By Sagar Shankaran, Founder of CallSphere
Flu season moves and January resets the formulary. How 2026 forecast models read a pharmacy's messy history, and what over-buying one generic really costs.
Key takeaways
Seventy-two percent. That is roughly the share of retail and e-commerce operators now using AI for pricing decisions, alongside about 66% using it for inventory and 48% of manufacturers using it for demand forecasting. Those numbers describe chains with category managers and a data team — the mass merchant three blocks away that already knows, to the case, how much children's ibuprofen to send its stores in the second week of October.
Meanwhile the ordering technician at an independent is standing at the terminal at 4:40 on a Tuesday, working a suggested order out of the wholesaler portal, deciding by memory whether last week's spike in one antibiotic was a real trend or one pediatrician's bad week. The wholesaler cutoff is 6:00 pm. She has forty minutes and about nine hundred line items she could look at.
Demand forecasting in a pharmacy means predicting, item by item and week by week, how many units of a specific NDC you will dispense — so the bottle is on the shelf the hour the patient asks for it, and not three months earlier. The reason independents have never had this is not that they didn't want it. It is that their history is short, jagged and full of one-off events, and until recently the tools choked on exactly that kind of history.
Every independent lives with four patterns that a simple minimum/maximum setting handles badly.
Respiratory season, which moves. Flu shots start arriving in August and the clinic wave runs from mid-September through November, with Medicare open enrollment sitting on top of it from 15 October to 7 December. But the illness curve itself moves by weeks from year to year, and when it lands early, oseltamivir capsules, pediatric suspensions and rapid tests all go at once against a min/max set from last year's calendar.
The January reset. On 1 January, deductibles restart, formularies change, preferred products flip, and prior authorizations that were approved in November come back as reject 75. For two weeks your dispensing mix changes without your demand changing at all — same patient, same condition, different NDC. A forecast built on units of a product rather than on patients and their plans is wrong in both directions at once: dead stock on the product that lost preferred status, out-of-stocks on the one that gained it.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
Shortages and substitutions. When supply of one strength goes short, demand does not vanish — it lands on the next strength, on another manufacturer's product, or on a compounded alternative. The pattern looks like a collapse in one line and an unexplained surge in another.
Lumpy institutional volume. One group home, one hospice contract, one prescriber who does most of the diabetes management in town. Lose that prescriber to a health system and a slice of your line items goes quiet at once. Add a twelve-bed facility on cycle fill and you need a month of specific items you have never carried.
The forecasting sold to independents in 2024 was, underneath, a moving average with a seasonal bump. It needed years of clean weekly history per item, it treated every gap as a zero rather than as "we were out of stock," and it had no way to know that 1 January is different from 2 January. So it recommended stock for the flu season you had last year and quietly punished you for every stock-out.
The 2026 tools do three things the older ones could not. They work from short, messy, gappy history instead of demanding clean years. They can read the difference between "nobody wanted it" and "we didn't have it," so a stock-out no longer teaches the model the wrong lesson. And they can take in things that are not sales numbers at all — your own plan-change notices, the shortage bulletins you already receive, the fact that the local school district went back a week early — and use them as reasons rather than noise. That is why forecasting and pricing are now two of the highest-adoption uses of AI in retail rather than a science project.
flowchart TD
A["Nightly dispensing history pulled from the pharmacy system"] --> B["Forecast reads season, plan resets and shortage notices"]
B --> C["Suggested order lands in the ordering tech's queue at 4:30pm"]
C --> D{"Does the tech agree with the odd lines?"}
D -->|No| E["Tech overrides and types the reason in one line"]
D -->|Yes| F["Order transmitted before the 6:00pm wholesaler cutoff"]
E --> F
F --> G["Next-day receiving and actual fills recorded"]
G --> B
The new version of 4:40 pm is not "the computer orders for you." It is that the suggested order arrives already sorted by where the money is, with a written reason on every unusual line. Eleven lines say "up 40% because the plan year reset moved these patients to this manufacturer." Six say "hold — you have 90 days of stock and the 30 patients on this came from one prescriber whose fills dropped 60% in three weeks." Two say "buy short, national shortage bulletin last Thursday." The technician's job becomes reading twenty flagged lines instead of scanning nine hundred, and her overrides go back into the record so the next forecast knows the reason.
The pricing half is the piece most owners have never touched. Third-party reimbursement you do not set — the plan sets it, and your margin on a generic can be under a dollar. But you do set the cash price, and that is where independents both lose patients and leave money on the table. The same tools that forecast units can compare your cash price on a common generic against what a discount card pays and against your real acquisition cost from the primary and secondary wholesalers. The same applies to the front end: the cold and cough aisle in November, the greeting card rack, the durable medical equipment nobody has repriced since 2023.
Assumptions, all illustrative: a single-store independent filling about 1,100 prescriptions a week, average gross profit of $9.40 per prescription, one over-bought generic and one under-bought seasonal item.
| Line | Assumption | Amount |
|---|---|---|
| Over-bought generic, 40 bottles at $32 | bought on a deal in October | $1,280 |
| Bottles still on the shelf at expiry | 26 of 40 | $832 at cost |
| Recovered from the reverse distributor | non-returnable generic | $0 |
| Days out of stock on the seasonal item in December | early flu season | 9 days |
| Prescriptions transferred out over those days | illustrative | 31 |
| Direct gross profit lost | 31 × $9.40 | $291 |
| Households that moved maintenance scripts too | 6 households, 14 scripts/month | $1,579 per year |
| One season, two items | $2,702 |
Nobody has exactly one over-bought item and one stock-out. Size it with your own numbers: last year's expiry write-offs, plus the transfer-out report your pharmacy software will produce, plus your average gross profit per prescription.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
It cannot see a prescriber leaving town until the fills stop. It will not know the family practice on the corner hired a nurse practitioner who prescribes differently, or that the group home changed management companies, until those show up as numbers weeks later. Anything driven by one human decision rather than a pattern is a blind spot, and that is a large share of an independent's volume. The fix is not a better model — it is that the pharmacist who plays golf with that prescriber tells the ordering technician.
It will also be wrong about brand-new drugs and about the first six weeks of any shortage, because there is nothing to learn from yet. And it should never set order quantities on controlled substances without a pharmacist signing off — your limits there are a DEA question, not an inventory question. Keep C-II ordering exactly where it is.
A wholesaler's suggested order is built from your purchase history and your min/max settings, which means it is very good at telling you to buy what you bought before. It does not know your plan-change notices, it does not distinguish a stock-out from a lack of demand, and its incentive is not identical to yours. Run the two side by side for a month on a hundred fast movers and compare the misses in both directions.
Partly. It will spot the substitution pattern — one strength collapsing while a neighbouring one surges — faster than a person scanning reports, and it can read the shortage bulletins you already receive. What it cannot do is tell you which secondary wholesaler will actually have product on Thursday. That is still a phone call and a relationship.
It will show you where your cash price sits against your real acquisition cost and against what discount cards pay for the same item, and flag the ones that look badly off. Deciding what to charge in your town, for your patients, is judgement — the tool's job is to stop you from finding out two years later that you were selling a common generic below cost.
Expect a first useful pass within a few weeks, because it works from short history, and expect it to get materially better after its first respiratory season and first January with you. What speeds it up most is not more data — it is your technician typing the reason on every override.
Before you buy anything, pull two reports out of your pharmacy system: prescriptions transferred out in the last twelve months, and inventory written off to expiry in the last twelve months. Put those two numbers next to each other. That is the size of the prize, in your store, in your handwriting — and it is the only number that will tell you whether this is a $2,000 problem or a $30,000 one.
One knock-on effect worth planning for: better stock means more callers asking "do you have it?" and "is it ready?", and those calls land during the exact hours your staff are filling. CallSphere builds voice and chat agents that answer those calls, check whether an order is ready, take refill requests and pass anything clinical to a pharmacist — so a forecasting win shows up as a filled prescription rather than as a phone that rings out at 4:40 pm.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Idle crew days and wrong merchandising cost real money. How 2026 demand models read gappy scale-ticket history and call mill quota shifts weeks ahead.
A 250-claim PBM desk audit eats three weeks of a technician's time. Four agents splitting the pile turn it into an afternoon and a 19-claim exceptions list.
Newsprint is ordered nine weeks ahead and the draw is a guess. How 2026 forecasting models cut returns, catch sellouts and set renewal prices by cohort.
A 12% miss on the treated buy costs a three-branch dealer up to $59,000. How 2026 demand models handle messy, seasonal building materials history.
Why optical frame boards die in January while December runs out of chairs, and how 2026 demand forecasting and pricing models fix both from your own exports.
Peak is 33 operating days of staffing risk. How 2026 demand models handle lumpy courier history, price stops by density, and where they still get it wrong.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI