This is the age of AI. We have progressed through the stage of early adoption into an exciting phase of rapid evolution. Across all sectors, AI is making business more accurate, performative, customer-centric, contextual and secure.
In London’s iconic Gherkin building, Expleo held a fascinating panel discussion on integrating AI into enterprise organisations. Featuring use cases and valuable lessons learned, the event delved into the practical applications of AI, best practices, and its impact on people and culture.
Here are nine takeaways from the lively, interactive session with innovators and thought leaders.
1. What is AI, anyway?
The lazy definition of AI is: ‘getting computers to do things that humans can do’ – but that could also describe simple automation. Besides, this wrongly assumes that humans are the benchmark for intelligence. A more elegant definition is ‘goal-directed adaptive behaviour’. The key word here is adaptive, as the system can learn to make better decisions. In reality, very few AI systems meet this definition.
2. Six ways to alleviate friction
There are six applications that can be used on any friction across any supply chain in any industry to help navigate the complex world of AI safely.
- Task automation: simple algorithms can drive a massive amount of value without the need for shiny new technologies like Generative AI and machine learning.
- Content generation: Large Language Models (LLMs) give everybody the ability to create any generic content. (However, the true battleground is creating brand-specific, production-grade, differentiated content – and that’s very hard.)
- Human representation: this means replacing human beings with things that look and behave exactly like a human being. The next step is ‘audience brains’ that recreate how people perceive content, such as promotional material.
- Insight extraction: The power in machine learning and data science is explaining predictions. By extracting insights about the world, you can make even better decisions.
- Complex decision-making: Use a computer for decisions based on multiple factors (i.e. there are over a trillion ways to allocate 15 people to 15 jobs). Humans struggle to compute anything with more than seven variables.
- Human augmentation: This last one sounds creepy, although LLMs can create digital twins (known as digital assistants) that are trained on an individual’s personal and professional data.
3. Three golden questions to ensure ethical governance
There’s a huge amount of misunderstanding and misinformation associated with ethics and security. When deploying AI in production, ask yourself these three questions.
- Is my intent appropriate? The key difference between AIs and human beings is that human beings have intent.
- Are my algorithms explainable? The difference between software and AI is that AIs tend to be opaque in terms of how they make their decisions. If you make them explainable, then you make them more transparent, auditable and governable.
- What if my AI goes right? Imagine the unintended consequences of overachieving your goal. How might you cause harm elsewhere?
4. Put the necessary guardrails in place
5. Overcoming the highest hurdle
The toughest challenge for AI is moving from the exciting stage of testing, experimenting and creating cool proofs of concept into the reality check of production. The AI application loses its novelty factor when it needs to deliver on value targets. How do you re-architect your stack so that it can cope with increased demand? How do you make your data transformation layer fit for purpose? You will have made huge progress and be buzzing to kick on. However, when the tyres hit the tarmac, you may need a big rethink.
6. Applying the right tech to the right problem
Broadly, there are three areas of your organisation where AI can get involved:
- Core productivity tasks (e.g. writing emails faster)
- Supply chain differentiation (using AI to accelerate your transformation)
- Disrupting your own business
There’s a temptation to see emerging technologies as a panacea – and also a danger of misallocating the wrong tech to solve the wrong problems. Technology then gets the blame, when it’s really a case of driver error.
7. Installing a top-down/bottom-up process
We all get excited about emerging technologies. Seven years ago, it was machine learning. Today, it’s Generative AI. The hype means that board members exert pressure on the CXO to demonstrate rapid progress in AI. At the grassroots level, meanwhile, there are bright minds who are doing the work and know how best to automate tasks. They need encouragement to bubble up ideas. To move the needle, you need an AI process that unites the whole organisation around the goal of creating meaningful value.
8. Choose big, start small and move quickly
When implementing an AI deployment strategy, you need to identify the most compelling and valuable problems in your organisation. However, don’t try solving the problem at the first attempt. Find a target area to improve and move quickly to show results. That momentum will attract attention and take you forward to bigger projects and bigger budgets. Also, any pilot must tie back to the current strategy cycle, or it will quickly lose funding to a higher strategic priority.
9. Give yourself the best chance of success
If your AI team have never developed software in your company, then you could be asking too much of them. Far better to work with partners who will build your AI capability over time. Your people are invaluable for identifying frictions – but it could be that buying applications from an ERP or acquiring a value-driven start-up is a better way to differentiate your business. A prudent CIO might hire a chief AI officer to bring deep expertise and apply the right bets to solving the right problems.
Deep dive into the insights
Watch the panel discussion from Expleo’s ‘Integrating AI’ event to hear our speakers delve into the practical applications of AI, best practices, and the impact on people and culture.
