Data questions answered: Which business problems can effectively be solved by AI, ML, or BI?">
Why data matters when it comes to business problems
A big challenge in data today, and one I’m a big proponent of getting out in the open, is that data needs to build better links to business teams. It is true that data teams have significant demands on their time, modernising infrastructure, building capabilities, and so on. But ultimately, it boils down to one key objective: how to deliver more business value.
So, before we discuss whether AI, ML, or BI is the most appropriate solution, let’s start with a more fundamental question: How can organisations align their data efforts with meaningful business outcomes?
8 ways to help understand how to use data to solve business problems
1. Get to know your business stakeholders
One of the most important things a data team can do is build strong, trusted relationships with its business stakeholders. They also need to ensure that they’re brought into conversations at the very start so that they can best advise on how to deliver against business goals.
I have seen many data teams whose primary focus is on firefighting ad-hoc queries from business stakeholders and lifting existing, largely Excel-driven reports into BI dashboards.
A big challenge with data is that it feels familiar “enough” for most stakeholders to understand how reporting and some basic data analysis can support their strategic goals. On the other hand, it feels completely alien for its more advanced capabilities, such as ML and AI.
This degree of half-familiarity is one of the major challenges for data ecosystems, as stakeholders may have preconceived ideas of what they want. They simply need an implementation team to execute on them, rather than creating room for a data team to bring their ideas and potential solutions to the table. That’s why data teams must embed themselves across the business and understand the big strategic initiatives at the earliest outset. Doing this will enable them to identify problems where more advanced techniques could deliver significant value.
2. Follow the money and spot where data can affect it
It may sound obvious, but focus on the P&L. Where are the major costs? Where are revenues generated? These areas should be the focus of data-driven improvement. For example, I’ve had several projects in my career where, by predicting demand on staff (e.g. call centre lines, demand on caseworking staff), we’ve saved significant money and improved customer experience. This has been as a result of reducing the number of unanswered calls and the knock-on cost of onboarding expensive permanent employees.
3. Look for the Excel
If a manual process exists-often underpinned by Excel or ad hoc scripts-it’s a sign that the business already sees predictive value. These predictions present a perfect opportunity for both modernisation of the prediction itself, e.g. improving accuracy to deliver more value, and updating the business processes that rely on the predictions to make them faster and more cost effective, etc.
For example, if rebuilding an algorithm out of Excel and VBA into Python can cut execution time from 2 days to 1 minute, a world of near real-time opportunities becomes possible. Opportunities that can increase business agility and responsiveness to changing marketing conditions.
4. Seek out operational pain points
Speak with individual contributors-they often have the clearest view of repetitive, manual, or error-prone processes. Their input can reveal areas where predictive or automation solutions would have high ROI and help to drive consistent quality.
Consider third-party data sources, such as understanding complaints (e.g., to an ombudsman) or customer feedback where they have felt undercatered for. These can give good insight into business failures, where potentially moving to better predictive analytics would assist in preventing recurrences.
5. Ask the right questions
Statements like “I wish we could predict…,” “summarise…,” or “prevent…” often signal problems suited to data or AI solutions. If all else fails, ask: “What headaches would you remove if you had a magic wand?”
6. Lead with relevant examples
A common early step in my discovery workshops is demonstrating how AI or ML has driven results in similar use cases. For example, walking through how a retail product recommendation algorithm may also identify clients with similar profiles to make more personalised investment recommendations in asset and wealth management.
7. Consider unstructured data opportunities
Many valuable business processes rely on documents, forms, conversations, or images -ideal candidates for modern AI approaches like GenAI. Automating decisions based on these sources can generate significant savings and improvements.
In a prior project, the client used a two-part questionnaire to assess applicants. They built their own (Excel) based prediction engine, running over the structured part of the questionnaire to offer recommendations. We managed to build a tool that sat alongside their own, which used the unstructured data captured for the first time. This helped them automate the processing of applicants and reduce their mishandling rate, saving millions of pounds in fines and fees.
8. Identify and mitigate risks
ML and AI are also powerful tools in risk mitigation and management, as evidenced above. Understanding parts of the business that are subject to significant risk - whether risk of fraud, regulatory risk, etc - is an important part of identifying use cases. For example, in banking or wealth and asset management, there are significant obligations around anti-money laundering and Know Your Customer (KYC). This is where ML and AI offer the ability to increase scale, consistency, and automatic detection of potential breaches, in order to prevent them, rather than just reporting after the fact.
How to understand where to use AI, ML, or BI in your business
Clarify the use case
If the goal is to monitor KPIs, BI is likely sufficient. If the aim is to anticipate events or automate decisions, ML or AI may be more appropriate. Understand the optimal outcome that maximises business value, and use the right solution to deliver that outcome.
Distinguish actionable insight vs. automated insight
Ask: Is this insight actionable, or should it be automated?
It’s not unusual for business executives to be drowning in a sea of data reports organically built over the years to solve specific needs. The chances are that many are unlikely to be delivering any value. It is critical to focus on the data that aligns with the key metrics a stakeholder needs to be informed about their decisions. I’ve seen clients who have thousands of analytics and reports running every week, with the vast majority of those never being used or opened.
Focus on iteration
If it’s worth predicting, it’s worth predicting well. Predictive models aren’t “set and forget.” A 1-2% improvement in forecast accuracy can yield significant gains. Treat models as products to be evolved. At the very least, models will decay, as will data, and they will require continuous training and deployment to maintain performance. However, in addition to maintaining performance, we should always aim to incorporate additional data sets and more advanced techniques to improve predictive power.
Critical lessons to consider whether your business is using AI, BI, or ML
No matter the technology, evaluating whether a business problem is suitable for AI, ML, or BI comes down to these three steps:
1. P&L impact
Can we clearly explain how this solution will improve profitability—by increasing revenue, reducing cost, or mitigating risk? If you can’t explain how a project drives business impact (not necessarily the specific impact, but at least the mechanism of the impact), it likely needs further refinement before you invest.
2. Data availability and data foundations
Do we have the right data captured, structured, and accessible in the appropriate platform? Can we deliver the insight or prediction into the operational system? Do we have the governance framework in place allowing us to use this data?
This is especially critical in industries dealing with highly sensitive or special category data, such as trade unions, financial services, healthcare, etc. It’s not uncommon for many legacy, disparate systems to lead to data silos that need to be migrated, and for governance and frameworks to be updated. Both need to be done in order to fully exploit the data’s potential.
3. Technical feasibility
This is where a data team can make an impact by using their experience to navigate what are pipedreams (e.g. "I'd love to use ML/AI to predict the lottery numbers") versus what are plausible problems to tackle and solve. Doing this will reduce the risk of focusing on projects that might not succeed or deliver useful results for the business.
Taking the steps covered here will help data teams and business leaders co-create the right problems to solve, balancing ambition with practicality, and prioritise projects with the highest potential for return on investment.
Is your data ready for AI and ML?
Is your data quality, infrastructure and governance ready to support the outcomes your business needs? Do you have the foundational readiness for AI and ML? Find out with our Data Foundations Readiness.
Follow me on LinkedIn for more insights.