The Frame Order Goes In at Vision Expo in March. The FSA Rush Hits in December. 2026 Forecasting Closes That Gap

Count them. Walk the frame board on the last Tuesday in January and pull every piece hanging there since Vision Expo West in September. In a lot of independent practices that number lands between sixty and a hundred and twenty frames. At $55 to $90 wholesale apiece, that is four to ten thousand dollars of the practice's own cash on a wall, in colors and sizes the optical manager already knew would not move.

Now the other end of the calendar. Between 8 and 31 December, the front desk turned people away: patients whose flexible spending money died on the 31st, patients whose benefit reset on 1 January, parents who did the math on the twelve-month clock. Told the first opening is 14 January, a fair share drove to the mall. The practice never billed the 92014, never sold the $340 optical ticket, never got the annual contact lens supply.

Dead frames in January and a full book in December are the same mistake pointed in two directions: the demand was guessed, nine months early.

The same error, twice a year, in opposite directions

Demand forecasting in an optical practice means one thing: knowing how many exam chairs, and how many of each frame group, lens design and contact lens modality you will actually sell in a given week, early enough that you can still change the order or the schedule. It is not analytics; it is deciding in March what the board should look like in November.

Optometry has a demand shape almost nobody else has. The August back-to-school block, driven by school screenings and sports physicals, weighted to pediatric exams and cheap durable frames. A soft stretch through late September and October. Then the last seven weeks, when flexible spending deadlines, met deductibles and expiring benefits land at once. Then January, when every benefit resets.

Spreadsheet forecasting handles none of this. Those peaks follow calendar rules rather than trend, one shifts with when Thanksgiving falls, and frame-level history is sparse: a model in one color might sell nine units in three years — not enough for a trend line, but plenty for a model that also reads the twenty-six similar frames beside it on the board.

What the buying decision actually looks like today

Here is the honest version. The frame rep comes in on a Thursday with the new collection, the optical manager walks the board, whatever looks tired gets swapped. The mental rule is last year's order plus ten percent, minus ten if the year felt bad. Twice a year the owner goes to Vision Expo East in March or Vision Expo West in September, writes orders on show terms, and comes home with a spreadsheet nobody reconciles against actual sales until the accountant asks about inventory in February. The scheduling side is less deliberate still: exam templates get set once and left alone, and the December crunch gets handled with a double-booked slot at 4:40pm.

The workaround everyone pretends is fine: the slow-moving-frame report inside RevolutionEHR, Eyefinity OfficeMate or Compulink. It is accurate, and it tells you what already went wrong. Nobody runs it before the order, because reading it takes an hour and interpreting it takes judgment the report does not supply.

What changed in 2026: the models stopped needing tidy history

Demand forecasting became one of the most-adopted uses of AI in manufacturing — around 48 percent of manufacturers now run it — and pricing optimisation reached roughly 72 percent in retail and e-commerce. Those are the two trades an optical straddles: a small manufacturer of a custom product and a specialty retailer of frames, in the same building.

What matters for a nine-chair practice is that the current generation of models copes with the exact data you have: short, gappy, seasonal, full of one-off distortions like the week the doctor was at a conference or the month a plan changed its frame allowance. The older approach needed that cleaned up first. It does not any more. You hand over three years of messy exports, say in plain words what happened in each odd stretch, and those stretches get accounted for rather than thrown out.

Then there is cost. Capable models have fallen roughly tenfold in price since 2025, so running this weekly on every frame group is cheap. Claude Cowork, launched January 2026, and ChatGPT Work, which arrived on 9 July 2026, take a goal and a folder of exports and hand back a finished spreadsheet.

flowchart TD
  A["Weekly sales export from RevolutionEHR"] --> B["Model reads 3 years of frame, lens and exam history"]
  B --> C{"Will this frame group sell before the January reset?"}
  C -->|Likely| D["Reorder at the rep visit, adjust the size run"]
  C -->|Unlikely| E["Mark down, move to the trunk show, stop reordering"]
  D --> F["Optical manager approves the order line by line"]
  E --> F
  F --> G["December exam template opened to match forecast demand"]

A Tuesday in September, run the new way

The optical manager exports three files before she opens: frame sales by vendor, collection and color for 36 months; lens orders by design and add-on; completed exams by type and week. No patient names, no dates of birth — just what sold and when. She asks for two things: what to order at the September show, and what the December book should look like.

What comes back is a buy sheet, not a chart. Nine hundred and forty pieces on the board today; suggested board at 1,010 by 1 November; forty-one pieces to stop reordering, with sell-through next to each; a warning that the practice is under-bought in petite metal in the 47-19 to 49-20 range, because the August pediatric block converts into second-pair sunwear faster than the buying caught up with. And: the fourth week of December is forecast 31 percent above the December average, leaving the practice 26 exam slots short.

She reads it, disagrees with six lines, deletes them, takes the rest to the owner. Twenty minutes on a Tuesday. The old version took two evenings and usually did not happen.

The refraction fee and the second pair: the prices nobody sets on purpose

Pricing is the quieter half. Most practices have three prices set years ago and never revisited: the refraction fee billed under 92015 when the medical carrier will not cover it, the second-pair discount, and the premium lens tiers — anti-reflective coating, photochromic, progressive designs — where the margin lives. Practices sit anywhere between $25 and $75 for the same refraction. A model that has read three years of your own transactions can tell you what happened when your fee last moved, and whether the patients who declined it also declined the AR coating. That is not a guess about the market; it is your own history, finally read.

One hard boundary: your contracts with VSP, EyeMed, Davis Vision and Superior Vision govern what you may charge for covered services and materials. Nothing here overrides that. What you can move is the non-covered work and the private-pay optical.

One frame board, run through the arithmetic

Illustration only — use your own numbers.

Treat the halves differently. The $4,900 is cash you already spent, coming back once. The $10,142 from December recurs, because those 42 patients come back next year.

Where the forecast is wrong and the optical manager is right

A model cannot forecast a frame that has never sold anywhere. New collections have no history, and the tool will quietly under-order them. That call stays with the person watching what walks in the door.

It cannot see a plan change coming either. When a large local employer moves from VSP to EyeMed, your history stops predicting your future for two quarters. Tell it; do not expect it to notice. And it has no opinion about your town: the optical manager who knows the high school principal wears the tortoise acetate, and that eleven people asked for it after the graduation photos, holds information that exists nowhere in your export. Keep the human veto on the buy sheet, every line.

First step: export 36 months of frame sales and completed exams by week, and ask for one thing — a list of frames to stop reordering before the next rep visit. If it is wrong you lost an hour. If it is right you found the money for the new edger.

Frequently asked questions

Does this mean handing over patient records?

No, and you should not. Everything here runs on sales and schedule history. Strip names, dates of birth and record numbers before the file leaves your practice management software. If a vendor wants identified patient data, that needs a business associate agreement under HIPAA first.

My software already has a slow-moving-frame report. Why is this different?

The report tells you what died. The forecast tells you what to order before it dies, at the level of a size run and a color, in the time it takes to make coffee. It also touches the schedule, and no frame report ever told you the third week of December is short of chairs.

What about the January reset — same problem?

Same tool, different shape. December is capacity-limited; January is conversion-limited, because patients arrive with fresh benefits and no idea what they cover. Forecast the call volume for the first ten business days and staff the desk to it.

How long before I can tell it is working?

Two buying cycles, six to eight months, before the board turns. The schedule side shows faster — you will know by 5 January whether you captured December.

One last note. December capacity is not only a scheduling problem; it is a phone problem. The calls that decide whether those 42 slots fill arrive at 4:50pm, at 7:30 in the evening, and on the Saturday after Christmas, asking the same question: do I still have benefits this year, and can you get me in. CallSphere builds AI voice and chat agents that answer the practice line and website chat around the clock, book into the slot you opened, and pass the message to the front desk in the morning. It will not tell you which frames to buy. It will stop the forecast being wasted on a phone nobody picked up.