Identifying and Prioritising AI Projects
A crucial component to your data strategy is undoubtedly knowing when AI is suited to your project and which use cases to prioritise. With reports that so many data science projects are never making it into production (a whopping 87% according to a VentureBeat study), knowing when and where to apply AI in the first place is key to getting on the path to impact and business value from your AI.
In 2021 together with Talent Insights, Data Futurology launched the first edition of our Data Leaders Who’s Who articles, where we asked leaders to share their insights on some of the frequently asked questions we receive around strategy and leadership. One of the first questions we put to them was around project prioritisation. Below we share some of their key tips.
Marek Rucinski, Deputy Commissioner Smarter Data, Australian Taxation Office (ATO)
AI presents huge opportunities for organisations to tailor customer experience and present more convenient offerings to your customer base. How do you determine which projects to use AI for and prioritise your projects?
We consider the use of supervised and unsupervised AI techniques to derive actionable insights for businesses where the solution to the problem requires predictive insights, consumption of large and multiple data sets, or data that is semi-structured or unstructured. We apply our data ethics principles to guide us on how to apply these techniques.
AI is not always the most suitable answer to a data and analytics project. Basic slice and dice analysis and visual reporting of high-quality data can also deliver significant value. Visual data representations can be a simple, but powerful tool to create new insights and consume complex data patterns. Whether you use AI or not, you can realise additional value by integrating the insights or data into human and/or robotic workflows. But human oversight will always be needed!
We use several criteria to prioritise our data and analytics projects, including:
The extent to which it is required to deliver legislative changes and government programs
Alignment to the ATO’s strategic objectives
Our overall data and analytics capability to deliver it.
Where the project is not related to legislative changes or government programs, we assess the value potential, strategic fit and feasibility of execution.
Felipe Flores, Head of Data and Technology, Honeysuckle Health
How do you determine which projects to use AI for and how do you go about prioritising your projects?
AI can be used for operational decisions and for strategic decisions. They both need to be handled differently. Operational decisions happen thousands, if not hundreds of thousands of times per day. Hence there is a lot of work that can be done at the MLOps and ML engineering area that can help us improve the scale at which these services are delivered.
On the other side, there are relatively few strategic decisions. If you think about an acquisition, for example, it is an important strategic decision for an organisation, but it would happen a couple of times a year at most. But there's a lot of analysis and insights that needs to go into large strategic decisions and AI can be used for insights to assist there.
In terms of which projects to use AI for – I’m usually thinking about operational. We need well-prepared, high-quality data presented to the model creation part which will create better models over time. With strategic AI it is much slower, in small groups, using AI/ ML to create insights and consult or advise on decisions. There are four components that help with project prioritisation:
Feasibility – can it be done with the data, technology, people and resources we have now?
Desirability – does the customer want this? Can we quantify that market desire and how important that feature is for the market?
Viability – is there a financial return on this? Is there a path to profitability?
Organisational readiness – is the organisation ready to bring the product to life for the customer or ready to consume that product? This readiness encompasses two sides – technical readiness and the people side – are people aware that this is a problem or that customers want this, and do we have the momentum to bring this product to life? Timing here is critical.
For more insights from data leaders and to read the full interviews with Australia Post, L’Oreal, ATO and Standard Chartered Bank, you can access the entire archive of interviews in our MemberSpace.
To have your voice in the community and share your data stories, either in our live event, podcast, articles or blog - get in touch to learn how you can contribute: info@datafuturology.com
OpsWorld: Deploying data and ML products
24-25 October 2023
DataOps is evolving! AnalyticOps, MLOps, AIOps, and DataGovOps are enhancing workflows with reliability, reusability, and reproducibility. From value-add to necessity, we’ll reveal how your organisation will benefit from automation, scalability, and consistency of delivery at its core.