A recent meeting with AntWorks’ marketing team reminded me that the various parts of technology companies communicate in different ways.
At one end of the scale, the R&D team is often so detail-focused and wary of over-promising that I end up with pages of nuance rather than the eye-catching numbers that help me start conversations with would-be customers.
At the other, marketing types like to keep messages simple to catch the attention of busy people like you. But, as head of sales for the UK and our global insurance business, if I let marketing get over excited they’ll say something that causes me problems down the line when I’m having conversations with potential customers.
Focusing on the practical
I prefer to focus on what we can do for people; practical solutions to their problems.
Few things hurt companies as much as IT implementations gone wrong. So smart customers stay wary. The sales process rests on building and maintaining trust; explaining what we can deliver and then delivering it. Allowing a bunch of promises to get made that I and the project teams can’t deliver on kills that trust. That can’t happen.
So, when the marketing guys wanted to put out a line, from a video by my colleague Ben, that “AntWorks can deliver extremely high levels of accuracy with small sample sets, “
I look at that and I start to worry. It’s not that it’s not true. It is true. But I need to manage expectations.
I’ll talk to people who may not know how many documents you typically need to train the A.I. for Intelligent Document Processing.
Most systems need thousands, even millions of documents to train them. It takes months. They require supervision. Every time they meet a new document layout someone has to hold the system’s little hand.
AntWorks’ CMR+ typically needs only a couple of hundred. It can be as few as 50 or 60, it could be, in some situations, say, 500. It’s a step change. (There’s also a critical advantage for any business that operates in a stringent regulatory environment; this low training load radically reduced the hardware required for training so that it’s easily affordable to do on premises.)
But what if someone has in their mind that it should only need 20 or 10 or five documents to train their AI? I’m going to have someone saying ‘well, 200 seems like a lot to me,’ so I’m a bit cautious.
And what do people mean by high levels of accuracy? Some systems claim 95% plus on the documents they can read, but only with the documents they can read – and they can’t actually read 80% of them because they’re non-standard or contain signatures, handwriting, odd fonts, images, you name it.
So, what do we mean by ‘accuracy’?
There’s out-of-the-box accuracy and then there’s accuracy after the customer has continued to train CMR+ for a few weeks and shown it a few typical forms.
If you do that, put a human in the loop so exceptions get reviewed, before you know it you’re getting closer to 100% accuracy across all your documents.
I’d rather tell people we offer ‘high levels of accuracy’ or that we are in the business of minimising human intervention because I would much rather under promise and over deliver than the other way around. That builds trust.
But what I’d really like to say is this: we get you to a point where you see real business benefits fast. Not years, but weeks.
So, we have the solution that can deliver the highest levels of accuracy, can deal with real-world data, understand a number of languages, process high volumes, business benefits can be felt in just weeks and all for a sum that doesn’t cause your calculator to go up in flames. So, in summary, a truly global enterprise solution that can address the needs of our customers.
That’s what I want to say, but putting it all into six words is the job of the marketing bunnies.
