TerminusDB Logo all white svg
Close this search box.
how to take on the big-tech monopoly of ai

How do companies take on the big-tech monopoly of AI?

ChatGPT has reached the masses and everyone is talking about it. Eyeroll (yes we see the irony about opening this blog talking about it). Even so, ChatGPT, as we know it today, will not be around forever. The initial test period will fade, and as Microsoft’s investment in the technology increases ($10 billion already!) the helpful AI assistance will most likely cost content marketers a small fee to use.

The fee is reasonable considering the huge infrastructure costs to support ChatGPT and the years of research and development that have gone into it. These AI chat tools are extremely useful for individual content creators, a boost for productivity, an angle to write, and hours saved. However, what about from a company-wide perspective, how do companies leverage artificial intelligence across their internal and external communications?

It’s not just OpenAI’s ChatGPT that’s taking AI into the mainstream, Google has released Bard too, or LaMDA (language model for dialogue applications). There was even a frenzy around this last year after an engineer claimed it was sentient. These huge tech giants have a big advantage over regular businesses. The key to this success is the vast amount of data they have collected and curated from the web. AI needs data and lots of it.

A regular company has a great deal of data. This data is often stored in applications and databases in different locations. Accessing and using transactional and operational data is a well-documented problem for large organizations. However, there is another source of data that often gets overlooked and is just as important – content. PDFs, websites, spreadsheets, and applications all surface content to relay messages. This content has an impact on sales, customer service, and countless other quantifiable metrics. Yet this content remains siloed and machines are unable to understand or process it.

For example, in a large food manufacturing business, product management produces a PDF product specification for every product. Product specifications contain everything from ingredients, allergens, cooking instructions, and packaging information. It is useful to wholesalers, distributors, marketing, sales, customer services, and customers eating the products. However, the information can not be reused programmatically. Marketing needs to copy and paste information for the website, salespeople email the PDF to customers, and these tasks are duplicated several times over during a product’s lifecycle. Not efficient.

Now, if the company did things differently and product management added product specifications into a back-office application, this information would be available for other teams to use. It would also help with other activities too. For example, if the marketing department wants to run a campaign targeting gluten-free products, they would be able to search the product specifications to list all relevant products.

But what does any of this have to do with AI? It’s the start you see. We bring together all of the relevant information in an organization and build a picture with data – which includes your content. Say we have machine-readable web pages and offline documents, combined with customer CRM data, sales data, Google Analytics, and operational costs, this picture becomes so much richer. It becomes a valuable dataset that can train AI and machine learning to serve customers and internal processes to greater effect.

How do you connect such disparate and diverse datasets together? With a knowledge graph. What is a knowledge graph? Essentially it’s a type of database, a graph database and unlike columns and rows in a traditional SQL database, it stores things as triples. A triple is two things (nodes) and the relationship between them (edge). It’s how we humans think about the world. John (a thing) is friends with (a relationship) Jill (a thing). With this sort of database structure, we can model just about any complex environment. For example, a webpage (a thing) promotes (a relationship) a product (a thing). Things can also have many relationships, the example webpage receives 10,000 visitors per month and generates 1,000 sales, and 60 percent of those sales come from people aged 26-35 with children under five.

Now imagine that you can build this knowledge graph to encompass the cost of manufacturing a product, the time it takes to make a sale, supply chain constraints, the complete demographics of customers, and countless other metrics that can provide valuable insights. The picture becomes vivid and the information is machine-readable and can be used to train AI. The potential is massive, dynamic websites serving exactly what a customer needs, recommendations based on many factors, preemptive emails to stop customers shopping elsewhere, onboarding documentation that delivers the right information at the right time, and back-office applications providing salespeople with the tools to close deals.

Good communication is about relevance. Good content that is useful to the recipient. In the noise of the digital age, ensuring the right message is delivered to the right person at the right time is difficult. AI can help remove some of these challenges with automation at a granular level. However, this comes with a huge caveat, without adequate training data, AI will provide no value.


Latest Stories