I'm often asked the question, "What skills should I develop to advance my career?" The asker may be in operations, customer success, or go-to-market, and obviously the answer varies by function. But a common reply that I'll give goes something like: "It will become increasingly important that you are good with data."
When I shared this during a recent roundtable, someone replied, "That makes me nervous." And I understand that sentiment, because not everyone can be a data scientist or machine learning engineer. The good news is that you don't have to be, and you don't have to have access to an ML engineer, either.
New products are democratizing machine learning so that everyday analysts can use predictive models to help their business partners make decisions. Want a continually updated prediction for your net dollar retention? Your data team, or the data person on your team, can now give you that, and help you understand the underlying model.
I sat down with Tristan Zajonc, founder of Continual, to learn how he's making machine learning more accessible and operational.
What motivated you to start Continual?
I've been working in the data and machine learning space for the last 15 years. After Cloudera acquired my last company, it became the basis for the machine learning platform that they sold to some of the largest and most sophisticated global enterprises. Everyone was excited about its potential to accelerate AI and ML. But customers also struggled to use it to make an impact on their business. The problem was moving predictive models from research to production, and also scaling beyond one or two use cases. Models needed to be continually maintained and the predictions continually updated. But each model required a team of data scientists and machine learning engineers to build and maintain. The result was a level of complexity that undermined the potential for AI to transform businesses.
I started Continual to solve this problem. I wanted to make the infrastructure invisible, automate as much as possible, and make operational machine learning and AI accessible to everyone.
Why hasn't someone solved this problem already?
First, we're at a critical moment in time, when the data platform is being transformed.
Twenty years ago, the data warehouses ran on premise running on servers, and couldn't handle much data or AI’s complex workloads. With the rise of Big Data platforms like Hadoop, you could handle massive amounts of data and AI workloads, but it introduced massive complexity. You could say we went from the era of Big Data to the era of Big Complexity.
Now we're seeing the rise of cloud data warehouses like Snowflake, which put SQL back at the center, eliminated management complexity, and made data more accessible across the company. This in turn has led to a reimagination of data tooling and the rise of a broader modern data stack ecosystem. Tools like dbt [where Allison is on the board], which is taking over data engineering, are a product of this transition to data warehouse centric architectures and a search for simplicity. But there's a gaping hole in this new ecosystem for predictive analytics and AI. We're looking to take the lead in that area, reimagining AI and ML for the modern data stack and modern data teams.
So what does this look like? We think that we have a distinctive point of view, which is that operational machine learning should be centered on the data and fully automated. All you should care about is what data is feeding into your predictive model. That’s what will make your predictions better, and what is the unique asset of most businesses. You shouldn’t have to care about infrastructure, data pipelines, or machine learning algorithms. All of that complexity can be automated away.
I imagine that's why you named your company Continual?
Yes, exactly. Your data is continually changing, and we need an ML platform that embraces that reality and automates the continual maintenance of models and predictions.
Let's say I'm a Chief Customer Officer at a SaaS company, looking to predict my retention rate. Being able to forecast this number accurately is typically a core part of the job description of a CCO, but it’s often extremely difficult to do that. Can I benefit from Continual?
Yes, absolutely. Continual is the predictive layer on top of the customer and operational data that's sitting in your data warehouse. For example, you may have gathered data on customer churn by cohort, categorized customers into high activity users and low activity users, etc. Then you can answer questions such as, which users are likely to churn in the next 30 days and why?
Unlike ad hoc analysis, models in Continual continually adapt to ensure that the latest data is informing the predictions. You can maintain the churn predictions in your data warehouse, so that you can leverage them in your workflow to take action to prevent the churn (e.g. send an email with a discount to a high-churn-risk user), and leverage them in your BI tools to report your forecasts to your CEO and board.
What we’ve found is that once you make ML easy, you often want to ask far more questions of your data. For instance, you could also predict expansion or contraction. Or how is usage likely to grow (or decline) in a particular account? Or how many seats are they likely to add (or subtract)? All these are variations you might want to predict if you originally think you just need to understand “churn.”
And predicting customer outcomes is just one use case for Continual. There are plenty of other KPIs that you can forecast, including sales, inventory, and others depending on your industry and function.
When am I ready to start using Continual?
You’ll need to have a cloud data warehouse, since that’s how we connect to your data. And you’ll actually need data. Typically you need at least a thousand data points to start using Continual, which is actually a quite low number when you think about it. For example, that might mean a thousand user-months, which Continual could analyze to understand those likelihood to churn or expand.
You don’t have to have all the data on potential predictors of churn to get started. You can start with a few simple ones – e.g. usage, support tickets – to build a baseline model. Then you can improve the model over time by adding more signals.
What other tools exist to solve this problem?
In terms of horizontal ML platforms, there are two segments of tools. First, the traditional machine learning platforms, which I’ve spent the last ten years trying to build in previous companies. These tend to be very code-centric. As a result, few companies can put more than one or two models into production, in the sense of leveraging the models for continuous, everyday decision-making in the business. So these platforms are incredibly powerful and flexible but suffer from the need for large and highly skilled teams to support them, plus expensive infrastructure to manage.
There’s a second segment of products that I’d call point-and-click AI tools, or no-code AI tools. They imagine a world where the business user can solve their own predictive analytics problems through a UI. The problem with this is that those models never make it into production. Most mission-critical data work within enterprises is done by professional data teams. They want tools that fit in with their existing workflow, and they value governance and robustness.
So customers have to choose between complicated platforms and no-code products that don’t deliver.
We’re looking to create a third category – a data-first operational AI platform – that is governed by software engineering best practices but without the complexity of traditional machine learning platforms. Those best practices include version control, the separation of production and development, declarative management of operational systems. But unlike traditional ML platforms, the core of Continual is focused on modeling your business data, not writing code or pipelines. We think this “data-first” mindset is the future of operational AI/ML.
What’s next for you and Continual in 2022?
We just launched our public beta and announced our seed funding from Amplify Partners. So 2022 is going to be a big year for us. We’re most excited by seeing the use cases everybody is bringing to the platform. You can literally sign up and try it out in 10 minutes.
More broadly, I believe machine learning and AI will be among the most exciting technologies that will unfold over the next decade. It’s honestly hard to even imagine where AI will all end up and the full range of impacts it will have on business and society. It’s fun to be able to participate in this moment and try to turn some of the hype into reality.
***
If you’d like to learn more about Continual, check ‘em out here.
If you'd like to hear from more founders like Tristan every couple of weeks, sign up for my newsletter here:
If you’d like to learn more about what I’m up to, see my website here.