The Hidden Costs of AI as a Service: Unpacking the Pitfalls of Subscription‑Based AI Tooling

In September 2022 I left DataRobot — the titan of automated machine learning — carrying with me a decade of insight and a profound sense of burnout. Over the following year I explored new passions and reflected on the pitfalls that plague enterprise SaaS solutions for ML and AI. I’d begun my journey in 2012, smitten with autoML and the dream of democratising a pioneering technology; I ended it by confronting the stark realities of market adoption.

As I navigated these experiences one truth became painfully clear: AI‑as‑a‑Service (AIaaS) — a clumsy mash‑up of enterprise SaaS subscriptions and “AI platforms” — has promised investors and customers far more than it’s delivered. The idea of universally accessible, plug‑and‑play AI is compelling, yet the practicalities of subscription platforms built for everyone have consistently fallen short. The disconnect stifles innovation and burdens organisations with hidden costs few acknowledge.

It’s not that I didn’t see this coming.

Democratisation is a four‑letter word

I lost my belief in fully democratised AI/ML back in 2015 when Jepson Taylor and I launched an early autoML solution and 120 people submitted datasets. Most of them broke our pipeline. One submission was an Excel file where the data started a few rows down and a couple of columns over — with a logo inserted. Jepson looked at me:

“Gonzo, what f*** are we supposed to do with this?”

After years of consulting I caught platform‑fever. Over the next seven years I chased myriad paths to “democratisation”, but none delivered the turn‑key setup and modest‑effort maintenance cycle required for an enterprise SaaS win‑win.

Re‑entering the fray at Data Kinetic, it’s time to unpack these challenges. The lessons of the past have shaped a new vision for AI integration — one that prioritises genuine utility over ubiquitous access. Let’s explore why traditional AIaaS struggles and how a focused, strategic approach can better serve enterprises.

POCs, pre‑sales exhaustion, and the Disney+ effect

Looking around, there’s a clear winner in enterprise AI: “bolt‑on, good‑enough ML” sitting beside data storage and processing. Everything else trails by a wide margin.

Standalone AI/ML products plunge teams into endless cycles of POCs and procurement, where the best minds spend days selling rather than innovating. Tool abstractions rarely accelerate delivery; they simplify technical steps yet leave the hard work — problem framing, change management, focus and investment decisions — to users motivated mostly by FOMO.

The average enterprise SaaS sales cycle is ≈ 9 months. New LLM‑based platforms and the “X‑for‑Y‑with‑LLMs” startups (founded FY 2023) tout key innovations — except in the business model they still rely on.

Current LLM market through the lens of the Gartner Hype Cycle

Data or compute wins the war

Value‑priced AI SaaS faces a brutal truth: the real winners either own the data or provide the compute. NVIDIA, Databricks, Snowflake, and the hyperscalers (AWS, Azure, GCP) succeed by embedding “good enough” ML and hosting open‑source tooling — driving prices down to the cost of compute.

Compute costs are falling, yet remain significant during training and inference, especially for LLMs. Once you’ve paid for compute there’s little margin left; vendors without data or infra are squeezed.

If you buy AI from a non‑cloud vendor you’re likely betting on a partner with no clear path to profitability. Their value‑pricing had better be pegged to data if they hope to survive.

Good tools are great, but the subscriptions add up, so you need to know where the budget yields the biggest bang.

How many streaming services will you pay for?

If you have kids you probably subscribe to Disney+; if you have a sports fan you still pay for cable TV. Disney+ ≈ $20 / month. Cable ≈ $200 / month.

Your cloud/IaaS bill is cable TV in this analogy. Disney+ and Netflix are your must‑have SaaS (Salesforce, ERPs). Only a few budget $/headroom slots remain for everything else.

You’ll probably buy Netflix, but you need a very niche reason to pick up Paramount+. Everything else had better be bundled. AI is Hulu or Prime Video — easier to stomach when included with something else.

Subscription AI SaaS sucks

The issue isn’t the fee itself; it’s the circus around SaaS: interminable pre‑sales, pricey annual contracts, low adoption, high churn. Organisations must streamline Applied‑AI adoption so they can binge on specific needs, cancel or pause anytime, and grow both operational and organisational capacity.

That’s what we’re building at Data Kinetic: simple, effective access to AI tech, talent and advice on‑demand. Industry‑specific strategy, portfolio‑management, and advisory services, backed by a platform‑agnostic catalogue of outcomes‑as‑a‑service that run in your environment. No hype — just pragmatic experience proven to create value and drive innovation.

And with that framing I’m no longer burnt out. I’m excited to actually help.