Using AI to unlock big business potential: A five step guide

AI and data can be powerful tools for organisations to unlock operational efficiencies, inform new product and service creation, and improve the overall customer experience. Yet many organisations struggle to capitalise on this potential, often because goals are unclear, milestones and budgets are unrealistic, and challenges are underestimated. In this blog, we’ll look at five key elements you’ll need to tick off in order to build an effective data analytics function.

1.     Use cases

2.     Team

3.     Data platform

4.     Data assets

5.     Users

Step 1: Getting started – finding the low hanging fruit and securing buy-in with clear and well-supported use cases

I always recommend starting with one small, well-defined project, with clear (and inarguable) business value. Setting out to solve one incremental problem will help you show value and secure buy-in from the top.

Any organisational change requires careful planning, and setting up a data science team is no exception. It is important to emphasise that AI and data products will enhance the quality of your employees’ work, and that automation will allow them to focus on the elements of their work where human skills are most needed.

The best way to demonstrate value is to find a business problem where you can build a firm case for improvement, identifying measurable and sustained value with high confidence. Often, a good place to start with is with your customer data, typically a large and under-used dataset. This data could allow you to identify use cases such as classifying high value customers for marketing purposes. Other frequently revealing datasets are those related to operations, staff, supply chain, and premises and equipment or facilities.

Then, create your first prototype to test the pipeline, develop your team’s skills and generate buy-in from stakeholders. This self-contained product will incorporate the major elements of your project, and provide a platform to develop the professional and technical capability within the team.

Step 2: Hire the right people to drive success – building a data science team

An effective data science team will be made up of people with diverse but overlapping skillsets. On the technical side, this includes data engineers, data visualisation developers, DevOps engineers, software engineers, data architects, analysts, and data scientists. You will also probably want some product managers and business analysts to handle the non-technical aspects of the product and understand user needs.

Using Agile methods can improve your team’s ways of working and increase throughput. It is important to ensure that your team fits your organisation and balances the requisite skills and experience. It is also important that your team maintains close alignment with your users, as this allows them to respond quickly to the users’ changing priorities. Agile teams that work in this way can rapidly create proofs of concept and show results – which offers the ability to fail fast and pivot to a more productive line of work if necessary.

Step 3: Laying the foundations of your data-driven future – building a data platform

The infrastructure is a critical success factor in building your data and AI capability, and allows you to make it easier to explore data and build and deploy data products. There are a number of areas your infrastructure should support:

• Data processing

• ETL pipelines

• Test and deployment

• Data exploration and research

• Software development

• Data visualisation

• Enforcement of prescribed compliance, audit, and governance standards

The foundation of your data platform will be a data lake, which is similar to a data warehouse, but which accepts many diverse data types and file formats. It is important that metadata are generated on ingest to a data lake, to ensure usability.

Considering whether to use the cloud or on-premise systems for your data platform is an important consideration. I generally recommend the cloud, especially for organisations with less data maturity, especially considering the benefit of managed services that cloud providers offer to support rapid set-up, configurability and scalability.

From the outset, you should also consider the deployment and maintenance of your data products. Focusing on this at the outset, rather than at the final stages of development, ensures you can seamlessly integrate and deploy your product. Your choice of infrastructure can play a role here. A major advantage of cloud deployment is that it if often much easier to deploy attractive and user-friendly dashboards and visualisations to help demonstrate the value of your first project.

Step 4: Creating data assets - your data engineers are the key

The data asset is a central idea in effective use of data and AI. Data assets are stored in a database or other robust format, and must meet a number of qualifying criteria:

1. They meet a clear business need, and are often used for business-critical decisions

2. They act as a single source of truth – i.e. each data element must be generated in a uniform, standardised way, agreed upon across the firm.

3. They are usually a composite, drawn from at least two or more source systems

4. They are typically subject to a further level of processing or aggregation

5. They are subject to a data quality and validation process to ensure integrity

6. They are live data, updated on a frequent, often daily or intraday, basis

These data assets can then be combined, where necessary, to answer higher-level business questions. This results in layers of assets that are each in a more highly aggregated and processed form. An example is a customer master data asset, which can be used to answer almost any question asked about a customer, from marketing techniques that are most likely to be successful, to predicting how to mitigate the risk of complaint. It is important to be able to update assets as and when new requirements emerge, and just establishing a robust update process can deliver substantial value to your team by avoiding the need for data scientists to manipulate old and messy data when developing each new product.

Step 5: To deliver value, focus on creating exceptional user experience by seeking feedback (and make the UX feel crisp and simple)

This step is (kind of) a cheat. It’s not really a step, more of an overarching point to remember when creating data products – always remember the user.

If you are doing a piece of analysis, its easy to get distracted by the modelling and forget to think about the people on the receiving end. Returning the results of your analysis to your users in a convenient and intuitive form is critical factor that often distinguishes AI and data science from analysis or statistics. So it is important to spend time ensuring that your data products are clear, especially in cases where the dataset is large and complex.

Make sure you assess things like how often your users need a dataset updated. If intraday data is critical, but your product only pulls data overnight, that will prevent users from getting value from the product you have laboured over. An agile delivery method can help you here – deliver iteratively, while staying close to customers.

Your data science team should have the skills to provide visualisations such as graphs, maps and charts, as well as text, to clearly demonstrate the business value of the product. Tableau or Power BI are your best friend here. If you need to build an application that isn’t too complex (such as an analytics app for senior management, or an HR and learning portal for staff), ‘no-code’ solutions such as Bubble.io are great. For anything more than this, some software engineering talent, at least with JavaScript or a frontend framework, can be a big help.

Conclusion

While these are high-level waypoints, and this is far from everything you’ll need (we haven’t touched on data governance, for example), if you can tick off these five elements, you’ll be well on your way to delivering major impact using data and AI.