We picked a low-code ML platform specifically because we didn't have a dedicated data science team and needed something fast. "No code, faster time to value" was basically the pitch on every vendor call. Five months later we finally had a model in production, and most of that time had nothing to do with the model itself — it went into wiring the platform into our existing data pipeline and getting IT, security, and the business team to agree on what "done" looked like.
Turns out that's not a us problem. G2's State of Low-Code Machine Learning in 2026 report analyzed 3,400+ verified ML platform reviews across five categories, and found that low-code ML platforms average 4.5 months to go live — the slowest deployment time of any machine learning category G2 tracks. Slower than full data science platforms. Slower than MLOps platforms. Slower than data labeling tools by 2.6x. The category sold on the promise of speed is, empirically, the slowest one to ship.
If you're at an enterprise it's worse: enterprise buyers average 5.47 months to go live versus 2.75 months for small businesses, and even after deploying, enterprises only put 35.5% of their licensed seats to actual use, against 49% at small businesses. A lot of adopted-in-name-only tooling out there.
The vendors G2 surveyed (Pecan AI, Acodis, Minitab, Kili Technology) all landed on the same root cause, and it isn't model quality:
"The real obstacle vendors are not talking about is the implementation work. Everyone expects automation to happen in a few clicks. But implementation takes time, on one side to fine-tune models, on the other to fit the model usage within a much bigger process."
That's Philippe Cayrol, Chief Revenue and Strategy Officer at Acodis. Another vendor exec put it more bluntly about enterprise deployments specifically:
"Currently, the biggest hurdle is the need for 3-4 teams to collaborate. For smaller organizations, this obstacle can be overcome. For larger organizations, it requires consultants, project managers, Forward Deployed Engineers, and very, very strong executive sponsorship to bring an idea all the way to a model into production."
The report also pokes a hole in the "no-code means no data scientist needed" pitch. When G2 asked the vendors how confident they were that non-technical users successfully deploy production-ready models without help, the average answer was 3.25 out of 5. In practice, a technical owner still builds and integrates the pipeline; the business side defines requirements and validates outputs, not the other way around.
None of this means low-code ML tools are bad — reviewers actually rate them highest on "Meets Requirements" (8.78/10) of any attribute measured. The tools do what they say. The gap is entirely in setup and admin, where the same reviewers score them lowest (8.44 and 8.43 respectively). If you're evaluating one of these platforms, budget your timeline around data readiness and integration work, not around how fast the demo looked.
Worth a full read if you're scoping a low-code ML rollout: https://learn.g2.com/low-code-machine-learning
Top comments (0)