About the Author: Aris Valtazanos is Head of Data & Analytics at Oakley Capital, a leading mid-market pan-European private equity investor. Aris holds a Ph.D. in Robotics and a B.Sc. (Hons) in Computer Science and Artificial Intelligence, both from the University of Edinburgh.
Everyone is talking about how to leverage Artificial Intelligence to develop new data products or pivot their entire business model. When done correctly, AI has the potential to give your business an edge over rivals.
There are typically three ways AI can achieve this: by helping build products that directly generate new revenue; by augmenting an existing product or enabling you to improve customer experience and retention; or by improving your business’ internal productivity and efficiency. Throughout this article I’ll share some insights from recent projects across Oakley’s portfolio of companies.
Building New Products
Let’s dive straight in with the first of these: building new products. Start by thinking about the advantages you have over your competitors, and how you can leverage them. Data is typically one of the strongest and most defensible moats. Any business that has differentiated proprietary data is in a very strong position to create new products and services.
For example, Oakley Capital portfolio company vLex has one of the largest and richest legal datasets in the world - a library with more than one billion legal documents from over 100 countries. This has been the foundation of their new "Vincent AI" legal assistant, which allows lawyers to ask complex legal questions in natural language and receive a well constructed answer complete with sources. The combination of this unique dataset and their domain and technical expertise places vLex in a very strong position to benefit from emerging AI technologies.
One of the techniques vLex has used to build its AI product is called RAG or Retrieval Augmented Generation, which combines a generative model with information retrieval. This method ensures that the generated output is coherent and anchored in accurate and trusted sources. This is an ideal approach for legal research but also other sectors where databases or document collections exist – and may often remain underutilised because there’s just too much information and/or it’s too difficult to access it. RAG can make it a lot easier for employees (and customers) to access specific information and data efficiently, while helping provide differentiated outputs compared to a standard ChatGPT model.
Just like vLex did, you should identify your true differentiation and ‘right to win’ before developing a new product. What doesn’t work so well is the alternative route that some companies have tried, which is to develop thin layers around existing and commonly available tools, such as ChatGPT. Many startups are taking this approach, building new products that are essentially wrappers around ChatGPT – or adding incremental features to it. This is something that others can easily replicate and provides limited additional value compared to what's already available. It also leaves these startups vulnerable to changes in the underlying tool that they cannot control. For instance, ChatGPT initially didn’t support querying PDF documents, and a few solutions (AskYourPDF, ChatPDF, LightPDF, and more) had emerged to plug the gap in this space. In late 2023, ChatGPT introduced PDF chat functionality, which brought the added value of a lot of these solutions into question. So when thinking about new AI products or companies, it’s always important to ensure they are future-proof.
New AI-driven data products may not always generate new revenue directly, but can still serve as a great way to attract or retain customers. IU Group, another Oakley portfolio company, is Germany’s largest and fastest growing university with more than 140,000 students. IU recently launched an AI-powered study buddy named Syntea, offering students personalised teaching and feedback. Students who were previously too embarrassed to raise their hand and ask questions in crowded lecture halls can now put questions to Syntea any time of day and night, while the AI tool monitors their progress and can personalise learning plans to meet their needs. This has led to a 27% reduction in required time to complete a course/degree.
Productivity Hacks
Even when not used to build or augment customer-facing products & services, AI can still provide a competitive advantage by improving the productivity and efficiency of internal teams. Staying with IU Group, it has harnessed productivity hacks across the business by harnessing ChatGPT. It launched an open call across the business asking teams to name their most time-consuming tasks and offering GPT workshops to teach teams how to prompt. Twenty-five teams (around half) have followed through and on average report 10-20% efficiencies achieved. One example is the production of personalised landing pages for new courses. Previously, when responding to particular Google searches for courses (for example, a BA in Food Science) multiple teams would take several weeks to produce and host a relevant webpage to capture that traffic. Using GPT has reduced this to hours, with a knock-on, positive impact on student sign-ups.
Code Writing
This is another area that is also ripe for disruption by AI. Alerce, a Spanish transport and logistics software business and another Oakley portfolio company, is exploring the use of AI assistants to improve the productivity of its software development teams.
While being a largely creative activity, software development also involves more mundane tasks such as writing tests to ensure the code performs as expected, or documentation to explain what each piece of the code does. In a similar way to how they are beginning to master human language, generative AI models can be trained to learn the structure of programming languages and code, and used to automate some of these more cumbersome tasks. They can also be taught to rewrite entire pieces of code to make them more computationally efficient, or even generate new snippets of code from human prompts (“write me code to do X”). Early results from Alerce suggest a potential saving of 10-15 hours per developer per month across these tasks, which can free up creative time for other efforts. The company is also exploring other emerging applications of AI models to programming; for instance, using human prompts to query structured databases (for example,“Tell me how many products did we sell last April?”). In addition to making developers more efficient, these applications can also lower barriers to entry and allow more users within an organisation to interact with data.
So what will make or break your AI project? Here are the key things to consider:
Factor in cost of development and maintenance: many AI solutions (especially Generative AI ones) require considerable resources to build, run and maintain – such as data storage or cloud compute. While there is a downward trend in the industry around some of the costs associated with these resources, a lot of them will still be sizeable, and required on an ongoing basis to ensure a solution runs smoothly. Depending on how a solution has been developed (bought off the shelf or built in-house), there may also be associated licensing and support costs, or investment in human experts (data scientists, software engineers etc.) to build and maintain these solutions. It is important to take all of these factors into consideration and weigh them against the financial benefits the solution is expected to bring.
Secure executive buy-in for your AI project: most AI solutions form part of a wider business process, so ensuring there is strong senior support is critical for both their successful development and adoption. For vLex, the leadership's vision for transforming legal research through AI was not just about adopting new technology, but reimagining the future of the legal industry itself. By championing a vision that leveraged AI to enhance legal research, the leadership not only set the strategic direction for the company but also galvanized the entire organization to embark on this transformative journey with enthusiasm and confidence. In summary, leadership buy-in is not just about approving projects or budgets; it's about fostering an environment where innovation thrives, risks are embraced as part of learning, and the transformative potential of AI is fully leveraged to redefine industries.
Avoid vendor lock-in and remain as technologically agnostic as possible: particularly in the fast-moving field of Generative AI, the landscape of companies and products is constantly changing. There is constant competition between LLMs such as OpenAPI, GPT, Mistral and others and this means performance is increasing all the time and the cost is reducing. So, whether buying off the shelf or building in-house, it is essential to have flexibility to change these tools as needed, and constantly monitor the market to see how alternatives are performing. This also ensures you have a fallback when one of them goes wrong, which is not uncommon with LLMs (think of the OpenAI board crisis in November 2023, which threatened to tear the entire GenerativeAI industry apart!).
Beware of reputational and legal risks: while providing great automation and efficiency, there is also a risk that comes with the outputs of AI solutions – including hallucinations (e.g. ChatGPT and similar tools) which are incorrect or misleading results caused by wrong assumptions used by the model or insufficient training data, as well as biased or unfair decisions. You may have heard the case about two US lawyers fined for unintentionally submitting fake citations generated by ChatGPT. You need to be aware of the needs and the severity of potential issues in each industry, and ensure that outputs are auditable and explainable. In this context, testing a solution is just as important as developing it; always try different ways to “break” it (and ensure it doesn’t!) before releasing it to customers or other end-users. Auditing the outputs of a model can also provide assurance that there isn’t any bias for or against specific cohorts, or indeed any falsified information. In this domain, there are many “explainable AI” tools that help interpret the outputs of a model, by highlighting which parts of the data typically drive these outputs, and using this to determine if there is any underlying bias that should be mitigated.
Beware of the implications of shifting your market position: often, developing these solutions means shifting the company from a "data provider" to a "product provider". While this opens up new business opportunities, it may also lead to competition with existing customers who are also active in the product space - so understanding the trade-offs is key.
From teaching your team to prompt, to buying off the shelf solutions or building your own product, AI has the potential to transform your business. Act now before your competitors do, and chose the right partner to help you navigate this fast-moving, fast-changing technology.
Have a question you'd like one of our experts to answer? Please contact us at editorial@boardwave.org
Comentários