When You Cannot Hire More Technicians: Managing Session-Level Variance With the Staff You Have

O*NET reports 78,800 ophthalmic medical technicians employed in the United States as of 2024, with employment projected to grow much faster than average, 7 percent or higher, over 2024 to 2034, and approximately 12,500 openings projected across that decade. The BLS Occupational Employment and Wage Statistics for May 2022 recorded 66,060 employed at a mean annual wage of $42,480, with 39,880 of those in offices of physicians.

On the ophthalmology side, AAO EyeNet reported in May 2021 fewer than 60,000 ophthalmic technicians supporting more than 19,000 practising ophthalmologists. Ruth D. Williams, MD, writing in EyeNet in December 2021, put it at 59,960 technicians against roughly 18,000 ophthalmologists, about 3.3 per physician, and noted the shortage predates the pandemic.

The training pipeline is the binding constraint and it is not expanding. Review of Ophthalmology reported in October 2019 about 30 accredited programs across 19 states plus the District of Columbia. Four years later, Mai et al. in Ophthalmology Glaucoma reported approximately 30 certified training programs and noted the count had not changed despite rising demand. New efforts exist, including a short-course program described in CRSToday in June 2025, but no published outcome data on placement or retention was available at the time of writing.

The practical conclusion for an administrator is not despair. It is that supply-side effort should continue and should not be the only thing you do, because none of it changes next month.

Ten patients is not ten patients. A postoperative cataract session and a new-glaucoma workup session with visual fields, OCT, pachymetry and extended counselling are different jobs wearing the same patient count. Staff both from a ratio and you are over on one and under on the other, systematically, every week.

This is arithmetically unavoidable if you staff from a mean. A mean is right on average and wrong on most individual days. The published benchmarks are explicit about this if you read the conditions attached. The ASOA throughput figure of 3 to 3.5 patients per technician per hour, cited in CRSToday from studies of 2009 and 2015, is stated as assuming “a balanced schedule combination of long and short exams.” Most sessions are not a balanced combination. They are a block of one thing.

The two failure modes cost different things

Overstaffed session. You pay a technician’s hourly wage for capacity you did not use. O*NET puts the 2025 median at $21.91 hourly. The cost is real, bounded, and appears in payroll.

Understaffed session. Patients wait. The provider waits between rooms. Technicians absorb the gap by working through breaks. Workup quality degrades under time pressure. Some of that cost lands on next month’s turnover, which returns you to the supply-side problem you were already losing.

These costs are not symmetric and they are not in the same units, which is precisely why a single ratio cannot balance them. It is also why administrators over-hedge upward when they can afford to, and why they cannot afford to in a shortage.

The friction compounds, and it drives people out

The demand-side and supply-side problems are connected, and Mai et al. make the connection explicitly. Their argument in Ophthalmology Glaucoma is that workflow inefficiency, not compensation alone, drives technician turnover: time lost transferring data among systems and moving patients between examination rooms and testing devices is time not spent on the parts of the job technicians find worthwhile. Their prescription is device integration and automated documentation.

Chronic understaffing on heavy sessions is the same category of friction and it is not addressed by better devices. A technician who is under-resourced on Tuesdays every week is being asked to absorb a planning failure with their own effort. That is a retention problem generated by an allocation problem.

Step 1: establish whether you can measure it at all

Does your practice management system or EHR timestamp when a technician begins and ends a patient workup? If yes, you can do everything below. If no, that instrumentation is the actual first project, and it comes before any conversation about forecasting or software. This is the most common place practices discover they are not ready, and finding out costs nothing.

Step 2: pull twelve months of sessions

For each historical clinic session: date, site, provider, appointment counts by type, diagnostic tests performed, and realized technician minutes.

Step 3: plot realized technician minutes against patient count

If the points sit close to a line, your patient count predicts your technician need, and a ratio is serving you adequately. Stop here; you do not have this problem.

If the scatter is wide, the vertical spread at any given patient count is the size of your problem, measured in minutes, on your own operation. That is a number you can take to a physician partner.

Step 4: compute your current baseline error

For each session, calculate what your current staffing rule would have predicted and how far off it was. Take the average of the absolute differences. That single number is your baseline, and every future claim by anyone, including any software vendor, gets measured against it.

Doing step 4 before talking to anyone is the most valuable hour in this entire process, because it converts a vendor demo from a story into an arithmetic comparison.

You can see across sites. Most groups plan site by site, because that is how the schedules are owned. Variance that is invisible cannot be smoothed.

Technicians can actually move. This is the constraint that quietly kills most reallocation plans, and it is not primarily about willingness. It is about credentialing. Technician minutes are not fungible. IJCAHPO certifies at three general levels, COA, COT and COMT, plus specialty credentials in ophthalmic surgical assisting, ophthalmic ultrasound biometry, diagnostic ophthalmic sonography, and scribing. Layer on device-specific competency sign-offs, site privileges, and hour restrictions, and the set of technicians who can genuinely cover a given session is much smaller than the roster.

Travel time is accounted for. A technician who spends 40 minutes driving between sites has not added 40 minutes of capacity to the organisation.

Where those conditions hold, allocation across sites is the largest available lever that does not require hiring anybody. Where they do not, the honest answer is that the group needs to fix credentialing visibility and cross-site scheduling authority first, and no software will substitute for that.

1. Credentialing rules must be hard constraints, not warnings. A plan that assigns an uncredentialed technician to a session should not be produced. A system that produces it with a caution has moved the compliance burden onto whoever is reading the screen at 6pm.
2. A human must approve every plan. Forecasting produces a number. Coverage, fairness, staff preference, and the thing you know about this week that no system does are judgment, and they belong to a manager. Any product positioned as replacing that person is solving the wrong problem.
3. Every change must be auditable. In a credentialed clinical environment, who changed an assignment, when, and why is a record you will eventually need, and it should be tamper-evident rather than merely logged.