Most people still think Large Language Models (LLM) like ChatGPT are best used for creative writing. While these applications can be helpful, they often yield mediocre results.

What if I told you that AI's real magic lies in its ability to perform menial, data-intensive tasks faster, cheaper, and more accurately than humans? Let me take you on a journey through a recent client experience that showcases AI's power in solving real, messy business problems.

The Challenge: Taming the Data Beast in Travel

A travel sector client approached me with a treasure trove of customer data from a questionnaire. They wanted to analyze responses to client questionnaires to determine travel destination trends, but they faced a massive, chaotic wall of messy text that resisted all attempts at analysis.

There were a variety of issues:

1. Volume: Tens of thousands of free-text responses. Imagine trying to manually extract country names from prose text responses a thousand times over.

2. Variety of Spellings: Take "Cambodia" for example. We found it spelled 15 different ways! Here's a taste:

• Kambodscha
• Cambodia
• Kambodia
• Cambodgia
• Kambotscha
• Kambodschah
• Kamboscha
• Kambodjia
• Kambodgia
• Cambodja
• Kombodscha
• Kamodscha
• Cambogia
• Cambocha
• Camodia

Now multiply this chaos across every country mentioned.

1. Outdated or Inaccurate Names: “Burma” instead of “Myanmar,” “Zanzibar” instead of “Tanzania.”

2. Irrelevant Information: Responses like "I don't know yet" or "First we go to Germany, then we are thinking of flying directly to LA".

3. Multiple Countries in One Response: "Thailand Malaysia Laos Vietnam Phillipinen Indonesien".

4. Abbreviations and Language Variations: "USA" vs. “United States of America,” “Zanzibar” vs. “Sansibar.”

Can you feel the data analyst's headache forming? Unfortunately, this is the reality of raw, human-generated data – messy, inconsistent, and resistant to traditional analysis methods.

Traditional Solutions

Before we dive into our AI-powered solution, let's consider the typical approaches many businesses might take:

1. Manual Processing: You could have an employee painstakingly standardize each entry. However, that would be slow, expensive, and prone to human error.

2. Outsourcing to Freelancers: Freelancer Platforms like Fiverr or Upwork have many freelancers who technically could do the project, but you'd still need to control their quality to ensure consistency.

LLMs to The Rescue: Teaching Machines to Clean Our Data Mess

Now, let's explore how we tackled this challenge using AI, providing a framework you can adapt to your own data processing needs.

1. Data Preparation: Respect Privacy, Reduce Risk Before proceeding, we scrubbed the data of personally identifiable information (PII). This isn’t just about complying with data protection laws—it's about being a responsible steward of clients’ information.

2. Initial Data Assessment: Showing AI the Ropes We selected about 10 of the most complex examples from our data set. Think of this as giving our AI a crash course in the messiest parts of our data problem. We're not just throwing data at the machine – we're curating a learning experience.

3. Prompt Engineering: AI Teaching AI In an innovative twist, we used one AI model (Claude) to generate instructions for another (ChatGPT). The result was a robust set of guidelines tailored to an LLM’s expectations.

4. Quality Check: Trust, but Verify We tested our AI solution on a sample set, meticulously checking its work. This isn't about blindly trusting the machine but refining and perfecting its approach. We're teaching the AI to learn from its mistakes, just like we would train a human team.

5. Error Handling: Embracing Uncertainty We defined clear protocols for how the AI should handle uncertainties or errors. Should it make a best guess? Flag for human review? This step balances efficiency with accuracy, ensuring our AI knows when to ask for help.

6. Scaling the Solution: From Sample to Tsunami After training and testing our AI, we created a script to process the entire dataset in batches using the OpenAI API.

The Business Impact: From Data Chaos to Strategic Clarity

By leveraging AI for this data processing task, we achieved results that would make any business leader sit up and take notice:

• Uniform, cleaned data ready for analysis: Like turning a jumbled jigsaw puzzle into a clear, coherent picture.
• A total processing time of just a few hours: Tasks that would take weeks were completed in less time than a long lunch meeting. A really long one, to be fair, but the point stands.

Key Takeaways for Forward-Thinking Business Leaders

1. Look beyond the AI hype: AI's true power often lies in automating tedious tasks, not just in creative applications. It's not about replacing human creativity but augmenting human capabilities in data-intensive tasks.

2. Identify your data bottlenecks: Look for areas in your business where messy or inconsistent data hinders decision-making. These are prime candidates for AI-powered solutions.

3. Start small, then scale: Begin with a sample set to refine your AI approach before processing larger datasets. It's about building confidence and competence before tackling your biggest data challenges.

4. Combine AI models for superior results: Different AI tools can work together to create more robust solutions. Don't be afraid to mix and match AI capabilities to solve complex problems.

5. Prioritize data privacy and security: Always consider data protection regulations when working with AI and customer data. Responsible AI use builds trust with your customers and protects your business.

6. Automate for the future: Once set up, AI-powered data processing can continue to deliver value as you collect new data. It's an investment in ongoing efficiency and insights.

7. Embrace the learning curve: Understanding and implementing AI solutions may seem daunting, but the potential rewards in efficiency and insight are immense. It's about evolving your business for the AI age.

Focusing on these practical applications of AI can unlock significant efficiencies in your business operations. The key is to identify the right problems—those repetitive, data-heavy tasks that are bogging down your team—and apply AI strategically to solve them.

It's a running joke at our house that Siri is about as smart as our hamster. With Siri’s newfound understanding of human language, interactions are set to become more intelligible. But while these improvements are nice, the big innovation at Apple’s WWDC24 lies beneath the surface – AI Agents and automation.

Introducing App Intents: A New Paradigm

The true game-changer introduced at WWDC24 is the concept of App Intents. At first glance, they might seem like just another incremental update, but they represent a significant shift in how we interact with our devices. App Intents are essentially standardized interfaces that allow apps to communicate more effectively and for users to automate actions across multiple applications seamlessly.

Imagine a world where your apps not only understand your commands but can also work together to accomplish tasks without constant manual input. This is where App Intents come into play. Siri, who you can also write with, serves merely as the human interface for this change, enabling it to act on our behalf more intelligently.

What Can Siri Do Now?

With these new capabilities, Siri can pull lunch suggestions from messages, suggest the right driving time to your destination, make reservations, edit photos, and more. These actions demonstrate the immediate potential of App Intents, but they are just the beginning. As developers start to explore and implement these standardized APIs, the next weeks and months will reveal even more possibilities, or reveal how much was marketing and how much is real.

The Future of Commerce?

To me, this hints at the future of commerce, for both B2C and B2B. We’re heading towards a paradigm where AI agents and apps largely interact on our behalf through APIs, creating a new kind of open web. This web might not be for human-to-human interaction but rather for AI-to-AI collaboration.

The implications are profound. I'm imagining automated systems handling everything from customer service interactions to complex supply chain logistics, all without or with very little human intervention. 

Experimenting with Today’s Tools

If we squint our eyes, we can already experiment with this future using today’s tools. Many services have standardized APIs, but we’re often not using them to their full potential. Automation tools like Power Automate, Zapier, or Make can help bridge this gap, allowing systems to talk to each other and perform tasks that would typically require manual intervention.

By experimenting with these tools now, businesses can gain a head start in understanding how to leverage automation and prepare for the more advanced capabilities that are just around the corner.

Embracing the Future of Automation

So, if you want to get ready for this brave new world (dystopian pun intended), now is the time to experiment with automation tools. By embracing these technologies, you can unlock new levels of efficiency and innovation, positioning your business to thrive in an increasingly automated world.

AI-to-AI interactions are bound to happen. Those who experiment with automation systems already at our disposal might have a leg up working with these systems once they're ready.

A recent survey by Deloitte revealed that 54% of Swiss business leaders believe their tech infrastructure is AI-ready. This level of confidence stands in stark contrast to their counterparts in Germany (37%), the UK (25%), and the global average (40%).

Having engaged in numerous discussions with business owners and leaders in Switzerland, it's clear that many are still navigating the best ways to integrate AI into their operations. Given the rapid advancements in AI over the past 18 months, this uncertainty is understandable.

The Reality of AI Readiness

The high confidence among Swiss leaders raises an important question: What does it truly mean to be AI-ready? Many businesses are leveraging out-of-the-box SaaS tools that incorporate AI features, often through large language models (LLMs). While these tools provide a level of AI integration, genuine readiness involves a deeper, more strategic approach.

For businesses embarking on significant transformations, AI integration requires building robust, tailored infrastructure. It's not merely about using AI-powered tools but embedding AI into the very fabric of the company's processes and strategies.

The Complexity of AI Integration

The confidence reflected in the survey might be misleading. AI integration is complex and multifaceted, often requiring significant investment in infrastructure, talent, and ongoing maintenance. Simply using AI tools like ChatGPT does not equate to comprehensive AI preparedness.

Conclusion

Swiss business leaders exhibit remarkable confidence in their AI readiness, but true preparedness goes beyond surface-level integration. It involves a strategic, well-thought-out approach to embed AI into core business functions. As the landscape of AI continues to evolve, businesses must critically assess and adapt their strategies to stay ahead.

For a deeper dive into the survey findings and practical steps to ensure your business is genuinely AI-ready, check out the full report from Deloitte here.