Why do annual recertification backlogs keep growing even after hiring?

Because the bottleneck is usually not staff review. Most elapsed time sits in the states where a case is waiting on documents from the household, often three to six weeks per case. Hiring adds throughput to a stage that is not constrained, so the backlog barely moves. Worse, waiting cases generate inbound calls that consume the same specialists who would otherwise close the ready ones.

What is the fastest way to speed up document collection for recerts?

Replace portal logins with one-tap secure links sent by text. No account, no password, camera-first, and plain-language requests that name the exact document — “a paystub from the last 60 days” rather than “income documentation.” Confirming receipt visibly to both the household and the specialist also removes a large share of status calls.

How should AI income verification work so specialists actually trust it?

Every figure must be traceable. Show the extracted number next to the source document image, state the pay frequency and annualization used, show the EIV comparison with the variance in dollars per month, and characterize what the variance looks like. A specialist can confirm that in seconds; an unsourced number just gets re-derived by hand.

How do you prevent EIV discrepancies from becoming audit findings?

Run the EIV-to-documents comparison on every case at the time of the recertification, not as a separate quarterly report, and make the result part of the packet. A small monthly variance caught during the recert is a two-minute conversation and a file note; found during a monitoring review, it becomes a finding, a repayment agreement, and a corrective action plan.

What does a finished recertification packet look like in a modern workflow?

One prepared case a specialist opens once: all required documents collected and legible, income extracted and annualized with links to source, EIV pulled and reconciled with variances flagged, any household responses captured, and a 50058-ready packet that exports clean to the system of record. The specialist reviews and approves — software never finalizes eligibility, income, or rent.

All articles

OperationsJuly 13, 2026 · 9 min read

How to Clear an Annual Recertification Backlog Without Hiring

Backlogs are almost never a staffing shortfall. They are a queueing problem with one dominant bottleneck, and it is usually not the part your specialists are doing.

Corina Alparaque

Public Housing Center — writing from the work we do with housing authorities every day.

Key takeaways

Recert backlogs are queueing problems. Measure where cases wait, not how many cases exist — the wait is almost always on documents, not on staff review.
The dominant delay at most PHAs is the gap between the notice going out and the household's paperwork arriving, often three to six weeks of pure waiting.
Every case in that waiting state generates inbound calls, which consumes the same specialists who would otherwise be closing cases.
Portals reduce document friction less than PHAs expect: passwords, app installs, and account recovery are their own drop-off funnel. One-tap secure links to a phone camera work far better.
Income verification and EIV reconciliation should produce a variance flagged before it becomes an audit finding, with every figure linked to the source document.
Aim for one prepared, 50058-ready packet per case that a specialist reviews once — not a case a specialist touches a dozen times.

Nearly every housing authority we talk to has a recertification backlog, and nearly every one describes it as a staffing problem. They are short two specialists, or the one who knew the process retired, or the caseload per specialist is well past what anybody thinks is reasonable.

All of that is usually true. It is also usually not the reason the backlog exists, which is why hiring into it produces less relief than expected.

Measure the wait, not the pile

The instinct with a backlog is to count cases. The useful measurement is different: for each open recert, where is it sitting right now, and how long has it been there?

Sort your open annuals into these states:

Notice sent, nothing received back
Partial documents received, waiting on the rest
Complete documents, waiting on staff review
In review, waiting on a third-party verification
Waiting on the household to respond to a discrepancy
Ready to sign

Then look at where the time is. At most authorities, the two waiting-on-documents states hold the overwhelming majority of the elapsed days — frequently three to six weeks per case of nothing happening at all. "Complete documents, waiting on staff review" is usually a much smaller pool that moves reasonably fast once a specialist opens it.

That distribution is the whole story. Your specialists are not slow. Your cases are sitting in a queue where the constraint is a household finding a paystub, remembering where the envelope went, and getting it to you.

Adding a specialist adds throughput to the stage that is not the bottleneck. This is why the backlog does not move proportionally when you hire.

The second cost of the waiting state

There is a compounding effect that makes the document-wait more expensive than it looks on a chart.

A case sitting in "notice sent, nothing received" is not passive. It generates a call — often several. The household calls to ask what is needed. They call again to ask if it arrived. Their adult daughter calls. If the notice was confusing or in the wrong language, they call before doing anything at all.

Those calls are answered by the same specialists who would otherwise be reviewing the cases that are ready. So the waiting state does not just delay its own case; it steals capacity from the stage that works. That is the feedback loop that turns a seasonal bump into a permanent backlog.

Why the portal did not fix it

Most authorities have tried to solve document collection with a resident portal, and most have been disappointed by the uptake.

The reason is not that residents are unwilling. It is that a portal asks for an account before it asks for a paystub. Every step in that chain loses people: find the site, create a login, satisfy the password rules, verify the email, remember the password next time, recover the account when they do not, find the right upload area, figure out what "supporting documentation" means, get a file off their phone and into a form field.

The household in front of you has a phone with a camera and a text message inbox. That is the whole reliable technology stack. The collection method that works is the one that starts with a text and ends with a photo:

A link that opens straight to the request, no login.
Plain language naming the exact document: "a paystub from the last 60 days," not "income documentation."
Camera-first, one tap, works on an old phone with a cracked screen.
The document lands attached to the right case, and both the household and the specialist can see that it arrived.

That last part — visible confirmation of receipt — removes a whole category of calls by itself.

Verification: the part that should be boring

Once documents are in, income verification is mechanical work that consumes real specialist time: read the paystubs, work out the pay frequency, annualize, compare against what EIV reports, decide whether a difference is a discrepancy or a timing artifact.

This is exactly the kind of work AI should do and exactly the kind of work where an unverifiable AI answer is worse than no AI at all. A tool that hands your specialist "annual income: $26,208" and nothing else has not saved them anything, because a careful specialist will re-derive it. The saving comes from the audit trail:

The extracted figure, next to the document image it came from, with the relevant line visible.
The pay frequency and the annualization it used, stated explicitly.
The EIV comparison, with the variance called out in dollars per month.
A note on what kind of variance it looks like — a raise mid-quarter, a new employer, an overtime spike, a genuine reporting gap.

A specialist can confirm that in seconds. That is the difference between AI that saves twenty minutes per case and AI that adds a verification step.

Catch the variance before the monitor does

The EIV discrepancy that becomes a finding is almost never a surprise in hindsight. It was visible in the file the whole time; nobody had the minutes to line the two numbers up on every case.

A $48-a-month difference between the paystubs and what EIV reports is small enough to be a mid-year raise and large enough to matter on a 50058. Flagged during the recert, it is a two-minute conversation and a note in the file. Found during a monitoring review eighteen months later, it is a finding, a repayment agreement, and a corrective action plan.

Run the comparison on every case, every time, and make the output part of the packet rather than a separate report someone runs quarterly.

One packet, one review

The target state is worth stating concretely, because it is different from "faster."

Today, a typical recert is touched many times: open the file, send the notice, take the call, log the partial documents, send a reminder, take another call, open it again when the rest arrives, key the income, pull EIV, reconcile, chase the discrepancy, prepare the 50058 fields, review, sign.

The target is that a specialist opens the case once, and what is in front of them is:

Every required document, collected and legible.
Income extracted, annualized, and linked to its source.
EIV pulled and reconciled, with any variance flagged and explained.
Any household response to a discrepancy already captured.
A 50058-ready packet, complete, exported clean to whatever your system of record is.

They read it, they check the flagged items, they approve. One review. And critically: they approve. Nothing about eligibility, income determination, or rent is finalized by software. The AI assembles; the specialist decides. That is both the compliance posture and the reason staff will actually use it.

A 30-60-90 that does not require a procurement first

Some of this you can start on Monday:

Days 1–30. Sort your open annuals by waiting state and measure elapsed days in each. Rewrite your recert notice in plain language, naming exact documents, in your top two non-English languages. Add a text-message reminder at day 7 if nothing has arrived.

Days 31–60. Stand up one-tap document submission for the top three document types, even if it is imperfect. Start logging which cases generate calls. Begin running the EIV comparison on every case at the point of review rather than as a periodic report.

Days 61–90. Take your worst cohort — the oldest open annuals — and run them end to end through a completion workflow with real files, real staff, and real approval. Measure days-to-signed-50058 against your baseline. That number is your business case, and it is also how you know whether the tool works before you scale it.

If you want to run that third step on your own caseload rather than a demo dataset, we do exactly that — bulk upload, real cases, your staff approving every determination. It is scoped to fall under small purchase rules, and the outcome is a number you can take to your board either way.

Frequently asked questions

Why do annual recertification backlogs keep growing even after hiring?: Because the bottleneck is usually not staff review. Most elapsed time sits in the states where a case is waiting on documents from the household, often three to six weeks per case. Hiring adds throughput to a stage that is not constrained, so the backlog barely moves. Worse, waiting cases generate inbound calls that consume the same specialists who would otherwise close the ready ones.
What is the fastest way to speed up document collection for recerts?: Replace portal logins with one-tap secure links sent by text. No account, no password, camera-first, and plain-language requests that name the exact document — “a paystub from the last 60 days” rather than “income documentation.” Confirming receipt visibly to both the household and the specialist also removes a large share of status calls.
How should AI income verification work so specialists actually trust it?: Every figure must be traceable. Show the extracted number next to the source document image, state the pay frequency and annualization used, show the EIV comparison with the variance in dollars per month, and characterize what the variance looks like. A specialist can confirm that in seconds; an unsourced number just gets re-derived by hand.
How do you prevent EIV discrepancies from becoming audit findings?: Run the EIV-to-documents comparison on every case at the time of the recertification, not as a separate quarterly report, and make the result part of the packet. A small monthly variance caught during the recert is a two-minute conversation and a file note; found during a monitoring review, it becomes a finding, a repayment agreement, and a corrective action plan.
What does a finished recertification packet look like in a modern workflow?: One prepared case a specialist opens once: all required documents collected and legible, income extracted and annualized with links to source, EIV pulled and reconciled with variances flagged, any household responses captured, and a 50058-ready packet that exports clean to the system of record. The specialist reviews and approves — software never finalizes eligibility, income, or rent.

Keep reading

Compliance