California’s AI Decision-Making Deadline Is Less Than 2 Months Away. Here Is What Retailers Need to Do in Q1
Published: Richard Mendoza 14 min read
California’s ADMT consumer rights take effect January 1, 2027. Retailers using AI for credit decisions, fraud detection, account access, or employee hiring need pre-use notices, opt-out mechanisms, and completed risk assessments before the deadline. Here is what it requires and what most retailers have not done.
In this Article:
- What California Means by Automated Decision-Making Technology
- Where Retailers Are Actually Covered: The Four Significant Decision Categories
- The Three Consumer Rights That Take Effect January 1
- The Risk Assessment Obligation That Was Due January 1, 2026
- The Inventory Problem Most Retailers Have Not Solved
- A Note on Applicability: Who CPRA’s ADMT Rules Cover
- What Your Organization Needs to Have in Place Before January
- Frequently Asked Questions
In less than 2 months from now, California consumers will have the right to receive advance notice, request access, and opt out of automated decision-making systems that affect them in consequential ways. The California Privacy Protection Agency’s ADMT regulations have been in force since January 1, 2026, and the consumer-facing obligations under those regulations take effect on January 1, 2027. For retailers already using AI or algorithmic tools in credit decisions, fraud detection, account access, or employment screening, January 1 is not a distant planning horizon. It is an operational deadline.
Most retailers are behind. A large share of them do not know they are covered. Some that have done a preliminary review have concluded their AI exposure is small because their initial inventory only captured the AI tools they knew about by name and missed the rules-based decision systems, applicant tracking configurations, and fraud scoring models that have been running for years without ever being labeled as artificial intelligence.
If your organization processes personal information of California consumers and uses any technology that replaces or substantially replaces human decision-making in a significant consumer interaction, the January 1 deadline applies to you. This article explains what the regulations require, where retailers most commonly underestimate their exposure, and what needs to be in place before the new year.
.gif?width=940&height=788&name=Copy%20of%20Stats%20-%20Blog%20(37).gif)
What California Means by Automated Decision-Making Technology
The CPPA’s definition of automated decision-making technology is deliberately broad. Under the finalized regulations, ADMT means any technology that processes personal information and uses computation to replace human decision-making or substantially replace human decision-making. The key phrase is “substantially replace.” A system substantially replaces human decision-making when it produces an output that a decision-maker uses without independent analysis of the underlying information, without the ability to override the result, or without understanding how to interpret the output in context.
The regulations also specify what ADMT does not cover: web hosting, domain registration, caching, data storage, firewalls, anti-virus software, spam filtering, spellchecking, calculators, and databases, provided those tools do not replace human decision-making. The list of exclusions is intentionally narrow. The design principle behind it is that tools affecting consequential outcomes are covered and tools enabling routine operations are not.
For retailers, the practical question is not whether a system is sophisticated. A rules engine that automatically declines a store credit application when a threshold is not met is ADMT. An applicant tracking system that filters candidates before a hiring manager sees the results is ADMT. A fraud scoring model that restricts account access without human review is ADMT. The regulations do not require a large language model or generative AI component. They require automated output that affects a decision.
Where Retailers Are Actually Covered: The Four Significant Decision Categories
The ADMT consumer rights obligations apply only when a retailer uses ADMT to make a “significant decision” about a California consumer. The CPPA defines significant decisions as those that affect a consumer’s access to or the terms of financial or lending services, housing, education enrollment, healthcare, employment, or access to essential goods and services.
For retailers specifically, the regulation explicitly excludes advertising, meaning a product recommendation algorithm surfacing items on a category page is not making a significant decision. A personalization engine ranking content is not covered. The exclusion is deliberate and applies broadly to advertising and marketing use cases.
What is covered for retailers falls into four practical categories.
-
The first is store credit and buy-now-pay-later decisions. Any automated system that approves or denies store credit, sets credit limits, determines repayment terms, or makes eligibility decisions for BNPL programs using consumer personal information falls within the financial services category. If the decision is made or substantially made by an algorithm, the January 1 obligations apply.
-
The second is fraud detection and account access restrictions. When a fraud scoring or identity verification system automatically restricts, suspends, or denies a consumer’s account access without meaningful human review, the access to essential goods and services standard potentially applies. Retailers whose fraud systems restrict access to online accounts, refund eligibility, or loyalty program benefits through automated rules should evaluate this category carefully.
-
The third is employee hiring and screening. Retailers that use applicant tracking systems with automated knockout criteria, resume scoring, or ranking systems that filter candidates before a human reviewer sees the application are using ADMT for an employment decision. This applies equally to the hiring of hourly store staff and corporate positions.
-
The fourth is consumer profiling for significant outcomes. High-risk profiling that automatically segments consumers in ways that affect their access to goods, services, or pricing for essential categories is covered. Dynamic pricing applied to essential goods and tenant screening in housing-adjacent retail are examples the CPPA has specifically identified as within scope.
The Three Consumer Rights That Take Effect January 1
Beginning January 1, 2027, California consumers whose personal information is processed through ADMT for a significant decision have three rights your organization must be operationally ready to honor.
-
The first is the right to pre-use notice. Before using ADMT for a significant decision, you must provide a clear, prominent notice that tells the consumer what ADMT is being used, what decision it supports, the categories of personal information it processes, and how to exercise their opt-out right. The notice must be delivered before processing begins, not buried in a privacy policy the consumer has never read. The CPPA has specified that the notice must be presented in a way that is easy to find and understand.
-
The second is the right to opt out. Consumers must be offered at least two methods to opt out of ADMT for significant decisions, and those methods must be easy to access and use. When a consumer exercises this right, your organization must offer an alternative decision-making process that uses meaningful human involvement. The regulations are specific about what meaningful human involvement requires: the reviewer must know how to interpret the ADMT output, must have access to the information needed to make an independent judgment, and must have the actual authority to override the ADMT’s result. A rubber-stamp review process does not satisfy this requirement.
-
The third is the right to access. Consumers can request information about how ADMT was used in a significant decision affecting them, what personal information was processed, and what the basis of the decision was. This right creates a documentation obligation that runs parallel to the operational obligations: your organization must be able to produce a coherent explanation of how an automated decision was made, which requires that the system’s logic be documented and that the documentation be accessible to whoever handles access requests.
The Risk Assessment Obligation That Was Due January 1, 2026
The consumer-facing obligations take effect January 1, 2027. The risk assessment obligation took effect January 1, 2026. If your organization has been using ADMT for significant decisions since before January 1, 2026, a risk assessment should have been underway since the beginning of this year.
Under the CPPA regulations, businesses using ADMT for significant decisions must conduct a documented privacy risk assessment covering the nature and purpose of the ADMT, the categories of personal information processed, the significant decision it supports, the benefits and risks of the processing, and the safeguards in place to mitigate those risks. For processing already underway before January 1, 2026, the assessment must be completed by December 31, 2027. For ADMT deployments beginning on or after January 1, 2026, the assessment must be completed before the system is deployed. The first filing with the CPPA, an attestation that required assessments were completed along with a summary of assessment information, is due April 1, 2028.
The practical implication for retailers discovering scope exposure now is significant. If a fraud scoring system or hiring tool that qualifies as ADMT has been running since before January 1, 2026, the risk assessment obligation is already running. Starting a compliant risk assessment process now satisfies both the ongoing obligation and the January 2027 consumer-facing deployment deadline.
The Inventory Problem Most Retailers Have Not Solved
The compliance obligation attaches at deployment, not at a future review date. Businesses already using ADMT for significant decisions before January 1, 2027 must be compliant by that date. Businesses that begin using ADMT for significant decisions on or after January 1 must be compliant immediately upon deployment.
The most common gap in retailer compliance programs is an incomplete inventory. Most organizations that have done a preliminary AI inventory have captured the generative AI pilots and named AI products they deployed intentionally. What they frequently miss are the systems that have been making automated decisions for years without the organization having labeled them as AI: the rules engine in the applicant tracking system that eliminates candidates before a hiring manager sees them, the fraud score threshold that automatically declines an order, the account suspension trigger that flags a customer based on algorithmic criteria. None of these require a large language model to qualify as ADMT. All of them can qualify as significant decision systems.
The starting point for January 1 compliance is not building a notice. It is an inventory of every system that processes consumer personal information and produces an output used to make or substantially influence a decision. That inventory drives everything else: which systems require pre-use notices, which require opt-out mechanisms, which require documented risk assessments, and which require a human review alternative.
For more on the broader state privacy law changes taking effect January 1, 2027, including Delaware’s lowered applicability threshold and Oklahoma’s new privacy requirements, our Q3 coverage provides the full picture: You Were Too Small for Privacy Law. As of January, You Aren’t.
A Note on Applicability: Who CPRA’s ADMT Rules Cover
The CPRA’s ADMT obligations apply to businesses subject to the CCPA that use ADMT for significant decisions affecting California consumers. A business is subject to the CCPA if it meets any of three thresholds: annual gross revenues above $25 million; buying, selling, or sharing the personal information of 100,000 or more consumers or households per year; or deriving 50 percent or more of annual revenues from selling or sharing consumer personal information.
For small and mid-sized retailers, the most commonly triggered threshold is the data processing one. A retailer with an e-commerce site, a loyalty program, and routine digital marketing analytics may cross the 100,000 consumer threshold without meeting the revenue threshold. If your business processes data on California residents, the applicability analysis is worth confirming before assuming the rules do not apply.
A business is subject to the CCPA if it meets any of three thresholds:
Annual gross revenues above $25 million;
It buys, sells, or shares the personal information of 100,000 or more consumers or households per year;
It derives 50% or more of annual revenues from selling or sharing consumer personal information.
Colorado Is Also Moving on January 1
Retailers and insurance agencies with customers in Colorado should note that Colorado’s Automated Decision-Making Technology Act takes effect on January 1, 2027. Colorado’s law, SB 26-189, requires a notice before automated tools are used in a consequential decision and a written explanation within 30 days of an adverse outcome, and it covers insurance underwriting and other consequential consumer decisions. A full breakdown of the Colorado obligations and how they interact with the NAIC AI governance requirements we covered in Q2 will follow in this edition. If your operations span California and Colorado, you are facing two simultaneous January 1 deadlines that address the same underlying technology but with different procedural requirements.
New York’s RAISE Act Also Takes Effect January 1
New York’s Responsible AI Safety and Education Act, known as the RAISE Act, takes effect on the same day, but it regulates a different group of companies. The law applies to developers of frontier AI models, which are systems trained using more than 10^26 computing operations, and its heaviest obligations fall on large frontier developers with more than $500 million in annual revenue. Those developers must publish safety frameworks, file disclosures with a new oversight office inside the New York Department of Financial Services, and report critical safety incidents within 72 hours. Retailers that use AI tools built by those developers are not covered by the RAISE Act themselves, but the law still matters to your planning for two reasons. The first is vendor oversight. The companies supplying your AI tools will soon answer to a state regulator, and their published safety frameworks give you public documentation to review during vendor due diligence. The second is the direction of state law. California and Colorado regulate how businesses use AI in decisions about consumers, while New York regulates the developers who build the underlying models. Retailers that rely on third-party AI should expect customers, insurers, and regulators to ask about both.
What Your Organization Needs to Have in Place Before January 1
Retailers that are using or suspect they may be using ADMT for significant decisions should work through the following before December 31.
The first step is a complete ADMT inventory across your operations. Document every system that processes consumer personal information and produces an output used to make or substantially influence a decision. Include rules engines, scoring systems, applicant tracking configurations, fraud tools, and account management automation alongside any explicitly AI-branded products.
The second step is a scope analysis against the significant decision categories. For each system in the inventory, determine whether the decision it supports falls within financial services, housing, education, healthcare, employment, or access to essential goods and services. Apply the advertising exclusion carefully to verify that marketing and recommendation systems are genuinely outside scope rather than assumed to be.
The third step is a gap assessment against the three consumer rights. For each in-scope system, confirm whether a compliant pre-use notice exists, whether an accessible opt-out mechanism is in place with at least two methods, and whether a human review alternative is defined that meets the meaningful involvement standard.
The fourth step is a risk assessment review. For ADMT systems deployed before January 1, 2026, confirm whether a compliant risk assessment has been initiated. If not, begin one now. The December 31, 2027 completion deadline for existing systems is approaching, and an assessment started today is better positioned than one started next year.
CompassMSP works with retail and franchise businesses on data privacy compliance, AI governance program development, and the multi-state compliance mapping that multi-location operators require. Learn about our retail and franchise compliance services.
YOU MAY NEED TO KNOW
Frequently Asked Questions
What is California’s CPRA ADMT regulation and when does it take effect for retailers?
The California Privacy Protection Agency finalized its automated decision-making technology regulations in September 2025, and they took effect January 1, 2026. The consumer-facing obligations under those regulations, including the requirements to provide pre-use notice, honor opt-out requests, and respond to access requests, apply to businesses using ADMT for significant decisions beginning January 1, 2027. Risk assessment obligations for ADMT used in significant decisions began January 1, 2026. Retailers already using qualifying ADMT should be conducting risk assessments now, with the first formal CPPA submission of assessment information due April 1, 2028.
Does California’s CCPA ADMT rule apply to my small or mid-sized retail business?
The ADMT obligations apply to businesses subject to the CCPA that use ADMT to make significant decisions about California consumers. A business is subject to the CCPA if it has annual gross revenues above $25 million, processes the personal information of 100,000 or more consumers or households annually, or derives 50 percent or more of revenues from selling or sharing consumer personal information. Many mid-sized retailers cross the 100,000 consumer threshold through e-commerce activity, loyalty programs, and digital marketing analytics even without meeting the revenue threshold.
My business only uses AI for product recommendations and advertising. Am I covered under CPRA’s ADMT requirements?
Probably not for those specific uses. The CPPA regulations explicitly exclude advertising from the significant decision definition, and product recommendation algorithms that surface items to consumers are generally not making a significant decision as the regulation defines it. However, the exclusion does not cover all retail AI uses. Credit and BNPL decisions, fraud detection that restricts account access, employment screening, and pricing or access decisions for essential goods are covered categories. Many retailers discover in an inventory exercise that systems outside their marketing stack qualify.
What counts as a “significant decision” under California’s CCPA ADMT regulation in a retail context?
Significant decisions under the California ADMT regulation are those affecting a consumer’s access to or the terms of financial or lending services, housing, education, healthcare, employment, or access to essential goods and services. In a retail context this covers store credit and BNPL approval or denial, fraud scoring that restricts account access, employment screening through applicant tracking systems, and potentially pricing or access decisions for essential goods. The regulations explicitly exclude advertising.
What does the CPRA ADMT pre-use notice requirement actually require retailers to provide to consumers?
Before using ADMT for a significant decision about a California consumer, the retailer must provide a clear, prominent notice that explains the ADMT being used, the decision it supports, the categories of personal information it processes, and how the consumer can exercise their opt-out right. The notice must be provided before processing begins, not buried in a privacy policy or general terms of service. The CPPA has indicated that the notice must be easy to find and presented in plain language.
What is the CPRA ADMT opt-out requirement and what happens operationally when a California consumer exercises it?
Consumers subject to ADMT for significant decisions must be offered at least two methods to opt out, and those methods must be easy to access and use. When a consumer exercises the opt-out right, the retailer must offer an alternative decision-making process that uses meaningful human involvement. The regulations define meaningful human involvement as review by a person who knows how to interpret the ADMT’s output, has access to the information needed to make an independent judgment, and has the authority to override the ADMT’s decision. A process where a human confirms an AI output without independent analysis does not satisfy this standard.
What is the CCPA ADMT risk assessment requirement for retailers and when is the CPPA submission due?
Businesses using ADMT for significant decisions must conduct a documented privacy risk assessment covering the purpose of the ADMT, the personal information processed, the significant decision it supports, the benefits and risks of the processing, and the safeguards in place. For ADMT deployed before January 1, 2026, the assessment must be completed by December 31, 2027. For ADMT deployed on or after January 1, 2026, the assessment must be completed before the system is deployed. The first formal submission to the CPPA, an attestation and summary of assessment information for 2026 and 2027, is due April 1, 2028.
How do I know if my retail fraud detection or account restriction system qualifies as ADMT under California’s CCPA?
A fraud detection or identity verification system qualifies as ADMT if it processes consumer personal information and produces an output that is used to make or substantially make a decision without meaningful human review. The sophistication of the technology is not the determining factor. A rules engine that automatically declines an order or restricts an account when a threshold is crossed without a human reviewing the specific decision qualifies. Retailers whose fraud systems result in automatic account restrictions, refund denials, or order cancellations should evaluate whether a human with override authority and independent analytical access is actually involved before the restriction takes effect.
Does California’s CCPA ADMT rule cover the applicant tracking and employee hiring systems used at retail locations?
Yes. Applicant tracking systems with automated knockout criteria, resume ranking, or candidate scoring that filter applicants before a hiring manager reviews them are ADMT for an employment decision and are covered by the significant decision standard. This applies to systems used for all retail employment, including hourly store positions. Many retailers have decades-old ATS configurations in this category that were never labeled as AI and have never been reviewed for ADMT compliance.
What are the CPPA penalties for CCPA ADMT non-compliance and when does enforcement begin for retailers?
The CPPA can impose civil penalties of up to $2,500 per violation for unintentional violations and up to $7,500 per violation for intentional violations. Each instance of a consumer’s rights being violated is a separate violation, meaning that a retailer using non-compliant ADMT for significant decisions across a large California consumer base faces compounding exposure. The CPPA also has authority to seek injunctive relief and conduct compliance investigations on its own initiative. The agency has signaled that its 2027 enforcement priorities will include ADMT, and businesses discovered to be using qualifying systems without compliant notices and opt-out mechanisms after January 1 will have limited recourse based on a “we did not know” defense.
Richard Mendoza
Richard is the Director of vCISO services with CompassMSP. He has over twenty-five years of experience as an Information Security professional with hands-on experience in engineering process and information security, and IT audit disciplines. With a wide-ranging knowledge as a Systems Engineer, Information Security Officer, and Senior Auditor, Richard has expertise in managing internal and external audits focused on reducing overall risk exposure and infrastructure redundancy for organizations.