Any series of tasks that you can clearly define has always been easy for automation. If every Friday morning you need to click something, download a file, put it into an email, address it to the same person, hit send, then it can pretty easily automated. (although this is still typically not taken advantage of, and is still manual).
But what if we don’t always get the file the same way? What if it needs to be sent to someone else instead, and what if it could arrive at any time and in any channel? The only certainty in the approach is we figure out what to do by opening it and reading it.
This is something that automation has always really struggled with.
So what we do instead is we do it ourselves manually or get low-cost or low-capability labour to do it, because for this all we need is someone to open the file, read the contents, and figure out what to do.
Now that computers can read well, and not just check for correct sentence structure and spelling, but kind of ‘understand’ what the text is saying, the capabilities of AI are bleeding into the world of work that is hard to code into a workflow.
The example of vague work like reading the contents of a document or email and then figuring out what the next steps are, and then executing those steps are now right very possible.
Let’s take an example, a customer complaint, actually the oldest known customer complaint, the tablet of Ea-nāṣi. The "Complaint Tablet to Ea-nāṣir" is a remarkable artifact that takes us back to ancient Mesopotamia, around 1750 BCE. A clay tablet discovered in the ruins of the city-state of Ur.
Nanni's message to Ea-nasir
“You promised to give Gimil-Sin top-quality copper ingots when he arrived, but you didn't keep your word. Instead, you offered my messenger, Sit-Sin, inferior ingots and told him to take them or leave them.
I'm insulted by your treatment. I've sent respectable men to collect my money, which I left with you, but you've repeatedly disrespected me by sending them back without it, even risking their journey through dangerous lands. Am I the only one among Telmun's traders you dare to treat this way? You're the only one who's shown such disrespect to my messengers. You think you can act like this just because I owe you a small amount of silver? Remember, I've already sent a huge amount of copper to the palace on your account, as has Umi-abum. And it's all recorded on a sealed tablet in the temple of Samas.
For all that copper, how do you repay me? You've kept my money bag in a place full of enemies. You need to return everything you owe me.
Be aware, from now on, I will only accept the finest copper from you. I will personally select each ingot in my yard and have the right to reject any that don't meet my standards. This is because of the contempt you've shown me."
For Nanni’s message, there are 3 key pieces of information we need to access to take action - the complaint (Nanni’s letter), everything we know about Nanni, and how we typically solve these types of issues.
Today this information would be stored in a Customer Relationship Management (CRM) system and knowledge articles (possibly in another system, and SharePoint, and Email...).
The CRM would hold Nanni’s address, lifetime customer value, recent transactions, their open and closed complaints, open opportunities, customer account team and others. The knowledge articles would hold all the company processes for working with customers.
It’s a weird problem in that the systems are built to help you, but they’re often only valuable to you once you’re so good at your job that you don’t need them. In the meantime they are confusing, duplicative and cumbersome.
Any relationship manager, sales rep, call centre agent or customer service specialist will know that stitching the data across these systems into a solution is a horrible process. The CRM data is in multiple tables and systems, and there could be thousands of knowledge articles. It takes months (years) of understanding customer problems, muscle memory and repetition to become capable.
But what if you had a grad who did it all for you? They’d go through all the information in the CRM, go through every knowledge article, find every relevant little piece, and then compile it and produce a briefing for you. You still get to decide what the action is, but all the prep work has been done by someone else. You focus on what you’re good at - fixing problems and not searching through systems.
A large language model combined with some smart can do this for you.
In the section below I’ll demonstrate an agent briefing based on Nanni’s complaint.
It was created using a fictional Salesforce record and knowledge articles as source data. It was written entirely by an LLM.
For any further clarification, please contact your team lead.
For this scenario, the agent still has the ultimate power to decide what and how to do it, but they’ve been able to get all the information they need, kind of like if they had a grad or an assistant to do it for them.
The AI assistant can not only do this with text, but with data too. If you wanted a report of all of Nanni’s transactions, and you know that they are spread across three tables and that one table has problematic data so double-check that there aren’t duplicates - you can instruct the AI assistant to do exactly that.
It’s reading it, combining it, picking up things you’ve told it to look out for, and then creating an output. It can even tell you the things it wasn’t sure about that you might want to take another look at.
Finally, this is a bit of a weird one, in all my work with AI I’ve discovered you learn their ‘style’.
You realise different levels of instruction result in different types of results. You realise that sometimes you need to give it very specific direction, but other things they’re great at being left alone to figure it out themselves…
Sounds familiar doesn’t it?
To get started, watch my webinar (no signup required) on how to use ChatGPT in Sales Masterclass.
Have a great week.
Dawid
.jpeg)
