Operational AI Readiness, the show. Short episodes, one obstacle each: what goes wrong, what good looks like, and one thing you can do this week. Episode 1 of 12.
Almost every AI tool at work is introduced with a gentle label: assistant, copilot, decision aid, “just a suggestion.” In this first episode (about 19 minutes), Viktoria Bakos, author of Operational AI Readiness, asks what that label actually tells you, and why the honest answer is often “not much.”
What the episode covers
The example is the Allegheny Family Screening Tool, used since August 2016 at a Pennsylvania county’s child welfare hotline to help decide which reports of possible neglect are investigated. The county says plainly that the algorithm was never meant to replace human decision-making (Allegheny County Department of Human Services [ACDHS], 2018, p. 4). Yet by its own 2024 description, the highest scores appear on screen as a “high-risk protocol” that defaults the decision to investigate, while supervisors keep the power to override, “and do so regularly” (ACDHS, 2024, p. 2). We chose this case not because the county did something uniquely wrong, but because it has published more about how its tool works than most organizations ever do.
The episode then walks through why soft labels are so tempting, what Viktoria calls automation laundering, and the five questions from Chapter 3 that make honest classification hard to dodge.
Key idea: decision-influencing AI
An AI system influences a decision when it shapes what people decide through scores, rankings, summaries, defaults or suggested wording, even if a human makes the final call. It matters because most AI that touches your job application, benefit claim or bank account will be described as advisory, and that description should not lower the protections around it.
Three key takeaways
- A label is not a classification. “Human presence does not erase AI influence” (Bakos, 2026, p. 29). A ranking, a summary or a pre-set default can steer a decision long before anyone clicks the final button.
- Classify by what the workflow can affect. Ask five questions: does it influence a decision or action; could that affect money, access, employment, eligibility, security, legal position, safety or rights; can it trigger or speed up downstream action; will people foreseeably over-rely on it; would you need to reconstruct what happened? If the answers cluster around yes, it needs a stronger control path (Bakos, 2026, p. 32).
- Override numbers are evidence. Allegheny reported that in the tool’s first full year only 63 percent of “mandatory screen-in” referrals were actually screened in (ACDHS, 2018, p. 10). Counting how often people disagree with a tool is how you tell a human who decides from a human who signs.
The one thing to do this week
Pick one AI tool your organization calls an assistant or “just a suggestion.” Answer the five questions for it in writing, treating “not sure” as yes. Then find its defaults: anywhere it pre-fills a choice, ranks a queue, tags something “low risk” or drafts the words of a decision. That is where its influence lives.
If AI is used on you: you can always ask whether any automated score, ranking or summary was used in your case, and whether the person reviewing it could overrule it. “A human decided” should be the start of the explanation, not the end.
Read the related article
For the written version, with the full record and its limits, read Is Your AI Really Just a Suggestion? Classifying Decision-Influencing AI.
Go further. Operational AI Readiness is FairByDesign’s free field manual for AI that is already live. Chapter 3, “Classifying Decision-Influencing AI Without Lying to Yourself,” walks through the five questions with worked examples from hiring, payments, procurement, customer disputes and security. Read the chapter free. The first 10 readers who send feedback can email their own questions to info@fairbydesign.org, and the expanded Premium edition is free for the first 10 with code FIRST10PREMIUM.
Your next step: take the free readiness assessment on one AI tool you use. It opens with the same five classification questions, where “not sure” counts as yes, then shows your weakest control domain, three next moves and the chapter to read. Nothing is stored; it is a self-assessment, not legal advice or an audit.
FairByDesign is reader-funded. If this helped, get the free Mercury Brief or see ways to support. Support never buys editorial control.
References
Allegheny County Department of Human Services. (2018). Allegheny Family Screening Tool: Frequently-asked questions (Updated August 2018). https://www.alleghenycountyanalytics.us/wp-content/uploads/2018/10/17-ACDHS-11_AFST_102518.pdf
Allegheny County Department of Human Services. (2024, April). Summarizing recent research on predictive risk models in child welfare. https://www.alleghenycountyanalytics.us/wp-content/uploads/2024/05/24-ACDHS-04-Predictive-Risk-Algorithms.pdf
Bakos, V. (2026). Operational AI readiness: A practitioner guide to governing AI that is already live (Free ed.). FairByDesign. https://fairbydesign.org/operational-ai-readiness/
Ho, S., & Burke, G. (2022, April 28). An algorithm that screens for child neglect raises concerns. Associated Press (via KTVZ). https://ktvz.com/news/2022/04/28/an-algorithm-that-screens-for-child-neglect-raises-concerns-2/
Leave a Reply
You must be logged in to post a comment.