The Gap Between the AI You Bought and the ROI You Expected
You bought the licenses. You built the tools. You told your team this would change how the work gets done — and most of that investment is sitting unused.
The usual diagnosis is technical: a better model, a sharper prompt, a bigger platform. The evidence points somewhere else. The rollout isn't stalling because the technology is broken. It is stalling at the human layer — how a real person, mid-shift and already busy, learns a new tool, recovers from the first failure, and builds enough confidence to keep using it.
That is not a software problem. It is a psychological one. It is the problem I have worked on since before it had a name in this market.
The science behind the screen
I did not arrive at this through a thread last year. I studied it. At the University at Albany, in the doctoral program in Industrial-Organizational Psychology, I was a Carson Carr Scholar, and my graduate research asked a narrow question: what happens to a person's self-efficacy — their belief in their own capability — when you put a complex, unfamiliar technology in front of them and ask them to perform?
The finding, in plain terms: how someone experiences that first encounter shapes whether they walk away believing they can do it, or quietly conclude the tool is not for them. Attention during training is part of the mechanism — when it scatters, self-efficacy never forms. I worked this in Dr. Jason Randall's lab, on the same question the field is now asking about AI.
Where the science met real stakes
I did not leave this in the lab. I took it into one of the least forgiving data environments there is: public behavioral health.
As Data Coordinator for the Prevention Resource Center Region 11 (PRC 11)at Behavioral Health Solutions of South Texas — the Texas HHSC-funded center for substance-use and behavioral-health prevention data across the 19 southernmost counties — I authored the region's 2023 Regional Needs Assessment, the public report that anchors prevention planning for 2.2 million residents. A document like that does not hold up if the data underneath it is sloppy.
Healthcare is where technology adoption is hardest: the compliance is heavy, the resistance runs deep, and a failed system is not measured in budget — it is measured in people who did not get help. I treat HIPAA-grade data discipline as the floor, not a feature. And I never left the biology behind — I am one course from finishing a post-baccalaureate in Biological Sciences, because understanding the whole human system is the work.
An operator, not only a researcher
I do not only theorize. I deploy. I build and run AI systems for construction crews, field-service teams, and professional firms — not in a demo environment, but in production, with real owners, real payroll, and real resistance on the floor. That is where the frameworks get tested, and where they have held.
And I build them myself. These are not slideware: document and receipt automation with vision-OCR, language-model pipelines that read messy business data, retrieval over a company's own documents, WhatsApp and CRM integrations, and the dashboards that make the output usable — wired together on n8n, Supabase, and the Claude and OpenAI APIs. The science is the edge; building the systems is how I prove it ships. That applied practice has its own home, Rebel Minds OPS.
Long before the academic work, I ran the aftermarket department at an auto dealership and built it into a model profitable enough that other dealerships copied it. I know what a payroll feels like from the inside. That is not incidental to this work — it is why the psychology stays grounded in what a business actually carries.
My roots are in Mexico, and building a career in the United States from the ground up — across two countries, two intellectual traditions, two ways of earning trust — produced a bilingual, bicultural perspective you cannot assemble in a classroom. It lets me work with the English-speaking executive who needs a grounded framework, and with the Spanish-speaking business community that rarely gets this level of rigor in its own language.
Why you bring me in
No single line on this page is the reason. The reason is the convergence: an operator who understands payroll, a researcher who understands how people actually learn under pressure, and someone who has handled data where the stakes were real. The research explains what the deployments confirm. None of it was built for the AI moment — it just happens to be what the AI moment requires.
If that maps to something you recognize — a rollout that underdelivered, a team that never fully adopted, a gap between what you paid for and what you got — the next step is not a new tool. It is a diagnostic: finding where the chain is breaking, and rebuilding it.
Research & publications
The peer-reviewed work behind the framework — the science, not the slogan.
Peer-reviewed publications
- Arredondo, M. L. (2022). The neurotic wandering mind and self-efficacy during training. Master's thesis, University at Albany, State University of New York. https://doi.org/10.54014/DKAX-FS1Smost relevant
- Jou, J., Arredondo, M. L., Li, C., Escamilla, E. E., & Zuniga, R. (2017). The effects of increasing semantic-associate list length on the Deese–Roediger–McDermott false recognition memory. Quarterly Journal of Experimental Psychology, 70(10), 2076–2093. https://doi.org/10.1080/17470218.2016.1222446
- Jou, J., Escamilla, E. E., Arredondo, M. L., Pena, L., Zuniga, R., Perez, M., & Garcia, C. (2018). The role of decision criterion in the Deese–Roediger–McDermott (DRM) false recognition memory. Quarterly Journal of Experimental Psychology, 71(2), 499–521. https://doi.org/10.1080/17470218.2016.1256416
Selected conference presentations
- Randall, J. G., Hanson, M. D., Arredondo, M. L., & Jandrew, M. (2019). Predicting self-regulation failures in training. Society for Industrial and Organizational Psychology (SIOP) Annual Conference, National Harbor, MD.
- Arredondo, M. L., Randall, J. G., & Nassrelgrgawi, A. (2018). Cognitive and non-cognitive predictors of mind wandering.
También disponible en español: Por Qué Mario Arredondo
Last updated: June 2026