Artificial intelligence and data analytics are transforming every enterprise. Salesforce Ventures recently invited a group of corporate leaders and members of our Innovation Advisory Board (IAB) to meet with select portfolio companies—including Protect AI, Astronomer, Hugging Face, and Starburst Data—for a discussion about how enterprises can leverage new cutting-edge technologies to streamline operations and improve efficiency.
In attendance were technology leaders from major enterprises, including Comcast, Amazon, T-Mobile, KPMG, Coinbase, and more. The conversation featured many great insights for startup founders and enterprises alike. Here were a few of our top takeaways.
Note: Quotes from our speakers have been edited for clarity and concision.
Protect AI co-founder Daryan Dehghanpisheh on the typical AI tech stack for enterprises…
“We work a lot with a lot of highly regulated industries like healthcare, life sciences, finance, and defense. There are pretty consistent build patterns for generative AI, and at some point they nearly all rely on open source software (OSS) in some way. OSS can be found in models, data sets, ML Ops tooling, and more. In LLMs, we see a lot of customers fine-tuning OSS models for specific use cases, such as an AI-augmented scientist for molecular discovery in biotechnology processes. Because there are so many regulations about how models, data, and other information is used in healthcare and life sciences applications, these enterprises all need greater control and analysis on the model and the entire AI system.”
“To do this, they use Protect AI to assemble a model and ML bill of materials. With Protect AI, they can manage new vulnerabilities and technical risks, create finer grain access controls to the system, model, and data sets, and have automated documentation for critical regulations. They can also monitor how clinicians, data scientists, and physicians or researchers are using the model and data assets so they can’t, essentially, ‘peek over the wall.’ That could violate some regulatory elements. Protect AI’s Radar policy engine is used to create those policies across all of their AI environments to help them build more secure, safe, and compliant AI.”
Salesforce Ventures’ Investor Emily Zhao on why enterprises may be slow to adopt AI tools…
“A majority of industries are still at the AI proof of concept stage. I think one of the big reasons for the slow adoption curve is challenges related to hallucinations, data privacy, and organizational alignment. We need to resolve those concerns and fine-tuning is one way to do that. Salesforce Ventures is interested in companies building products to help customers fine-tune whatever models they want to leverage for specific values they care about so the outputs are more aligned with what they want. These companies should prioritize user trust and data privacy.”
Hugging Face Head of Sales Bassem Asseh on AI innovations he expects in the near future…
“We’re starting to see use cases around image, audio, and video. What’s coming in the next few months is what the industry calls a multi-modal model. This means a model that is able to take into account language, image, audio, and video and generate an output.”
Hugging Face Head of Sales Bassem Asseh on balancing model cost and model quality…
“Thanks to open source AI, enterprises have access to models that can be efficient when it comes to fine tuning models with the customer’s processes and datasets. Fine tuning a huge model takes too much time and money. So enterprises are turning to smaller models that are able to focus on specific tasks in specific contexts. For example, if you’re building a feature that summarizes text, you get a model that is built for summarization and fine tune it on your data so it’ll be even better in generating content related to your context and also more cost effective.”
Corporate partner and IAB feedback
Corporate partners and IAB members who attended our innovation forum raised interesting questions and provided valuable feedback to our founders. We’ve collected a sampling of their most salient insights here:
- On the challenges of enterprise AI software buying: “The biggest obstacle is determining the right AI solution to solve for a given task. The past year has been characterized by experimenting, exploring, and trying to evaluate where in the organization we can deploy the right solution for the right task. At enterprise scale it can be hard to keep up with the rate of innovation in the market. We have to be very selective about what we’re evaluating and how many resources we’re putting behind exploring whether it’s the right fit for our organization. There’s a lot that goes into deep diving what the solutions are and if there’s actual feasibility behind exploring them further.”
- On business processes that can benefit from AI and ML technologies: “Digesting regulatory filings and large sets of data, transferring knowledge amongst workers, performing customer research by scanning social media, and writing more efficient code.”
- On measuring the ROI of AI-driven data analytics initiatives: “We measure by assessing how current systems behave prior to and after AI implementation coupled with data and analytics. KPIs are critical for determining ROI and have to be defined prior to establishing a strong AI presence.”
- On ensuring data privacy and eliminating bias when utilizing AI solutions: “We’re focused on data privacy using traditional data privacy systems as well as ethics and bias tools that we then integrate with AI-based solutions. We ensure accuracy and fairness via constant monitoring and measurement through traditional systems and parameters in preset models.”
- On leveraging AI while still maintaining a human touch: “The balance is a delicate one. We aim to make sure each focus area has human involvement.”
This Innovation Forum was part of an ongoing series that aims to connect Salesforce Ventures’ corporate partners and thought leaders with members of our portfolio. We’re energized by the current rate of innovation in the market, and look forward to nurturing this innovation through an ongoing series of conversations with members of the Salesforce ecosystem.