TL;DR
- Generative AI did not legitimize machine learning spending. It created urgency for enterprises to understand their own data, and harmonization, context and orchestration budgets opened as a result.
- Industrial AI is whitespace. The large industrial systems companies build extraordinary businesses and are not yet where they need to be on data applications.
- Pilots are not failing because models fail. They fail because the surrounding engineering, meaning evaluation, tool calling, data cleanliness and cataloging, is manual work that few teams have done.
- Professional services in AI are not custom software. They are data engineering, evaluation and workflow orchestration, which is the real systems integration work.
September was another busy month. I met 100+ potential clients, partners, and ecosystem players across power, utilities, datacenter, electrical engineering, property assessment, and industrial domains.
While my real eyes are astigmatic, in the world of AI, I am going toward 20/20. Each time I hit the road, I come back with a new understanding. Let me share them all with you:
- Infra spend is opening up I used to think that the GenAI wave has legitimized spending on machine learning and statistical modeling. That is not true. That is a second order effect. GenAI has created an urgency in enterprises to understand their own data. Harmonization, Context, and Orchestration spends are getting legitimized.
- Industrial AI is a whitespace With due respect to the massive businesses industrial systems companies like Siemens and ABB have created, they are not quite there when it comes to data applications. They push boxes and reluctantly, some APIs and digital twins. Unlocking the value of that data is still a greenfield endeavor.
- Midmarket Industrial needs new solutions 'Make in America' is a wave. The mid-market manufacturing company has not even started logging data, let alone use generative AI. This is an unspoken opportunity.
- Ignore the MIT report but pay attention to 'why' failures happen Pilots are not failing. The MIT report is wrong. Their own data shows that 30% of pilots go to production. But, product vendors are overselling their place in the enterprise. Pilots fail not because LLMs fail. The surrounding engineering that ensures consistent tool calling, evals that set a robust guardrail system, data cleanliness, accessibility, structured cataloging, and modeling of enterprise data is a very manual process and the failure of it shows when you do pilots. That most organizations (including the AI app layer vendors) don't understand how to build data foundations is a double-whammy.
- Speed and success comes from orchestration Enterprises are realizing that apps and general tooling provided through captive contracts by hyperscalers are limiting them. But only some realize that point solutions may also fail for the reasons in Point 4. Very few vendors bring together context layers, pre-built LLM-based applications, and custom workflows, and orchestrate all of them together. 4 and 5 are the reasons Moative exists.
- Professional Services go beyond custom build Enterprises have had a bad experience with professional services. In the past, professional services meant custom software and they were fleeced to death by 'change requests.' But professional services in AI transformation are about data engineering, tool calling, evals, and workflow orchestration. They are the true system integration work.
If you are thinking of an enterprise AI pilot, your checklist should be based on your data, context readiness and your vendor set should be a mix of point solutions, infra tools, and orchestrators/integrators.
Enjoy your Sunday!
Where Moative fits
Moative exists because of the last two observations above: point solutions fail for the same engineering reasons that platform tooling does, and almost nobody assembles the context layer, the applications and the workflows into one orchestrated whole. That is the work we take on, which looks less like custom software and more like systems integration for the AI era. In healthcare revenue cycle management it means the evaluation harness and the data cataloging come first, and the visible application arrives on top of them.
Frequently asked questions
Why are enterprises spending on data infrastructure now?
Generative AI created urgency around understanding internal data, which opened budgets for harmonization, context and orchestration that had been difficult to justify before. The spending is a second order effect of the hype rather than a direct result of it.
Is industrial AI an underserved market?
Largely yes. The major industrial systems companies have built substantial businesses around equipment, with interfaces and digital twins attached. Turning the resulting data into applications is still mostly unclaimed ground.
Why do enterprise AI pilots really fail?
Not because the models underperform. They fail on the surrounding engineering: inconsistent tool calling, missing evaluation and guardrails, unclean and inaccessible data, and the absence of structured cataloging. That work is manual and most teams have not done it.
Are AI point solutions safer than platform contracts?
Not inherently. Captive platform contracts limit flexibility, and point solutions fail for engineering reasons of their own. Neither substitutes for owning the context layer and the orchestration between them.
What should professional services deliver in an AI program?
Data engineering, tool calling, evaluation and workflow orchestration. If the engagement is producing custom software and change requests, it has reverted to the old model and will produce the old results.
What should be on an enterprise AI pilot checklist?
Your data readiness and your context readiness first, then a vendor set deliberately mixed across point solutions, infrastructure tooling and orchestration. A checklist that starts with model selection is starting in the wrong place.
