will be huge, however, businesses will need a fast path to production allowing automated deployment, monitoring and management.

In this AnalyticOps setup, the concept of deploying a model will become more fluid. As models are innovated at rapid speed, this will necessitate the ability to launch multiple algorithms, potentially using different frameworks and languages, in parallel, at scale. By having numerous versions and evolutions of a model in production, organisations will perform champion-challenger approaches to allow algorithms to compete for first place.

Self-Adapting AI to Mass Scale Real-Time Decisioning

With an increased focus on personalisation and effective customer journey interactions, there are potentially tens to hundreds of decisions to be executed per customer, requiring real-time capability. These decisions need to be driven through data that is fresh and up-to-date.

In order to execute on these decisions, a new paradigm to performing and acting upon analytics will emerge. No longer will models be left to execute on their own accord for weeks and months, inevitably becoming stale and leading to sub-optimal results.

Instead, automated processes will continuously feed machine learning and deep learning algorithms with fresh data to allow models to self-adapt. The humans in the loop, who would traditionally analyse algorithmic outputs and make execution decisions, will be replaced by machine processes that can automate, real-time decisions that take up-to-date insight to action at speed.

Governance of Automated Decisioning

Automated decisioning, involving analytics, will mature and increase to mandate higher levels of governance. As organisations implement an increasingly larger number of self-adapting, AI algorithms into production, they will look to ways and means to manage the algorithms now making instrumental business decisions.

Whether it is an Algorithm Office, a Chief Analytics Officer or an AI Competency Centre – companies will seek the right structures and oversight to combat algorithm bias and ensure sound proof decisions. Furthermore, as the field matures, regulation will catch-up to mandate not just data protection, but also responsible algorithmic insight execution.

To protect against regulation violations, and also meet consumer acceptability of data use, organisations will look to create frameworks and guidelines for the use of insight. For example, marketing will look for enterprise wide frameworks to govern customer communications that consider factors such as channel, message, timing, environment, competitiveness etc.

Self Service to Enable AI Commoditisation

To date we have primarily seen code driven AI, requiring specialist skills, but this is changing as more AI applications make the techniques accessible across the enterprise. As businesses move to adopt AI in more aspects of decision making, business users will demand self service capability to understand and interact with AI models.

Furthermore, as AI automates certain jobs across the enterprise, the human element of creativity and craft will be ever more important. The business domain knowledge to interpret and guide AI efforts will require more transparency and understanding of AI to the wider organisation outside of the data science elites.

Embracing the Cloud to Take AI to New Heights

Using the cloud allows organisations to tap into the new tools and frameworks supporting AI endeavours. We have seen the pendulum swing to various extremes as organisations battle to understand which programming language, software or platform, open source and proprietary, will emerge as the front runner to become the go-to choice. However, the reality of a balanced ecosystem is emerging.

The most successful organisations will be those that embrace the flexibility that comes with a mix of open source or commercial software, and those that recognise ecosystem architectures that allow for tool and technology choices to be adaptable and swapped with minimal impact: it’s these organisations that will find themselves able to solve a broader range of business problems using the latest AI techniques.

The cloud will be an integral part of this ecosystem, providing organisations with ultimate flexibility to scale and access tools on demand, in extension to on premise capabilities.

Developing Skills Fit for the Enterprise

The hype over data scientists will be replaced with the realism that taking ideas through to production requires a hybrid team of specialists. Companies will look to re-assess their internal capabilities, including the structure of their data and analytic teams. Structures will evolve to bring analysts and data experts closer to business SME enabling more effective execution of business relevant use cases.

Also, as businesses look to build a pipeline from innovation to production, they will seek to invest in skill sets that go beyond data science to data and software engineering, DevOps and architecture. It is only through hybrid teams that organisations can operationalise analytic endeavors to realise the true value of AI. This will require a mix of hiring and upskilling through training and mentorship.

Ultimately, the future of AI adoption means that organisations to have a strategic plan that focuses on a variety of elements that enables AI execution to create business value. Organisations need to move beyond capabilities for AI pilots to the right environment, skills and infrastructure that can take AI to productionalisation.

AnalyticOps frameworks, will enable enterprises to deploy, monitor and manage AI applications that self adapt and commoditize insights across the enterprise. As businesses move to rely on algorithms to make efficient and effective business decisions, organisations will need the agility to have models deployed in competition, with results feeding in real-time to business action.