In September of 2022 I left DataRobot, the titan of automated machine learning, carrying with me a wealth of insight about the industry I loved for over a decade and a profound sense of burnout. I was lucky to have over a year to explore new passions and reflect on the pitfalls that plague Enterprise SaaS solutions for machine learning and AI. I started my journey in 2012, immediately smitten with autoML and the allure of democratizing this pioneering technology and I was ending my journey confronting the stark realities of market adoption.
As I navigated through these experiences one truth became painfully clear: the prevailing business model, AI as a Service, a clumsy combination of Enterprise SaaS subscriptions, the lucrative dream business model and Enterprise AI platforms promised both investors and organizations more than they’ve been able to deliver. The concept of making advanced AI tools universally accessible and easy to implement across diverse industries is compelling. Yet, the practicalities of achieving this—through subscription-based platforms designed to serve a broad customer base—have consistently fallen short of expectations. This disconnect between aspiration and reality not only stifles innovation but also burdens organizations with hidden costs and complexities that go largely unacknowledged.
It’s not that I didn’t see this coming.
I lost my belief in fully democratized AI/ML knowhow in 2015 when Jepson Taylor and I launched an early autoML solution and 120 people submitted datasets. Most of them broke our pipeline. One of the datasets was an Excel spreadsheet where the data started a few rows down and a couple columns over from the top and it had a logo inserted. I looked over at Jepson: "Gonzo, what f*** are we supposed to do with this?"
At the time I felt I had gone as far as I could helping organizations directly. My consulting practice was all consuming and like so many I caught platform-fever. Over the next seven years I would see and try so many paths to democratization, but none of them ever delivered the kind of turn-key setup and modest-effort maintenance cycle required for Enterprise SaaS to materialize a win-win for both organization and strategic partner.
As I re-enter the fray with a renewed mission at Data Kinetic, it's time to unpack these challenges. The lessons learned from past mistakes and the insights gained from successes have shaped a new vision for AI integration in business—one that prioritizes genuine utility over ubiquitous access. Let’s explore why the traditional AIaaS model struggles and how a more focused, strategic approach can better serve enterprises embarking on their AI journey.
Looking back and assessing where we are, there’s a clear winner in Enterprise AI. Bolt-on, good-enough ML, sitting besides data storage and data processing has proven to be the option the market wants and everything else trails by a wide margin.
Standalone AI/ML-focused offerings plunge teams into an endless cycle of POCs and procurement woes, where our best technical minds spend their days not innovating, but rather, navigating customer problems and bolstering operational capacity. The meaningful abstractions in these tools rarely accelerate delivery of a solution, rather they simplify technical processes and leave the really difficult work of deciding where to focus, how much to invest, framing the problem solidly, and architecting and executing the change management to people whose primary motivation to use AI is their boss’s or board’s fear of missing out.
The average Enterprise SaaS sales cycle is 9 months. The newer LLM-based platforms and the “X but for Y with LLMs” startups all got going about a year ago (FY 2023) and are now entering the market with what appears to be some key innovations in several key areas but one: they’re relying on the same business model that has failed Enterprise AI for over a decade.
Current LLM market through the lens of the Gartner Hype Cycle
Value-priced Subscription AI SaaS is facing a significant challenge, regardless of the AI modality involved—be it deep learning, transformers, or LLMs. What we have observed, and expect to continue to see, is that the true market winners are those who either control the data or provide the computing power. Companies like NVIDIA, Databricks, Snowflake, and major IaaS providers such as AWS, Azure, and GCP succeed by integrating "good enough" ML and AI. This often involves hosting or leveraging open-source offerings, which aligns with market trends that drive prices down to the cost of compute.
However, while the cost of compute continues to drop, it remains a significant expense during the training or inference phases, particularly for models like LLMs. After paying for compute, there is little to no margin left, presenting a financial challenge for those relying solely on providing AI processing capabilities.
If you are looking for AI solutions from a vendor that’s not a cloud provider you’re probably betting on a partner without a path to profitability and fewer and fewer paths to acquisition. Their value pricing had better be pegged to data if they hope to survive.
But good tools are a great thing. Sure, the subscriptions add up. So, it’s important to figure out where the budget will get the most bang for buck.
If you have children in your household you probably have Disney+ and if you have a sports-fan in your household you pay for cable TV. Disney+ costs roughly $20/month. Cable TV costs roughly $200/month.
Your cloud/IaaS bill is cable TV in this scenario. Disney+ and Netflix are your must-have SaaS, vendors like Salesforce and ERPs. This leaves only a few slots in both the budget of dollars and the budget of organizational bandwidth for other SaaS subscriptions.
| Cable TV (Even startups pay for cable) | Disney+ & Netflix | The others |
| NVIDIA AWS, Azure, GCP Other IaaSs |
Salesforce ERPs |
Databricks, Snowflake, Data Science tooling, collaboration platforms and apps, communication apps, dev tools, project management tools, etc. |
You’re probably going to buy Netflix but you have to have a very niche need to buy Paramount Plus. Everything else had better be rolled into some other offering and that’s the problem with standalone AI SaaS platforms and “X but for Y” SaaS subscriptions. AI is Hulu or Amazon Prime Video. It’s a lot easier to stomach when it’s included in something else like when these two services are included with Amazon Prime and Disney+.
The problem is not paying the subscription itself. It’s the whole circus that surrounds SaaS, with its interminable pre-sales, high-priced annual contracts, low adoption, and alarming churn rates. Your organization needs to be able to streamline the adoption of AppliedAI so that it can “binge” on specific AI needs, cancelling or pausing anytime, allowing your organization to focus on successfully growing BOTH operational and organizational capacity.
And that’s what we’re offering at Data Kinetic: simple and effective access to AI technology, talent and advice as and when it is needed. This includes industry-specific strategy, portfolio management and advisory services backed up by a platform-agnostic catalog of industry-specific machine-learning outcomes-as-a-service that run in your environment. These are not based on superficial hype but deep, pragmatic experience that's proven to create business value and drive innovation. And with this framing I no longer feel burnt out. I feel excited to actually help.