We're shutting down CrowdPrisma, a bootstrapped applied AI company that was formed two years ago by my friend and I (both ML engineers) to solve free-text survey analysis. A ton of really important information can only be gathered by open-ended surveys and as NLP models were maturing it seemed like a nice and relatively niche market to go after.
I think we had a unique value proposition as our NLP engine was fairly sophisticated and we also integrated its output nicely with other non-text variables (check our demo on the site).
Getting significant business traction was difficult (for reasons which would warrant its own blog post), but now with GPT4, Office Copilot, etc coming online it'll be near impossible.
We learned a ton from the experience and wouldn't change one bit, but I do wonder, how many other companies must be out there like us..
Although it might seem like there's a generative AI startup frenzy going on atm (which is great), I do have the suspicion that in most domains, large tech companies will dominate the market, due to several reasons (another blog post maybe).. Will see.. anyways, interesting times ahead..
Basically MS is putting this into Office, and Google doing the same.. in a year's time this tech will be basically free to use for anyone w a computer and it's excel anyways where they gather their feedback from the surveys.. why would they then export it and go to another site when they can just ask questions about their data directly in excel.. we assessed this situation for weeks and basically long term, this product doesn't have a USP anymore..
That doesn't add any value at all to the conversation. They had a head start, even if tons of companies "will," they "do." Also, given the context limitations of transformers, the value in using a transformer here might be extremely dubious.
A head-start on the, now deprecated, tech they were implementing. Not on the interface and business idea, which is much more easier to replicate.
From a company with a tech advantage, one that they hoped to capitalize, they are now just a team with a front-end on what's now a publicly availabe commoditized technology - one where a third party controls the state of the art even.
It's a totally different business proposition, and a much less winning one. Not some VC money getting one either.
The technology they developed on that space has been superceded and deprecated. They once had a tech advantage, or so they thought, now they'd at best have a front-end to a widely available technology controlled by a third party.
The failure is that the company didn't have the goal of solving a customer issue. If it did, pivoting to use GPT in the survey space would be completely natural. The reason the company is shutting down is because it didn't just want to solve survey analysis, it wanted to do it using it's own tech. I can see how using GPT doesn't offer as large a moat as your own tech, but if you can't use GPT better than your competitors to solve your customer problems, you're not in the customer service business, you're in the software/tech making and selling business. Both may be ideal but the former is certainly feasible.
I would say don't let the hype bring you down. A very large general solution can never suit every single use-case that exists out there. In fact, you should think about bringing in an LLM-based solution as well. If you can't beat them, join them.
Show your customers where traditional ML beats LLMs hands down.
People are using LLMs as a hammer for everything. That hype will die down.
Hum... I've noticed that this post has a deeper explanation of what your product does than your entire site.
I am really curious if ChatGPT really beat a slightly modern highly specialized AI. Did you actually try it and get better results, or are you afraid of the hype?
Did you consider just using LLMs for your product? Even with LLMs it doesn’t seem trivial to format the survey data, design the prompts, validate them across a large dataset, and present the results with nice charts and links back to the original text etc.
With LLMs I’d imagine you could do even more interesting things with analyzing survey data. It seems like the core concept of your product is still useful, and could potentially do more now. Are you actually losing customers to GPT or just feel the end is near?
Also would love to read that future blog post! Thanks
OP here.. yeah, these are all great suggestions and we considered them at length (even validated some)... the problem is, before GPT, building a deployable (i.e. not just academic paper but something that actually works) NLP pipeline was fairly niche and hard and now the barriers to entry are almost zero (but over time it's definitely trending towards zero) .. we're a small bootstrapped company doing this on the side (until we can afford to switch over).. we can't afford to compete against 100x new companies doing this (some with way deeper pockets bc either they are VC or daddy funded).. so we're not going to.. also see my response about ms copilot etc above..
Respect for being in the arena and working on a useful problem. Sorry about the decision.
Can I try to spin this moment in a different way: you have a fantastic story to tell as you pivot go head first into AI.
You’ve built amazing, relevant skills in an adjacent space. You’ve shipped, which is more than what most people do.
Rest, refuel, and put together that narrative. It’s an interesting time right now and I think there’s a solid route to raising some kind of initial funding for round 2.
The only way you can make that claim is if you had X% (for a large X) of your paying customers leave and start using GPT4 instead, resulting in an unsustainable drop in revenue. My guess is you never had those customers, and your company never gained any traction at all. Hard to blame GPT for that.
It looks like your product is based on top of NLP tech, not the NLP tech itself. That's great because here's the good news: none of your customers care what you use as your back end. All they care about is whether you can solve their problem or not. With GPT-4 that capability should have improved.
>Or build a startup with an actual moat and not build on top of another product doing the heavy lifting.
That's exactly what they tried to do with a technical lead. This is what GPT-4 made obsolete.
Doing it with mere front-end or "execution" or marketing is close to impossible, as those are much more trivial areas and difficult to differentiate yourself. They're also mostly meaningless, as they depend on investment, which has no reason to go to them (as now they would just become one of many wanting to go to the area, and not the one's that has a tech-lead up its sleeve).
It's like they had an idea about a coffee place with their own secret syrup recipe for great tasting coffee, and they worked hard for 2 years on that recipe, hoping someone will invest in them to help them make it into a franchize.
Now someone came along and gives everyone a ready-made syrup mix, in the same exact vein, that is many times better tasting.
And all the suggestions here to "then execute better than others" are like "just use the new mix, and focus on making a better coffee place then, invest in the atmosphere and service" and so on.
Duh, well, everybody can try and do that. And there's no obvious way to differentiate yourself, or get investment to build a franchize just with this. Whereas if they still have their reciple (technology lead) and it still mattered, they'd have a major leg up the competition.
"He worked hard, built a tech lead, and we hope to get investment and beat the market with that"
is a totally different problem, and a much more hopeless situation than:
"He have a dead-end tech, there's a new better tech controlled by others, and we can only beat the competition by coming up with new ideas on execution, things that anybody can attempt, that have nothing to do with the hard tech work we did all those years".
As other comments have mentioned - I would see Chatgpt not as a startup killer, but an opportunity. Automatic surveying based on GPT just got a lot more viable. You still have significant headstart over others who would be approaching the same. Sure, you will have to throw out a lot of code, but you got customers, market recognition.
"Auto surveying" - give the "model" some basic information about the app/business and let it figure out what questions to ask, ask follow-up questions to customers and just come to me with the list of improvements by priority for customers. "AI product manager" Something like that is going to be huge, is definitely viable with current GPT model, but would still require a lot of work to get working (especially working around limited context and long term memory of GPT). If you start on this now, are you sure you won't be able to win this market?
I've thrown away some old code and replaced by the GPT. A lot less than you, but I am happy, as it ultimately allows my company to get more revenue from customers.
Everybody and their dog is now jumping on the AI bandwagon, which in many cases means they download LangChain, use a prompt they copied from someone's tweet, concatenate it with user input and call GPT.
Companies with actual domain knowledge will be in much better position because they will really know how to leverage the full power of these new models instead of being blinded by the bling that (cherry-picked) works nicely in 50% of the cases.
My friend and me have a good natured competition in making an AI that writes stories. He's using a custom tuned LLama, I'm calling ChatGPT.
Making cutesy little stories is easy. Making anything you'd actually be interested in reading is freaking hard! We're not writers! We don't know enough to prompt/select/guide the AI to do it well!
Same in other domains. LLMs are powerful tools but they'll make you more poweful if you have the right domain knowledge (and seems these guys do). Don't give up, leverage it!
Because instead of their own differentiating tech, they'll have a me-too product anybody can easily replicate - and they'll need to base it on a third party provider too...
I belive there is room for dedicated tools in most areas where gpt4 is good, even if they are built ontop of the gtp4 apis. A dedicated tool can have all the nuance and integrations that the broad tools never will have.
I see you are from the uk - have you approached entrepreneur first? Or any other similar accelerators. If anything i think now is the time to grow. Also i cant find examples of how your products works, nor do i see an api or anything else that would indicate you are have experience in marketing what you built - perhaps bring someone on board to cover that gap?
Anyways, I'd never do an accelerator kinda thing again or even take funding unless it makes a ton of sense.. turns out there are a lot of business (not unicorns) you can build with a buddy on the side and not get caught up in the craziness of raising, posturing, raising again, playing someone else's game, etc..
> turns out there are a lot of business (not unicorns) you can build with a buddy on the side and not get caught up in the craziness of raising, posturing, raising again, playing someone else's game, etc..
This! Precisely my take away from my own experience with EF (nothing against them, they are quite cool to be honest).
The people i've met at EF, failed startups or not, have been absolutely.brilliant. Keep it up!
Seems like a great vertical for AI. Many successful startups have taken commoditized technologies and glued them together in novel and interesting ways. Why can’t you guys do that with GPT4? Don’t confuse the technology you developed with the product you are offering to customers
Cold email wont work. You would need to get a warm introduction to the ceo of the company. Nurture that relationship a bit on the basis that you are both in the market and working on similar problems, then once a rapport has been established you can broach the topic of acquisition.
I think we had a unique value proposition as our NLP engine was fairly sophisticated and we also integrated its output nicely with other non-text variables (check our demo on the site).
Getting significant business traction was difficult (for reasons which would warrant its own blog post), but now with GPT4, Office Copilot, etc coming online it'll be near impossible.
We learned a ton from the experience and wouldn't change one bit, but I do wonder, how many other companies must be out there like us..
Although it might seem like there's a generative AI startup frenzy going on atm (which is great), I do have the suspicion that in most domains, large tech companies will dominate the market, due to several reasons (another blog post maybe).. Will see.. anyways, interesting times ahead..