You already use AI. Most doctors do by now, even the ones who don't mention it.
You've asked it to explain a guideline you half-remembered. You've asked how to price a package, or how to handle a staffing problem you didn't want to discuss with your staff. And it gave you a good answer — properly good — but a general one. Because it had no idea who you were.
From today, it does. You can connect NeftX to Claude or ChatGPT, and the same conversation you were already having now happens with your hospital's actual numbers in front of it.
Why we didn't just put a chat box in NeftX
That was the obvious move. Every hospital software company is about to do it. A little chat bubble in the corner and an "AI" badge on the pricing page.
We think that's the wrong thing to build, for a simple reason: whatever we put in that box would be worse than what you're already using.
You already have Claude or ChatGPT. It's already excellent. It already knows medicine, pharmacology, business, inventory, pricing — genuinely more than most consultants you could hire in this city, available at eleven at night.
The intelligence was never the missing piece. Your data was.
So we built the connection instead of the chat box. Your hospital's information, safely available to the assistant you already use.
Three things follow from that, and they all favour you.
You're not paying for a second-rate AI. We're not asking you to use our version of something you already have a better one of.
There's nothing new to learn. No fresh screen, no training session for your staff. You type into the same box you already type into.
It improves on its own. When Claude or ChatGPT gets better next year — and it will — your setup gets better that day. We don't have to ship anything. You don't have to upgrade anything.
What actually changes
A general question gets you a general answer. Here's the difference once it can see your hospital.
Before: "How do I reduce pharmacy wastage?" → a sensible explanation of reorder levels and expiry tracking. Correct. Unusable.
After: "Look at what my pharmacy has sold this year and what's on the shelf. What am I over-ordering, what keeps running out, and what should I buy this week?" → an actual purchase list, built from your sales, your stock, your expiry dates, and the fact that monsoon is six weeks away.
Some others worth trying:
"Why did my pharmacy revenue stay flat but my profit fall?" — the kind of thing you'd normally discover in March while doing the books.
"Am I buying the same medicine from two vendors at different prices?" — nobody has ever put those two invoices side by side, because it isn't anybody's job.
"This patient has been coming for three years. Walk me through what's changed." — eleventh visit, file too long to read, four minutes before they sit down. It has read every line and it also knows the pharmacology. A colleague who did the reading, not a diagnosis. The call stays yours.
And the one no software can answer:
"What's costing me money that I probably haven't noticed?"
There's no screen for what's wrong with my hospital. Most of what comes back you'll already know. One thing you won't — and that one is worth the twenty minutes.
When it stops being a novelty
Ask something once and it's a nice trick.
The shift comes when you save it. Both Claude and ChatGPT let you write instructions once and reuse them. Set one up for your Monday purchase list. Another for a one-page summary before your first patient. Another to prep any file you're about to walk into.
Written once, it runs every week. That's the point where it stops being something you tried and becomes how your week works.
About your data
Straight answer, because you should ask.
When you ask about a patient, that patient's information goes to Claude or ChatGPT. There's no version where it doesn't — that's how it reads anything.
What we control: it can only read, never change or bill anything. It runs on your own login, so it sees exactly what you're allowed to see and nothing from any other hospital. It's switched off until you turn it on, and you can cut it off any time. Only what's needed for the question gets sent.
Beyond that, it's Anthropic's and OpenAI's policy rather than ours, so read it — and use a paid plan rather than a free account if patient data is involved.
Worth noting: most of the questions above are about your business, not your patients. Every pharmacy, vendor and margin answer works without a single patient record going anywhere.
Where to start
Don't start with the impressive thing. Start with the question you've been wondering about for a year and never had a free Sunday to check.
Why margin fell. Which vendor is quietly expensive. What's dying on the shelf.
You've been typing that question into an AI for two years and getting general advice.