The 2026 immigration backlogs: what the numbers mean for firm capacity, and what to do about them

Every immigration practice in the United States is now operating against queues that did not exist at this scale five years ago. That is not a cyclical dip that reverses next quarter. It is a structural change in how long a matter stays open, and therefore in how much work a matter generates before it closes.

The response most firms reach for first is hiring, and the response that usually fits better is operational. This piece sets out what the current figures actually say, what they do to firm capacity, which of the resulting work can be delegated without touching legal judgement, and what has to be in place before any of it is.

Two backlogs, two different numbers

The single most common error in commentary on this subject — including in material written for law firms — is treating “the immigration backlog” as one number. There are two systems, they are administered by different departments of the federal government, and they do different things.

United States Citizenship and Immigration Services (USCIS), part of the Department of Homeland Security, adjudicates benefits: petitions, applications, and requests for immigration status or work authorisation. It held 11.3 million pending cases at the end of FY2026 Q1.

The Executive Office for Immigration Review (EOIR), part of the Department of Justice, runs the immigration courts. It is where removal proceedings happen. It held 3,195,137 pending cases as of June 2026, up 12% year over year, of which 2,310,698 were asylum applications awaiting a hearing or a decision.

A firm can be heavily exposed to one and barely touched by the other. A family-based practice filing adjustment applications lives almost entirely in the first queue. A removal defense practice lives in the second. Most firms sit somewhere between, and the mix determines which set of pressures actually applies to them.

How the backlog is actually measured

“Pending” and “backlogged” are not synonyms, and the difference matters enormously if you are trying to plan around them.

The gap between them is substantial: 11.3 million total pending against a net backlog of 6.3 million. For firm planning, the net figure is the meaningful one, because it isolates cases that are genuinely stuck rather than merely in progress. A case sitting inside normal processing time is not a problem; it is the system working as designed.

The distinction also matters when you are setting client expectations. Telling a client their case is part of an 11.3 million-case backlog is technically defensible and practically misleading — it conflates their recently-filed, on-schedule matter with cases that have been stuck for years.

Why throughput matters more than the queue

A large queue is not, by itself, evidence of a worsening problem. A system can hold a large queue indefinitely if it clears cases as fast as it receives them. What turns a queue into a crisis is the completion rate.

USCIS completions fell 41% year over year in FY2026 Q1. That is the number to watch, because it changes the direction of travel. A queue that grows while throughput falls does not stabilise on its own; it compounds. Every quarter of reduced completions adds to a base that the following quarter must also work through.

For a firm, the practical consequence is not that cases become harder. It is that cases stay open longer. And an open case consumes attention for the entire time it is open, whether or not anything is happening on it.

The agency side: what slows down, and where

Processing delays are not evenly distributed. They cluster by form type, by service centre, and by whether a case requires an interview or biometrics appointment. Understanding where your own caseload is concentrated tells you far more than a headline figure does.

The categories most immigration practices feel first:

• Family-based petitions. Long-running by nature, with the added complication that priority dates and visa availability sit outside the agency’s control entirely. A case can be administratively complete and still wait years.
• Adjustment of status applications. Frequently interview-dependent, which introduces scheduling as an additional bottleneck on top of adjudication.
• Employment authorisation. Time-critical in a way most filings are not, because a delay has an immediate consequence for the applicant’s ability to work. Renewal timing becomes an operational discipline rather than an administrative afterthought.
• Naturalisation. Interview-dependent and oath-ceremony-dependent, so two scheduling bottlenecks sit downstream of adjudication.
• Humanitarian filings. Evidence-heavy at the preparation stage and slow at the adjudication stage, which is the worst combination for a firm’s working capital.

The operational point is that each of these generates a different follow-up rhythm. A case waiting on visa availability needs almost no attention for long stretches and then needs prompt action. An interview-dependent case needs monitoring for a notice that can arrive with limited warning. A work-authorisation renewal needs a countdown started well before it becomes urgent. Treating them all with the same review cadence either wastes effort or misses deadlines.

The court side: how the EOIR queue works

Immigration court operates on a fundamentally different logic to agency adjudication, and firms that move between the two sometimes carry assumptions across that do not hold.

The most important difference: court dates are fixed. An agency processing time is an estimate that slips. A hearing date set by an immigration judge is a date. Filing deadlines set in relation to it are similarly hard. Nothing about a staffing shortage at your firm changes either.

With 3,195,137 cases pending and waits stretching to years, the interval between the master calendar hearing and the individual hearing can be very long. That gap is deceptive. It looks like time in hand, and it functions as a trap: evidence gathered early goes stale, country conditions material dates, medical and psychological evaluations age, contact details change, and witnesses become harder to locate.

A file that was ready two years before the hearing is frequently not ready at the hearing. The work of keeping it ready — periodic refresh, re-verification, re-indexing — is invisible in any case-count statistic and is one of the largest hidden costs a long court queue imposes on a firm.

Representation itself remains far from universal: the overall representation rate across completed cases was 55% (Vera Institute, FY2025), rising to 77% for completed asylum cases. Firms operating in this space are not competing for scarce clients. They are rationing scarce capacity.

What a long queue does to firm economics

The financial mechanics of a lengthening queue are worth stating plainly, because they are counter-intuitive.

In a flat-fee practice — which describes most immigration work — the fee is typically set at engagement and collected across the matter. The work required to deliver that matter, however, is not fixed. It scales with how long the matter stays open. When lifecycle lengthens by a year, the fee does not change but the servicing cost does.

Three effects follow, and they compound:

1. Effective realisation falls. The same fee now covers more hours of servicing. Nothing about the engagement looks different; the margin quietly erodes.
2. Open-matter count ratchets upward. If matters open at a steady rate and close more slowly, the standing caseload grows without a single new client being signed. Most firms notice this as a vague sense that everyone is busier without knowing why.
3. Capacity gets consumed by matters that are not progressing. Attention spent on a stalled case is attention unavailable for a new one, which caps intake without any explicit decision to cap intake.

The last of those is the one that most often goes unexamined. Firms conclude they cannot take on more work and treat that as a fact about demand, when it is usually a fact about how their existing caseload is being serviced.

The administrative load, itemised

If the argument is that backlogs generate administrative rather than legal work, that claim should be specific. Here is what a long queue actually produces, matter by matter:

Every row in that table is real work with a real cost. Almost none of it requires a law licence to perform, and all of it requires a licensed attorney to supervise. That combination is precisely what makes it delegable — and precisely why the delegation has to be structured rather than casual.

Doing the capacity arithmetic

Before comparing solutions, it is worth sizing the problem. The arithmetic is simple enough to do on paper.

Take your open matter count. Estimate the non-legal hours each consumes in a typical month — status monitoring, document chasing, packet work, client contact. Multiply. That is your monthly administrative demand.

Then take the people currently absorbing that work, in full-time equivalents, multiply by roughly 160 working hours a month, and multiply again by the share of their time genuinely spent on administrative rather than substantive work. That is your capacity.

The difference is the gap. If it is positive, that work is not disappearing — it is being absorbed by attorneys as non-billable time, or it is turning into slipped follow-ups and late refreshes.

The case capacity calculator runs this arithmetic with your own figures and exposes the per-matter assumption as an input rather than hiding it in a constant.

The four responses, and what each costs

There are realistically four things a firm can do about an administrative capacity gap. Each is legitimate; each has a cost that is worth naming honestly.

The fourth deserves one specific caution. Support that works outside your case management system — receiving documents, working on them elsewhere, returning them — creates a reconciliation burden that can consume the savings entirely. Small-firm operations data consistently identifies information scattered across disconnected tools as the single largest source of friction. Adding another disconnected location is not a fix; it is the problem with extra steps.

Where the delegation line sits

Assuming delegation is the route, the boundary needs to be explicit before anyone touches a file. Stated plainly:

Delegable: preparing draft petitions and applications from firm-supplied information; collecting, labelling and indexing evidence; formatting declarations; compiling country-conditions material; preparing draft response language for attorney revision; assembling filing and hearing packets; maintaining status records; tracking deadlines and escalating them; sending administrative updates.

Never delegable: giving legal advice; deciding or recommending strategy or the relief sought; assessing merits or likelihood of success; signing, certifying or submitting any filing; appearing before or communicating as a representative with any court or agency; quoting fees or accepting engagements; exercising independent professional judgement.

The reliable test is not whether a task is difficult. It is whether performing it correctly requires legal judgement. Indexing two hundred exhibits is laborious and requires none. Deciding which twelve of them answer the question actually being asked requires it entirely.

Our full treatment of this is in RFE response support, where the line is hardest to hold and therefore worth stating most precisely.

The compliance work comes first

Firms that delegate successfully settle the professional-responsibility questions before the operational ones. Firms that struggle usually did it in the opposite order and had to retrofit.

The obligations are not exotic. ABA Formal Opinion 08-451 addresses client disclosure and consent where non-lawyers outside the firm receive information protected by Model Rule 1.6. Model Rules 5.1 and 5.3 impose supervision duties over that work. Model Rule 1.1 keeps competence with you regardless of who performs the underlying task. Several states go further than the ABA position — Florida Ethics Opinion 07-2 requires disclosure for overseas outsourcing outright, and California, New York and North Carolina have issued their own opinions.

Practically, four things should exist before the first matter is assigned:

1. Client consent, in the form your jurisdiction requires, obtained before any access is granted rather than after.
2. A written supervision protocol naming the supervising attorney and setting out permitted and prohibited tasks explicitly.
3. A data processing agreement covering sub-processors, breach notification and what happens to data on exit.
4. Scoped system access under least privilege, limited to assigned matters and revoked on offboarding.

Assembling these after work has started is how firms end up with a disclosure gap on matters already in flight — a problem with no clean retrospective fix. The full set of questions is covered in our compliance FAQ.

Triage: matching review cadence to matter type

One of the cheapest improvements available to a firm carrying a long queue is to stop reviewing every matter on the same schedule. Uniform cadence is expensive in both directions: it wastes effort on matters where nothing can change, and it under-serves matters where something can.

A workable triage model sorts open matters into three bands.

The band that causes the most trouble is dormant, because it is the one where doing nothing feels safe. It is not. A dormant matter is precisely where evidence quietly expires, contact details go stale, and a long-scheduled hearing arrives with a file that was ready two years ago and is not ready now. The dated wake-up — a scheduled task set at the point the matter becomes dormant — is what stops dormancy turning into neglect.

Triage also changes the delegation question. Active-band work needs judgement close at hand. Monitored and dormant work is almost entirely procedural, and it is usually the larger share of a long-queue caseload. Firms that delegate the monitored and dormant bands first tend to get the cleanest result, because the escalation criteria are easy to write down and the risk of an unsupervised judgement call is lowest.

What to measure once you are managing it

Most immigration firms measure revenue, matters opened and matters closed. Those are necessary and insufficient for managing a backlog, because none of them reveals whether the standing caseload is being serviced or merely held.

Five measures make the situation legible:

1. Open matters per fee earner. The blunt capacity ratio. Watch the trend rather than the absolute number — a ratio drifting upward with flat headcount is the earliest signal that lifecycle lengthening is being absorbed silently.
2. Median matter age. Not average, which a handful of very old matters will distort. Median age rising while intake is flat tells you closure has slowed.
3. Matters past posted processing time. Your own equivalent of the net backlog. These are the matters where an enquiry or follow-up action may actually be available, and they should be identifiable on demand rather than discovered when a client calls.
4. Deadline near-misses. Not missed deadlines — those are catastrophic and rare. Count the ones met inside the last day or two. Near-misses are the leading indicator; missed deadlines are the lagging one, and by then it is a claim rather than a metric.
5. Non-billable administrative hours by role. The number that reveals whether attorneys are absorbing paralegal work. Most firms have never measured it and are startled by the answer.

The last one is worth a short tracking exercise even if you do nothing else. A week of honest time capture, categorised only as substantive versus administrative, usually settles the question of whether a capacity problem exists faster than any amount of discussion.

Talking to clients about delay

A queue this long changes what clients need to hear, and firms that handle the communication well spend markedly less time on inbound “any update?” contact. The pattern is consistent enough to be worth stating as principles.

Set the expectation at engagement, not at the first complaint. A client told at the outset that their matter type currently runs long, and why, reacts very differently from one who discovers it eighteen months in. The second conversation is always harder and always longer.

Distinguish normal from abnormal explicitly. Clients experience all waiting as the same thing. Being told plainly that a matter is inside normal processing time and that no action is available — as opposed to being past it, where an enquiry may be possible — converts a source of anxiety into a status.

Push scheduled updates rather than waiting for pull. A brief, scheduled status note on a defined cadence costs far less than the unscheduled contact it prevents, and it is administrative work that does not require an attorney to perform.

Do not quote the headline backlog number. Telling a client their case is one of several million is accurate and unhelpful. It conflates their circumstances with the whole system, and it invites the conclusion that nothing can be done.

Be careful who answers what. Administrative updates can come from support staff. Anything touching eligibility, merits, strategy or consequences must come from an attorney. That boundary needs to be explicit if anyone other than an attorney is contacting clients at all — it is the single most common place where a delegation arrangement drifts into territory it should not occupy.

Where technology helps, and where it stops

Immigration case management platforms have improved considerably, and any firm not using one is leaving obvious gains on the table. Modern systems auto-populate large numbers of USCIS, DOL and DOS forms from a single intake record, maintain form libraries, and track status.

It is worth being precise about what that does and does not solve, since this is where a good deal of vendor messaging is misleading.

What software genuinely solves: repeatable transcription from a structured record into a structured form; keeping form editions current; storing and retrieving documents; recording status; generating reminders from dates it already holds.

What software does not solve: deciding what evidence answers a question; noticing that a document received is not the document requested; chasing a client who has not responded; judging whether an explanation is adequate; reading an unstructured notice and working out what it actually requires; assembling material that only exists in someone’s email.

The second list is where the administrative load of a long queue actually sits. This is why “we have AI that fills forms” is a weak proposition to a firm that already licenses a platform doing exactly that. The binding constraint is not form population. It is the judgement-adjacent, unstructured work surrounding it — which needs a person, working under supervision, inside the system where the case already lives.

Two worked examples

Abstract capacity arithmetic is easier to trust with numbers attached. Both examples below are illustrative constructions, not descriptions of actual firms.

A three-attorney family and naturalisation practice

Assume 180 open matters, weighted toward family petitions and naturalisation, with one full-time paralegal spending roughly 60% of their time on administrative work.

At 1.2 administrative hours per matter per month, monthly demand is about 216 hours. Capacity is one FTE at roughly 160 hours, times 60%, giving 96 hours. The gap is around 120 hours a month — close to three quarters of a full-time equivalent.

That gap does not appear as unfinished work. It appears as three attorneys each losing something like eight to ten hours a month to administration, which at any realistic billing rate is a far more expensive way to buy those hours than any of the four responses discussed above.

A twelve-attorney practice with a removal defense component

Assume 520 open matters, a meaningful share in immigration court, four paralegals at 55% administrative time.

Court-weighted matters carry a higher administrative load — hearing preparation, evidence refresh across long gaps, packet assembly — so assume 1.8 hours per matter per month. Demand is about 936 hours. Capacity is four FTE at 160 hours times 55%, or 352 hours. The gap is roughly 584 hours: about 3.6 full-time equivalents.

A firm in this position is usually aware something is wrong but experiences it as a general sense of overload rather than a quantified shortfall. The value of doing the arithmetic is that it converts “we are stretched” into a specific number that can be compared against specific options — including the option of doing nothing, whose cost is now visible.

Run these numbers against your own caseload with the case capacity calculator, and see what covering a gap of that size costs.

What a working operating model looks like

Pulling the threads together, a firm that has this working generally has five things in place.

One system of record. All case work happens inside the firm’s own case management platform. Nothing important lives in an inbox, and nothing is copied out to be worked on elsewhere. This single decision removes an entire category of failure.

Deadline tracking as a process, not a person. Court dates and response windows are registered on receipt, tracked on a defined cadence, and escalated on defined criteria. Tracking that depends on one individual’s memory fails precisely when the queue is longest.

A review gate that cannot be bypassed. Draft work is stored in a state that cannot be filed directly, and attorney approval is recorded so it is auditable afterwards. This is what makes the supervision obligation real rather than aspirational.

A defined escalation list. Everyone knows what goes to the attorney immediately rather than into a queue: any deadline inside a set window, any notice indicating denial or a request for evidence, any apparent conflict, any suspected error in something already filed.

Continuity that survives absence. A single person holding all the context on an account is a single point of failure. A briefed backup is the difference between a colleague’s illness being an inconvenience and being a missed filing date.

If you are starting: a first ninety days

Firms that try to fix this all at once generally fix none of it. A sequenced approach works better, and ninety days is enough to get from “we think we have a problem” to a running arrangement.

Weeks one to two — measure. Capture administrative time honestly for one week across every role, categorised only as substantive or administrative. Pull your open matter count and median matter age. Do the capacity arithmetic. Resist the urge to fix anything yet; the number is the point, and acting before you have it means you cannot tell afterwards whether anything improved.

Weeks three to four — triage and write things down. Sort open matters into active, monitored and dormant bands and set a cadence for each. Write the escalation criteria explicitly: what goes to an attorney immediately rather than into a queue. This document is useful even if you never delegate anything, because it converts institutional habit into something a new person can follow.

Weeks five to six — settle the compliance position. If delegation is on the table, this is where consent language, the supervision protocol and the data processing agreement get drafted and reviewed. Do it now, before selecting a provider, so you are evaluating providers against a standard you have already set rather than accepting whatever they hand you.

Weeks seven to eight — scope and pilot. Start with the monitored and dormant bands, where the work is most procedural and the escalation criteria are cleanest. Grant scoped access to a defined subset of matters. Review output closely and frequently at this stage; the point of a pilot is to find the gaps in your own instructions, which are always larger than expected.

Weeks nine to twelve — widen or stop. Re-run the capacity arithmetic against the same inputs you used in week one. If administrative hours absorbed by attorneys have not moved, something in the arrangement is not working and widening it will not help. If they have, extend scope deliberately rather than by drift, and make sure a briefed backup exists before you depend on the arrangement.

The single most common failure in this sequence is skipping weeks one to two. Without a baseline there is no way to distinguish a genuine improvement from the feeling of having done something.

Glossary of terms that get confused

These come up constantly and are routinely used interchangeably when they should not be.

Frequently asked questions

The questions below come up most often when firms start thinking about this seriously. Fuller answers on the compliance side are in the compliance questions immigration firms ask before delegating.