We’re back with more from LiveRamp Ventures portfolio company Frame AI, which helps to apply machine learning and natural language processing to unstructured customer feedback.
After Part 1 of the Q&A with Frame AI’s co-founder and CEO George Davis, read on for George’s thoughts on risks, the enterprise and marketing data stacks, and more!
Q: You obviously believe there are great opportunities in AI adoption. What are the risks?
It’s tempting for companies to treat “AI” as some sort of core capability that needs to be acquired centrally. This can lead to intermediation, where implementing the simplest project for a business team requires fighting for and scheduling resources from a single “center of excellence”. Doing this both suffocates projects early, and prevents the rapid feedback and iteration loop that make AI projects ultimately successful.
In reality, AI is a whole box of tools – and finding the opportunities in different parts of the company requires empowering experts to explore use cases as directly as possible. And what makes this generation of AI exceptional is how accessible it is to non-experts! There has never been a better time for “data democratization”, and organic adoption of AI tools. I always advise CIOs and CTOs to focus on empowerment and training, and where necessary embedding resources at business teams.
The counterpoint to this “let 1,000 flowers bloom” approach is that it’s easy for AI projects to emerge around shiny ideas and justified as exploration, without a clear case about how current business objectives will be impacted. The best thing you can do for the success of AI projects is to orient them around an incredibly clear ROI case, and invest early in the instrumentation that will let you measure the impact of the project. This not only holds the project to account externally, it orients your teams around iteratively improving the metrics they have promised to work towards.
We’ve tried to build Frame AI around both of these principles. We are an AI tool that is operated by front-line customer-facing teams, and speaks directly to the metrics they already use as KPIs. AI is a means to an end, and business users drive the feedback that helps us reach that end.
Q: Let’s talk about AI for the Enterprise Data stack in particular. At this point we’ve had over a decade of heavy investment in datalakes, CDPs, and other data orchestration. What does AI add to this equation?
It’s funny – when you track down that now-cliché quote “data is the new oil”, you’ll see that as far back as 2006 marketers were emphasizing that data must be refined through analysis to be useful. But in spite of that, the industry has been far more successful at capturing data than making it consistently useful. Most modern enterprises collect mountains of data, but use a tiny fraction of their overall columns and rows in profitable business operations.
A big part of that is that our best solutions to “refining” data are still very expensive. Even with the most modern data science tools – ETL and reverse ETL pipelines, ML ops platforms for performing bespoke inference tasks, etc. – it still takes multiple data engineers weeks or months to identify and extract the most relevant data for a new analysis task or business automation application. This has a real chilling effect on which projects get greenlit. Worse yet, resource constraints mean that a lot of projects don’t get iterated on and maintained consistently after launch, leading to “rot” in data pipelines.
The net effect is that in spite of enterprise-wide data capture initiatives, many teams live and work within small data silos where they can invest continuously in making the data they need available. This is especially true when it comes to unstructured data – call transcripts, emails, survey responses, and other interactions that aren’t machine readable in the first place. And often, this is where the most surprising insights, which customers spent the most effort to provide for you, are buried.
We founded Frame AI because we think AI can unlock a tremendous value in stagnant data. General purpose language models can be adapted very, very quickly to extract specific data signals from text, and even from structured data like customer profiles and event streams. This lets us operate a very rapid feedback loop with business teams, helping them explore the factors that are potentially impacting costs, growth, and risk.
Everything I’ve described so far is about the analytic use cases for AI – how can we help extract value from mountains of data. But new advances in generative AI are going to change how data is consumed around the enterprise. It’s going to make it easier to reorganize support data in a way that Product or Marketing can understand. That’s something I’m excited about and which we are investing in heavily.
Q: How does working with LiveRamp boost the work Frame AI is doing?
Frame AI’s mission is to help businesses build an understanding of their customers across all channels. That mission can’t even begin until we can reliably tell one customer journey from another.
LiveRamp’s identity framework is a powerful standard against which to align our identity resolution. By enabling marketing to real people, working with LiveRamp helps marketers to connect their customers’ omnichannel journeys, connecting all of their data from every digital and offline interaction.
Q: This latest AI revolution is happening in the middle of already-seismic shifts in the marketing landscape. Do you see the relevance of AI to how companies shift from relying on third-party to first-party and federated data?
Third-party data networks provided us with so much information about consumers, we could afford to ignore messy data and use it inefficiently. The reality of the first-party world is that data is higher quality, but there is much less of it. This makes it more important than ever to squeeze everything you can out of each customer interaction.
It shouldn’t be satisfactory to record the fact that a customer “viewed a page about product X” – that record should be annotated with which topics on that page spent most of their time in view. Customers shouldn’t simply “complete a marketing chat related to pricing” – the key inquiry should be categorized. Each of these facts should be readily available for both historical analysis and profile building.
Another important change is that full lifecycle marketing is becoming more crucial. Both because high initial customer acquisition cost (CAC) means every business must maximize lifetime value (LTV), but also because one company’s post-sale data may be useful pre-sale data for others in a data federation. We’re excited to partner with LiveRamp in addressing these use cases.
Q: Any parting advice for execs on how to weather this AI wave?
Focus on your goals. Focus on the data and distribution that make your business, and your role, unique. Get your hands dirty – this is the most accessible technical revolution any of us have lived through! And help your teams do the same.
To learn more about LiveRamp Ventures, visit the website.