When someone first told me that AI could predict which medicines my pharmacy would need next week, I thought it was the kind of thing that works in Silicon Valley but not in a medical store in Lucknow.
My pharmacy does about Rs 8 lakh in monthly sales. We carry 4,200 SKUs. My ordering system for the last twelve years has been simple — my helper walks the shelves every morning, writes down what is running low on a notepad, and I call the distributors. Supplements and slow movers get ordered when I remember, which is not often enough.
It worked. Until I started tracking what "worked" actually meant.
Over six months, I measured three things: how many times a customer asked for something we did not have (stockout), how much stock we threw away due to expiry, and how often I panic-ordered at higher prices because we ran out of a fast mover.
The results were uncomfortable. We were stocking out on at least 8 to 10 products per day. We wrote off Rs 1.4 lakh in expired stock over six months. And we paid a "rush premium" of roughly 3% to 5% on emergency orders — about Rs 15,000 over the same period.
That is when I agreed to try AI-powered forecasting. Not because I believed the hype — because the status quo was expensive.
What AI Forecasting Actually Does (In My Pharmacy)
Let me demystify this. The AI is not some science fiction robot deciding what to order. Here is what actually happens:
The software (GoMeds AI, in my case) looks at my billing data from the last two to three years. Every transaction. Every product. Every day. It identifies patterns that I cannot see because I am too close to the business.
What the AI noticed that I did not:
- Cetirizine sales spike 40% in the second and third week of October — not just "allergy season," but specific weeks
- My Metformin 500mg sales are growing 8% quarter over quarter because two new diabetologists opened clinics nearby
- Demand for specific paediatric antibiotics drops by 60% during summer holidays (fewer school-going kids = fewer infections)
- One particular antacid brand was declining steadily because the popular gastroenterologist in our area switched to prescribing another brand six months ago
I knew some of these things vaguely. The AI knew them precisely and adjusted reorder recommendations accordingly.
The First Three Months: What Changed
Month 1: Scepticism
The system generated its first set of weekly reorder recommendations. I did not trust it completely, so I ordered what it suggested plus a 20% buffer on fast movers. Some of the recommendations surprised me — it suggested reducing my order of a particular cough syrup by half. I ordered the full amount anyway because "winter is coming."
Result: The cough syrup sat on my shelf. The AI was right. That particular brand had been losing share to a newer entrant, and the model caught the trend from my own billing data.
Month 2: Cautious Trust
I started following the recommendations more closely, overriding only when my gut strongly disagreed. The stockout complaints from customers dropped noticeably. I was not checking exact numbers yet, but my counter staff mentioned that fewer customers were leaving without buying.
The system also flagged three products that were overstocked relative to their sales velocity. I stopped reordering them and let existing stock sell down.
Month 3: The Data
After 90 days, I pulled the numbers:
| Metric | Before AI | After AI | Change |
|---|---|---|---|
| Daily stockouts (products not available) | 8–10 | 2–3 | -70% |
| Monthly expiry write-off | Rs 23,000 | Rs 9,500 | -59% |
| Emergency/rush orders per month | 12–15 | 3–4 | -73% |
| Inventory value (same sales level) | Rs 14.2 lakh | Rs 11.8 lakh | -17% |
The inventory value reduction was the surprise. I was carrying Rs 2.4 lakh less inventory while actually stocking out less. The AI was not just ordering smarter — it was ordering less of the things I did not need and more of the things I did.
How It Handles Seasonal and Unpredictable Demand
This was my biggest concern. Lucknow has extreme seasonal variation — winter respiratory infections, monsoon dengue scares, post-Diwali gastrointestinal issues, and summer dehydration products. Can an AI handle this?
The answer is yes, but with a caveat.
What the AI handles well:
- Predictable seasonal patterns based on historical data (cough and cold medicines in December, antihistamines in October)
- Gradual shifts in prescribing patterns (a doctor changing from Brand A to Brand B)
- Consumption trends tied to day-of-week patterns (higher OTC sales on weekends)
What the AI handles with help:
- Sudden outbreaks (an unexpected dengue cluster in a neighbourhood)
- New product launches (no historical data to learn from)
- One-time events (a health camp driving temporary demand for specific products)
For these unpredictable scenarios, the best AI systems let you manually adjust. "Increase dengue-related products by 50% for the next two weeks" — you tell the system, and it adjusts its calculations. It does not fight you. It incorporates your input and learns from the outcome.
The Distributor Relationship Effect
Something I did not expect: AI forecasting improved my relationship with distributors.
When you order predictably — consistent weekly orders instead of erratic panic calls — distributors treat you better. My delivery reliability improved. My credit terms improved slightly. And because the AI generates purchase orders automatically, my ordering process went from 45 minutes of phone calls to 10 minutes of reviewing and approving recommendations.
The Honest Limitations
AI forecasting is not perfect, and anyone who tells you it is is selling you something. Here is what I have experienced:
It needs data to work. If you have been using billing software for less than a year, the AI does not have enough data to make accurate predictions. It needs at least 12 months of transaction data — ideally 24 months — to identify seasonal patterns.
It does not replace your judgment entirely. I still override the AI's recommendations about twice a week. Sometimes I know something it does not — a new doctor opening nearby, a local health event, a distributor offering exceptional terms.
It works best for fast and moderate movers. For your top 500 products (which probably represent 80% of your revenue), the predictions are very accurate. For long-tail products — the speciality eye drop that one patient orders every three months — the AI has less data and less accuracy.
It does not fix bad purchasing habits. If you accept every distributor scheme regardless of demand, the AI will keep recommending lower orders while your shelf fills up with scheme stock. The AI shows you the right path. You still have to walk it.
Is It Worth It for Your Pharmacy?
My honest answer: if your pharmacy does more than Rs 5 lakh in monthly sales and carries more than 2,000 SKUs, yes.
Below that threshold, the volume might not justify the software cost, and manual ordering can work reasonably well.
Above that threshold, the cost of stockouts, expiry, and overstock is almost certainly more than the cost of the software. In my case, the AI saved me roughly Rs 3.5 lakh in the first year (reduced expiry + reduced emergency orders + reduced inventory carrying cost). The software costs me about Rs 60,000 per year. The math is not even close.
If you want to see what AI forecasting looks like with your own sales data, GoMeds AI Pharmacy Management Software includes demand forecasting trained on your specific transaction history. Book a demo and ask them to show you the forecasting module with real pharmacy data.
Rahul Bansal owns and operates a retail pharmacy in Lucknow. He has been in the pharmaceutical retail business for 14 years and adopted AI-powered inventory management in 2025.
Tags
Written by Rahul Bansal
Published on 15 February 2026
