Before You Automate, Count What You'll Lose
mlj.solutions
mlj.solutions Většina lidí si myslí, že automatizace zákaznického servisu jednoznačně sníží náklady. Běžné dotazy jako změnu adresy nebo reset hesla bot skutečně vyřeší levněji. S nimi ale často zmizí i člověk, co uměl poznat výjimku: reklamaci po výpadku, nestandardní smlouvu, citlivý případ. Ušetřené náklady tak mohou stát firmu odolnost pro situaci, kterou žádná tabulka nezná. Tohle je potřeba zvážit.
View on InstagramWhat the study is about
The usual automation ROI calculation overlooks risks that never show up in a spreadsheet. The study presents the PHP-AIO protocol, which quantifies these uncounted risks role by role and leads to an auditable decision on whether to automate a given job.
Key numbers
4
categories of overlooked risk
Loss of the tacit knowledge of experienced people, lower organizational resilience, regulatory exposure, and erosion of social and institutional capital.
5
gates in the decision protocol
The decision passes through five consecutive gates and ends with a composite check of overall risk.
40
fields in the input schema
Scoring relies on a structured input with 40 fields, each of which is observable and verifiable.
What this means
- The economics of automation are incomplete, classic ROI leaves out long-term systemic risks that can call the entire saving into question.
- Decisions are made per role, the protocol assesses risk for individual roles, not for the company as a whole.
- Risks add up, the automation debt formula shows how decisions about individual roles accumulate in multi-step processes.
- Only human oversight neutralizes the warning, the warning about accumulated debt disappears only when a regulator requires that a human stays in the loop for the decision.
- Decisions can be audited and tailored, an institution can set weights and thresholds according to its risk appetite, but the default configuration is based on expert estimates for financial services.
Our take
Automating what you can doesn't mean automating what you should. It's always necessary to assess what will be lost after deployment and what impact that will have.
Michal Dobrovolný
Founder & CTO
Zdroj
de la Chica Rodriguez, J. M., Rodriguez Arias, J., & Chouliaras, S. (2026). When Not to Automate: A Formal Protocol for Human Preservation in AI-Optimized Organizations. arXiv:2607.15944. https://arxiv.org/abs/2607.15944
Don't miss these posts
When More Agents Pay Off and When They Don't
More AI agents don't necessarily mean a better result. A study shows that what matters is how much information is lost when it is passed between agents.
AI Remembers a Claim Even When You Tell It It's False
When a model is fine-tuned on documents that deny a claim, it often learns the claim as true anyway. The authors named this effect Negation Neglect.
The Skill That Explains Anything to Anyone
ELI5 is an open-source skill for Claude Code that adapts explanations to the audience. With the skill it passed 10 of 12 criteria, without it only 5.
Want to implement AI in your company?
We'll go through your needs together and propose an approach that delivers measurable results.
Write to us
