AI should never determine eligibility
A hard product rule, not a slogan. Agents may organize files and surface exceptions. Qualified humans keep eligibility, approval, denial, and every notice that carries legal weight.
Affordable housing already has a compliance officer, a file, and a set of program rules. Adding AI does not change who is allowed to decide whether a household qualifies. If a vendor is vague about that line, they are asking you to take a risk they will not put their name on.
We treat it as a technical boundary, not a preference. Agents do not determine eligibility. They do not approve or deny. They do not send adverse, legal, or eligibility notices on their own. If a workflow needs one of those actions, it stops and waits for a person.
What agents may do
- Collect and classify applicant or recertification documents against a required list.
- Map files to household members and flag what is missing.
- Track third-party verifications and age unanswered requests.
- Draft missing-document reminders and appointment notices for staff to review and send.
- Surface exceptions: complete enough for review, stale, or waiting on an agency.
What qualified humans keep
- Eligibility and income determinations.
- Applicant approval, denial, and prioritization.
- Legal and compliance-sensitive communications.
- Final certification and reporting sign-off.
That split is the only version of 'AI for lease-up' a compliance manager should be willing to forward internally. The machine makes the file ready. The person makes the call.
If the agent can say yes or no to a household, it is in the wrong layer of the system.
Why the line has to be enforced, not promised
A policy in a slide deck is not a control. Least-privilege access, human approval gates, and a record of what was read, drafted, and sent are how the rule survives a busy Friday. The agent can prepare a summary for review. It cannot skip the reviewer because the file looks complete.
This is also why we do not sell a generic chatbot for occupancy. A chat window will answer the question you typed. It will not refuse the question it should not own. An operating layer can. If you want a system your compliance lead can live with, start there, not with a smarter prompt.
Common questions
- Can AI determine LIHTC or HUD eligibility?
- It should not. Eligibility, approval, denial, and certification sign-off belong to qualified staff. Agents organize the file and stop before the decision.
- What is safe to automate in affordable housing compliance?
- Document collection and classification, completeness queues, verification tracking, exception reporting, and draft communications for human review.
- How do you enforce that AI will not deny an applicant?
- Least-privilege access, human approval gates on eligibility and legal notices, and a record of what was read, drafted, and sent. A policy slide is not a control.