Real-World AI Use Cases and What Lies Ahead

The pace of AI over the past few years has been nothing short of staggering. Every notch up in quality and speed unlocks new use techniques and use cases. In this blog I'm going to look at the use cases of today, tomorrow and in the (not too distant) future.

‍

Use Cases Today

Summarisation

Long, boring documents can be summarised into key points. The best models to use here are GPT-4o and Claude 3.5 Sonnet. They allow you to grasp something quickly. Lower-level models are good for a one-line summary.

Categorisation

Depending on the use-case, you may not need an LLM for this, but you can often get pretty good results with Open Source models, or cheaper commercial models like GPT 3.5 and Amazon Titan Text. 

Data Extraction

You have a bunch of messy, unstructured data. This could be an email, a scan of a form, or a form completed as a PDF. You need to turn that into structured data to feed it into a system. No more data entry!

Data Cleansing

If you have a big spreadsheet with postcodes, phone numbers and addresses in a big jumble of formats, an LLM can understand context for the whole entry and make sensible corrections and tidy ups.

Customer Service Agents

If users have the same questions again and again, the simplest implementation just answers those. More advanced implementations can actually perform actions on behalf of the customer, for example changing a booking, or processing a refund.

Thinking Partner

One of my favourite use cases is to use LLMs as a creative thinking partner.

Simply asking ChatGPT to "ask me 10 questions, one after another, wait for me to answer before asking the next question", then explaining what your desired outcome is - is really powerful. You can often discard the output, as what's really important is the thinking you've done along the way.

A great example of this is our Ticket Helper which helps us think through user stories when writing up Jira tickets.

‍

‍

Use Cases Soon

With specialised knowledge, some of these near-future use cases can start to be unlocked now.

Automating Undifferentiated Knowledge Work

Any workflows which involve receiving an email and performing a basic action or data entry task from the back of it are ripe for automation. In very large enterprises this grunt work that helps glue processes together can be a surprisingly large amount of effort. Freeing employees from this will allow them to think about the edge cases or spend more time talking to people.

Personal Assistants that Work

These assistants will be able to order things for you online while operating within budget. Understanding your preferences for delivery and which days you’re likely to be home. They’ll be able to fight parking tickets (we’re already seeing some of this in the US), claim refunds, and cancel subscriptions. There are some hurdles around terms of service and SMS authentication, but technologies like passkeys may be able to help here.

‍

In the future

John McCarthy once said; “As soon as it works, no one calls it AI anymore.” You can see this with number plate recognition. Initially an amazing breakthrough in AI, it is now just a bog-standard part of traffic and car park management.

Many of the use cases that seem amazing to us today will just be part of everyday software - as 'boring' and ordinary as using a magical piece of glass to communicate with people on the other side of the world.

‍