Guide · Published November 14, 2026 · Reviewed November 14, 2026 · Ingenious AI team · Not legal advice
How to Inventory Your Automated Decision Tools: Finding Every System That Decides
The short answer: Every ADMT duty starts with a list: the tools that materially influence consequential decisions about people. Most businesses' first list is incomplete — it names the flagship screening platform and misses the scheduling optimiser, the lead-scoring tool that quietly sets terms, and the spreadsheet macro a team lead built. A complete inventory takes five passes: procurement records, decision workflows, data flows, shadow tools, and a register that records, per tool, the decisions it touches, the states involved, its role's weight, and where its notices and records live.
Check your duties and dates — free Build your first notice — free
What is an ADMT inventory? It is the register of automated tools that replace, substantially replace, or materially influence decisions about people — the population the notice duties apply to. The scope tests come from the kit's data: California asks whether the technology replaces or substantially replaces human decision-making; Colorado asks whether the tool materially influences a consequential decision. Both tests are applied per tool in How to Tell If a Tool Is ADMT; this guide is about finding the tools to test.
Pass one: procurement — what you bought that decides
Start with money: contracts, subscriptions, and invoices for software in HR, lending, housing, education, and customer operations. Vendor products that score, rank, screen, assess, verify, or recommend are candidates. Procurement finds the sanctioned tools reliably — and only the sanctioned tools, which is why it is pass one of five, not the method.
For each purchased candidate, capture the contract facts the duties will later need: the tool's exact (trade) name for the notices, its vendor contact for documentation duties, and its renewal date — the moment your leverage to demand documentation is highest. Colorado's developer-documentation duty (intended uses, training-data categories, known limitations) is described in Colorado Developer Duties; the inventory is where you note which vendors owe you that material.
Pass two: workflows — follow each decision backwards
List your consequential decisions first — hiring, tenant screening, lending, admissions, service eligibility, scheduling — then walk each one backwards from the outcome and note every system that touched it. This direction finds what procurement cannot: the scoring step inside a larger platform, the 'recommendation' panel in the CRM, the rules engine in the legacy system nobody calls AI.
For each tool found, record its weight honestly: does it decide, substantially contribute, or merely inform a person who genuinely decides? The weight entry is a first draft of the scope test, and it will be revisited — but writing it during the workflow pass, with the operator present, captures facts that are never written down anywhere else.
Worked example: Great Lakes Logistics Group
Example — fictional business, real duty structure. Great Lakes is a fictional logistics employer with hiring, scheduling, and contractor-assignment decisions across Colorado and California. Its first inventory, run over three weeks, found:
- Pass one (procurement): 4 candidate tools — a résumé screener, a background-check platform with a scoring step, a scheduling suite, and a contractor-assignment app.
- Pass two (workflows): 3 more — a recommendation panel inside the CRM used for customer credit terms, a rules engine in the legacy dispatch system, and the screener's separate interview-scoring module, initially counted as part of the screener.
- Pass three (data flows): 2 more — an embedded identity-verification scoring API and a third-party data service returning applicant risk categories.
- Pass four (shadow tools): 2 more — a depot manager's spreadsheet ranking model and a free AI tool used to summarise applications; the spreadsheet is adopted and governed, the free tool's use is stopped.
- Register result: 11 tools registered with decisions, states, weights, and data categories; 6 confirmed in scope for notices, 5 documented as out of scope with reasoning — the documented negatives matter as much as the positives.
Pass three: data flows — follow the person's data forwards
Now follow personal data forwards: where does an applicant's or customer's data go after collection? Every destination that returns a score, a flag, a category, or a recommendation is a candidate tool — including API services embedded in other products, whose names appear in data-processing agreements and privacy notices rather than in any user's vocabulary.
This pass has a second yield: it drafts the data element of California's risk assessment and the data-categories field of Connecticut's written notice at the same time. An inventory that records, per tool, the categories of data consumed and their sources is doing three duties' groundwork in one document.
Pass four: shadow tools — the ones nobody bought
Every organisation of any size has them: the spreadsheet scoring model, the free-tier AI tool a manager pastes applications into, the script an analyst wrote. Shadow tools carry the same duties and none of the vendor documentation. Find them by asking team leads a direct question — 'what do you use to rank, score, or shortlist?' — in a tone that makes honesty safe. An inventory with an amnesty finds tools; an inventory with consequences finds silence.
For each shadow tool, the options are adopt-and-govern (bring it into the register with notices and records), replace with a sanctioned tool, or stop the use. What is not an option is leaving it off the register: the register's credibility is the inventory's whole product. Employment shadow tools are the most common find — see the sector walkthrough in ADMT Notices for Hiring Decisions.
Pass five: the register — one row per tool, kept alive
The register is a living document with one row per tool: name and trade name, vendor or internal owner, decisions touched, states involved, weight assessment, data categories and sources, notice status, record location, and review date. The kit's setup mirrors this shape — a business profile, its tools, its decision types and states — because the notices are generated from exactly these fields.
Give the register a cadence: review on procurement changes, on vendor updates (a material update can change a tool's weight), and at least quarterly. A register reviewed quarterly stays a register; a register written once becomes archaeology. Run each registered tool through the free checker per state and decision type, and file the results with the row — the inventory then ends not with a list, but with a duty set per tool.
- Pass one — procurement: list every purchased tool that scores, ranks, screens, assesses, or recommends, with trade names and renewal dates
- Pass two — workflows: walk each consequential decision backwards and record every system that touched it, with its weight
- Pass three — data flows: follow personal data forwards to every service that returns a score, flag, or category
- Pass four — shadow tools: ask team leads directly, under amnesty, what they use to rank or shortlist
- Pass five — the register: one row per tool with decisions, states, weight, data categories, notice status, and a quarterly review date
Sources, as carried in the kit's own data: the duty summaries in app/rules.py and the checker data in app/checker_data.py, verified against primary or authoritative sources on October 28, 2026. Run the free duties checker for your jurisdiction, decision type, and the tool's role, and compare the kit's plans for the record, builder, and review queue. More from Ingenious AI is on the hub.
Related guides
- How to Tell If a Tool Is ADMT: The Role Test in Colorado, California, and Connecticut
- Colorado Developer Duties: The Technical Documentation You Owe Your Deployers
- ADMT Notices for Hiring and Employment Decisions: A State-by-State Employer Guide
- Colorado AI Act Automated Decision Notices: What to Send
See all ADMT guides for the full set.
Frequently asked questions
Why does the inventory come before everything else?
Every duty — notice, explanation, opt-out, assessment — attaches to a tool used for a covered decision. Tools missing from the inventory get no notices and no records, which is the most common root cause of non-compliance.
What counts as an automated decision tool for the inventory?
Anything that replaces, substantially replaces, or materially influences a decision about a person — including scoring steps inside larger platforms, embedded APIs, and internal or shadow tools, not just products marketed as AI.
Should out-of-scope tools be in the register?
Yes, with the reasoning recorded. A documented negative — this tool is a minor input to a genuinely human decision, and why — is evidence the scope test was applied, and it gets revisited when the tool or its use changes.
How often should the inventory be reviewed?
On procurement changes and material vendor updates, and at least quarterly. A tool's weight can change when its use changes, even if the software does not.
What fields does each register row need?
Tool and trade name, owner or vendor, decisions touched, states involved, weight assessment, data categories and sources, notice status, where its records live, and a review date.