AI: 101 (part 2) – Making a business case for AI
A quick recap: In our previous blog, we discussed the challenges and complexities behind the rush to embrace AI. We talked about what you needed to run AI (lots and lots of good data, and specialised hardware and software), the GenAI hype cycle, and the potential for use cases to simply bomb without delivering value.
Lastly, we finished with some somewhat worrying stats that questioned the readiness of Australian organisations to harness AI, let alone understand what they’re going to do with it.
So, now you’re up to date, let’s move on to use cases, those that are well-defined and understood, and why others are just pie in the sky.
AI legal eagles
Earlier this year, Melbourne law firm Lander & Rogers set up their own AI Lab within the practice. They are currently working up “three or four” prototypes a day (mainly using Microsoft Copilot) to test how they can leverage AI to interact with various data types. Some of the most valuable use cases uncovered (from a flood of ideas generated internally) are those that save lawyers time in finding and surfacing the information they need.
Amongst Lander & Rogers’ winning use case is engaging AI to build a chronology of events in legal cases.
What’s not on their agenda, though, is using AI to rewrite a lawyer’s work. Their Head of AI Engineering, Jared Woodruff, says, “That’s not what AI is meant to do. The AI is there to give them [lawyers] all the information that they need to execute that decision and execute it with precision, saving them time.”
Lander & Roger have taken a strategic approach to their areas of focus and have defined where AI can be used to deliver definable business value in the legal profession.
More great legal tech
Working with AI service provider Automatise, Ethan, an Australian-owned technology service provider, has invested heavily in building Cicero, a pioneering AI tool specifically designed for the Australian legal sector.
Ethan says that Cicero has already been adopted by several mid-tier and enterprise law firms in Australia and is transforming workflows and enhancing productivity. To quote: “As these firms integrate Cicero into their operations, they experience firsthand the benefits of high quality, coherent summaries and analyses of legal documents, a feat made possible by the fine tuning of LLMs for local use cases.”
Again, this is another great, well-thought-out use case that meets specific industry needs. If all goes to plan, it will transform the Australian legal industry and deliver an impressive ROI.
AI pie in the sky
There are numerous high-profile examples of poor AI use cases. Some are simply ill-conceived, ethically irresponsible, dangerous, or just plain thoughtless. Others have used insufficient or inadequate training data, which produced skewed and reprehensible outcomes.
This hasn’t daunted the would-be AI adopters, though.
McKinsey’s 2024 global survey on AI reports that 65% of respondents said they regularly use generative AI for at least one business function. However, only 10% of those organisations had implemented gen AI – at scale – for any use case.
Further to this, a senior partner at McKinsey, when speaking at the MIT Sloan CIO Symposium, said that while there are many organisational initiatives, “a lot of the efforts are scattershot and don’t contribute to the bottom line.” McKinsey’s survey confirms this, saying that only 15% of the responding companies realised an improvement in earnings for those AI initiatives.
AI isn’t cheap or easy
Why is the failure rate (or inability to generate an ROI or measurable business value) so high? This is where we dig out the old axiom: ‘fail to plan, plan to fail.’
Like any technology project, there needs to be rigour around the ‘what, why, and how’ of the business case. Major considerations include:
Worryingly, ADAPT’s CIO Edge Survey from February 2024 says that 66% of Australian CFOs say their organisations are unprepared to harness AI. 25% are non-committal, and only 9% say they’re AI-ready. AI-ready or not, 48% of the CIOs surveyed say they haven’t even defined any clear use cases for AI.
- Setting out your commercial objectives – in other words, defining the problems you’d like to solve within your business, as well as the desired outcomes. Then, deciding if AI is, in fact, the right solution.
- Ensuring your data is up to scratch – remembering that AI is data, your data needs to be up to date, accurate, relevant, ample – and used appropriately. All of this requires preparing and adhering to a sound data governance strategy.
- Realistic expectations – yes, AI can be wonderful, but it’s not a magical cure-all. It’s critical not to overestimate the capabilities of AI, and it is essential to test and validate systems to ensure they meet the basic requirements of safety, compliance, accuracy, ethics, transparency, fairness, and security.
- Making sure you’re resourced up – adopting AI comes at a cost. Just as you wouldn’t let a newly qualified driver lose in your brand-new Tesla, you wouldn’t (or shouldn’t) place your trust in anyone who doesn’t understand the legal, ethical, and data considerations mentioned earlier. A successful AI project also requires an investment in technology, data and infrastructure. Poor infrastructure can result in performance issues and failure to support the implementation of advanced AI models, compromising both their efficiency and reliability.
- Scalability – it’s also critical to test AI projects at scale. What works perfectly as a test project may disappoint in terms of efficiency and reliability when rolled out to the entire organisation.
Blue skies or uncertian horizons?
We know of many businesses that are keen to increase their compute power so they can train their own AI. And we’re supportive of that; we love to see organisations innovate.
But what concerns us is that few know what they actually want to train AI to do.
Without clarity of purpose, a strong business case, and a structured, disciplined approach, AI has the potential to become an expensive toy rather than a transformative technology that contributes to the bottom line.
