Extending Product Roadmaps with Generative AI: Acrolinx CEO Volker Smid (Part 4)

Sramana Mitra: The objective is learning. What your technology is doing is, it’s streamlining that learning.

Volker Smid: By helping the author to align the content with the objective. The objective is learning. You need to use technical, simplified English. The content needs to be measured against clarity. Sometimes it’s as easy as saying you can’t use long sentences. The assistance of the authoring environment is one aspect.

The other aspect is that, in the backend, we provide all the insights and scoring of the content. Which content is below the threshold? Where do you have content that has a high degree of engagement?

Sramana Mitra: Is learning the primary objective around which enterprises hire you?

Volker Smid: No, there are very different use cases. I’ll take a completely different use case in the finance sector. Imagine you run a bank. Part of your business is mortgage. You have 2,000 underwriters and you have content for these underwriters.

Let’s assume that the content is confusing for the underwriters. It creates the risk of litigation. In this case, the objective is to get the confusing part out, eliminate certain pieces, and make sure that your content is aligned with the need. In this case, it’s risk avoidance.

The third use case is around inclusive language. You run a website with a million pages. The million pages have been created over 15 years. Today, you have the ultimate need that the language you use is free from discriminatory aspects. You need to have a platform to audit your content to identify pieces of content that you should change.

This whole process is important. In an open-source environment, content that we check can even be source code. In source code, there are language constructs that people have used in the past that you cannot use in today’s world. A typical construct is master-slave which cannot be used in today’s world. You have to go back and identify areas that you need to change.

The optimization of the user interface is another one. In general, our ideal customer is a Fortune 2000 company. It must be a company with a significant amount of content that needs to be holistically managed and governed.

Sramana Mitra: How many customers do you have?

Volker Smid: About 200.

Sramana Mitra: I’m going to ask you about what’s happening in the current landscape. Generative AI has entered the conversation. What impact does generative AI have on your product, roadmap, and the prospect of being disrupted?

Volker Smid: As we discussed in the very beginning, I’ve seen many technology changes throughout my career. Some of them were significant and some weren’t. When I saw Chat GPT in November the first time, it reminded me of the introduction of the PCs into the world of mainframes. It was more intuition than anything else. I had a feeling that this is something that we need to take a very close look at because it will be very disruptive. We started to look into this.

After a long assessment, my current hypothesis in the context of content supply chain is the following: Generative AI can help create an enormous efficiency in the content supply chain and the content assembly line. The creation of structure and the creation of initial content have become a lot easier. The average cost of the creation of an HTML page was between $800 and $1,200. With the help of generative AI, this can come down to $150 to $200 per page.

If you think of the dimension of seven million pages that I mentioned, there’s an enormous impact on the efficiency to create pages and to maintain pages. You need to have guardrails in place because these large language models can actually create risk for an enterprise. Our current hypothesis is, it is here. Enterprises will use it because of the efficiency gains, but they have a lot of concerns about the pattern of usage, copyright infringement, the uncontrolled publishing of a complete generated piece of content.

This is where we believe that the concept of content governance has gone to a very different dimension. Today, we need content governance to be able to engage with generated AI. There’re two aspects of generated AI. One is helping an author to become more productive and efficient. The other aspect is if you take existing content pools and you tune an existing model, the tuning or the transformation creates a lot more opportunities than just using an existing model.

The tuning aspect to create your own language model has additional value. The thing is, if you have not established a sense of quality of content, it is very difficult to identify what content pool you want to transform and tune. If you take low-quality content and tune a model, what you get back is low quality. If you take high-quality and brand-aligned content and tune a model, what you get back is a very high quality transformed model and efficiency for the authoring environment. Then combine that with the ability to take existing high-quality content and tune a model. Put these two pieces together.

This segment is part 4 in the series : Extending Product Roadmaps with Generative AI: Acrolinx CEO Volker Smid
1 2 3 4 5 6