So I run a DTx bot that I cant quite say is therapy legally, I mean it also isn’t therapy because therapy is based on a relationship between two people but I digress…
I run an emotional reflection bot that helps people clarify their emotions get through their day etc…
So to answer your question “How do you be transparent about it”
That is up to you - what I did was I found the average tokenage per message, and set my rate at 100% that. So my tokenage is .06 / 1000 - you’ll notice this is the same price as gpt4 bot tokens. This is because the average of your tokens is always going to lean toward .03 / 1000 because prompt tokens compound - bot tokens are almost non existent at the end of the day.
I also found that after I had a 20 message chat - so really 10 messages each, that I would find significant relief by that point
So I simply made a calculator that would show people if you pay X amount of dollars (this is the variable they put in) - for example (making it up)for 6 dollars you can have 10 20-message chats in a month.
It takes maybe 20-30 minutes to have that long of a chat depending on if you’re really using it and sharing and looking for clarity.
The idea is try to be as transparent as possible about tokenage, but also - the consumer has no idea what a token is (and lets be real, do we even?) so explaining things to them in other metrics makes sense to them. But then leaving the burden onto them. A lot of people will have a bill they don’t want to pay at the end of the month but thats just the growing pains, theyll realize its an expensive service and taper their usage or unsubscribe, but thats just the nature of SaaS (assuming you’re running a SaaS business)
Why though? Explain the rules of the game, hand them the rope and if they hang themselves its not your problem, they’re adults who were explained the rules. That is literally how a credit card works.(That is a horrible analogy to make for someone in the business I’m in, but I’m talking business here.)
People are too caught up on this “tier” system. So for me right, what if someone is having a deep conversation and their pre-paid tokenage shuts off? I don’t feel like that’s helping the person. The person should always be able to talk (but that’s MY specific industry).
If it works for your userbase, then great, but you can’t tell me that a 3 x 7 matrix of values and descriptions is simpler for an end user than “it’s 20$ a month”
I’m not trying to say that at all. I’m just trying to cover a reasonable range of options. And, I don’t know if it will work for my userbase, because I haven’t tried it yet. Heck, I haven’t even started marketing yet – because I’m still trying to figure out this part!
Not trying to limit users tokens, but not trying to be on the hook for tokens used, but not trying to confuse the end user, but trying to explain the relationship between queries, tokens and costs… Man, that’s hard!
I mean listen we are the early adopters if you want to succeed in your respective spaces and be the first to market than you have to do these bullshit conversions and figure it out because unfortunately OpenAI doesn’t do anything to make it easy for us.
You just have to find a way to communicate to the user ROUGHLY how much they will pay. If they get use out of it - 5 dollars a week for something you find incredibly useful is a small price to pay
There’s 0 way to stay profitable unless your users are paying at least 100% mark up on their tokenage. Figuring out the freemium is tricky. But we’re all in the same position of “how do I market a chatbot and stay profitable”
For early adopter products and services 1000-10000% is not unheard of, you are all early AI adopters and your initial customer base will be likewise risk taking early adopters, typically they are well read and have a larger disposable income than the median, they also typically understand that new technology is a risky venture and that premium surrounds the risk.
I think a range of offerings makes sense, some will opt for pay as you go and some will opt for fixed services for a fixed fee, most modern market research shows that an average non technical user prefers a fixed monthly outlay as it aids in budget management, but that may not be your typical early adopter of your service.
As has been said, if you are offering value and your offering is novel, at least for now, you have a captured market.
Or your goose might already be cooked. Where’s your product’s multi-modal computer vision and linked output of searched web citations and images? Rapid deployment on active directory and knowledge integration?
Microsoft was able to cut a great deal for itself. The corporation bargained for what Nadella calls “non-controlling equity interest” in OpenAI’s for-profit side—reportedly 49 percent. Under the terms of the deal, some of OpenAI’s original ideals of granting equal access to all were seemingly dragged to the trash icon. (Altman objects to this characterization.) Now, Microsoft has an exclusive license to commercialize OpenAI’s tech. And OpenAI also has committed to use Microsoft’s cloud exclusively. In other words, without even taking its cut of OpenAI’s profits (reportedly Microsoft gets 75 percent until its investment is paid back), Microsoft gets to lock in one of the world’s most desirable new customers for its Azure web services
(from Wired magazine October 2023 issue)
I just was going to recommend a lower tier than ChatGPT plus to a forum user, but checked, and the provider drastically increased the price of entry. They likely did a re-calculation of actual costs of a $10/mo plan for 10 gpt-4 per three hours.
Honestly, I think that’s possibly the most insane pricing strategy I’ve ever seen for anything.
Maybe that will work for you, but it’s so very convoluted.
I don’t think anyone is clamoring for anything like this.
Beyond paying the flat fee, they need to pre-load credits, then they need to start thinking about tokens, which you’ve “helpfully” approximated as number of queries, but then there are tier levels with different token costs per 1,000 tokens so different number of queries…
Just charge $20 / month and give everyone 40 messages / month with no roll-over and if you really want to let people use more, charge $10 for an additional 40 messages which do roll over.
I honestly believe, 90% of your potential customers are going to look at this and say,
I have no idea how much this is going to cost me…
And just move on to something with a simpler, consumer friendly pricing scheme.
So, it looks like we’re on the same page. My pricing mock-up achieves exactly what you suggest: 38 “queries” per month for $20 as opposed to 40 “messages” per month at $20. I can do 40 messages per month, no problem.
The question, then, appears to be: How to present the concept in a less convoluted way?
I just want to be able to point out that that same $20 will buy more messages on a lesser model. I mean, somebody might be able to get by using Bronze (aka. gpt-3.5-turbo-16k), in which case they can get a whopping 384 messages for the same cost. For this particular dataset, I wouldn’t advise it, but it is nonetheless an option.
Or, as I think about it, perhaps the answer is two tiers: Simple and Convoluted. No, I’m serious.
Simple Plan = $20 a month, 40 messages.
Convoluted Pan = The pricing mockup above.
I’ll give it another name, but that’s the idea. Thank you, @elm. You may have just helped me to the next stage of figuring this out!
Additional services. User can (in addition to submitting questions and receiving responses via our website interface).:
query via email
query via sms
Ask questions and receive responses in English, Spanish, Chinese, Korean and more
keyword search documents
browse documents
export query history to csv file for future reference or model training
generate analytics on individual usage via text to sql
All that, in addition to getting conversational chat access to the most extensive library of documents on the niche subject available anywhere (that would include Lexis-Nexis and CaseText) with, so far, a 95% success rate on every exam (on it’s core subject matter) it’s taken.
If that ain’t enough, then they can go subscribe to one of those “Chat with your PDF” services listed earlier. Clearly not a customer I’m going to ever get under any circumstances.
Your basic tier should cover a worst case scenario of typical usage and have rate-limits to prevent taking damage. Simple consistent fee. If it’s not understandable within a minute then it’s a bad pricing model.
You can offer a pro tier which is much more nuanced.
My current working basic tier is $20/month for 40 messages (i.e., queries). That is at Gold (gpt-4) level. That should be pretty simple. In fine print I can add the total messages for same amount at the lesser levels.
I’ve already worked out what the average tokens per query are with my dataset, given myself a 100% markup and a buffer so that I’m pretty comfortable with the advertised rate. And, it’s actually tokens I’m counting, not “messages”, so the typical user is going to end up with more messages per month than the posted limits anyway.
