Enterprise AI — the integration of AI into various aspects of enterprise technologies — is growing and expected to transform how companies function, influencing every part of a business from process automation to marketing and risk management.
Sifted and Sapphire Ventures’ latest report, which recognises the 100 most promising privately held software companies in Europe and Israel, featured 22 AI startups for enterprise use.
How is enterprise AI changing the way companies work — and what are the opportunities and challenges?
Transforming industries
Bob van Luijt, cofounder of AI database platform Weaviate, sixth on Sifted’s Rising 100 list, says that what differentiates enterprise AI is that “you need to adhere to the wishes of the enterprise”.
He says that ever since GenAI came to the fore, C-suite executives have started to take interest in developing their own companies’ strategies for GenAI — and this is the gap that enterprise AI fills. “It's easy to read your own data; it's secure to write your own data. We [Weaviate] come to you so people can run it inside their own private cloud,” van Luijt says.
You can create content in any language, on message and on brand
Last month, Sapphire Ventures announced its plans to invest over $1bn in AI enterprise startups. One imminent area of potential in enterprise AI is document automation, says Andreas Weiskam, partner and head of European operations at Sapphire Ventures.
“This is true for all kinds of forms processing that you can see, be it invoices in large organisations that get thousands of invoices that come in, or bills of lading where not everything is in a structured format, but maybe there are stamps on there,” he says. “These things that needed people to process it, now can be completely digitised and automated by AI.”
Weiskam adds that content creation for advertising and marketing is another area that will be transformed by enterprise AI: “You can create content in any language, on message and on brand. You can create more and more personalised content in real time, based on information that you can gather of people who visit your website and their assumed interests.”
For Alberto Rizzoli, cofounder of V7 Labs — a platform which turns images into training data for AI and ninth on the list — one of the biggest advantages of adding AI to B2B SaaS is that “you can add a copilot to almost any piece of software. Every piece of B2B SaaS requires a lot of work to set up and to learn — and the promise that AI holds is that it will make this process a lot easier”.
He added that an example of integrating AI with products and internal processes is one of their enterprise customers, a biotech company that makes microscopes, that can now make microscopes with AI. Another is a company that makes chainsaws which now understands what it's sawing so “it can work more independently and also tell you when you're doing something unsafe or when it's being worn out”.
Current and future trends
Van Luijt says that enterprise AI could take different forms depending on the needs of different businesses. Weaviate, for example, offers B2B SaaS as well as a “Bring Your Own Cloud” service where they operate the software inside the private clouds of the enterprise.
He adds that there’s already an inclination towards running enterprise AI software within private clouds. Weaviate’s open-source technology has 2.4m downloads at the time of writing.
Areas of companies that were previously mainly doing data science and analysis are now being rebranded as AI
Weiskam believes GenAI represents a platform shift and predicts that it will pervade into “all the different layers of B2B software, adding “it will even optimise the legacy software, updating solutions that have been around for a long time”.
Rizzoli says that AI favours incumbents by nature and predicts that the world’s most popular products will adopt AI
“Areas of companies that were previously mainly doing data science and analysis are now being rebranded as AI because AI tools are able to do a lot more,” he says.
Challenges to adoption
Despite the massive potential, there are some challenges to adopting enterprise AI.
“The rules that apply for any type of enterprise software, apply for AI as well. And so you have to think about data privacy, governance, those kinds of things, everything that comes with explainability of AI is very important for the enterprise,” says van Luijt.
The rules that apply for any type of enterprise software, apply for AI as well
Rizzoli says that to avoid risks such as data privacy, model errors and bias, companies should try to spend as much time on their data as possible.
“Training is something that happens in the background, so you can just let it run and then come back and check it,” he says. “It’s key to make sure that you focus on really high-quality training data on any kind of implementation.”
For Van Luijt, however, fine-tuning models to avoid errors and other risks is already an outdated method. Instead, he suggests Retrieval augmented generation (RAC), “where you basically intertwine the database with a model”. This means the data sits in the database that runs inside the virtual private cloud of the customer.
Weiskam says that the opaque nature of LLMs is also an issue when it comes to adoption as people don’t really understand what actually happens inside the models. Due to their non-deterministic nature and risks such as hallucinations or model errors, he adds that AI models “still require some human interaction and checks — so that is potentially one area that could hamper the growth in adoption”.
Rizzoli says that another key challenge is that many enterprises don't have as much in house talent in ML as AI companies do.
With the “paradigm shift” sparked by generative AI in enterprises, training is key for everyone to be able to use it well, as companies are still figuring out enterprise AI, says van Luijt. He suggests workshops, training courses, webinars and other educational initiatives.
“People know that they want to use it — much like a child who knows that they want to ride a bike and that they want a bike for their birthday, but they don't know how to ride a bike yet. If you just give them the bike, it’ll lead to accidents; you first need to teach them to ride the bike.”
Discover the ultimate B2B SaaS startup full list here and download the report here.