Ask HN: Do AI/ML companies actually have real customers?
58 comments
I find it hard to wrap my head around this question because "AI" is a relative (and moving) target.
Alexa would be considered advanced AI in the 90s, wouldn't it? Google Search or Facebook ad targeting might be, too. They certainly fit some definition of ML.
It sounds like you're thinking of companies that are doing simple data analysis and calling it AI, right?
Alexa would be considered advanced AI in the 90s, wouldn't it? Google Search or Facebook ad targeting might be, too. They certainly fit some definition of ML.
It sounds like you're thinking of companies that are doing simple data analysis and calling it AI, right?
Agreed. Google Search encompasses some of the biggest ideas of AI.
Also all the advanced image recognition/classification/manipulation that's powering a lot of Facebook's products. Definitely AI territory.
The most difficult thing to do is to pitch a product instead of a technology. This is more accentuated by the fact that we're mostly engineers and we're not good at pitching benefits vs features.
But I can assure you that customers do appreciate the AI component, especially when explained to them. I do a LOT of education in commercial meetings and demos. Some of our customers we invited to our offices for a more in-depth explanation of our technology.
My experience is that customers' interest is piqued by buzzwords, but if you are able to explain the tech in plain words, they are astonished and the sale is much easier.
To summarize, if you are raising money it may be worth it to buzzword the hell of your deck, but when selling the product the key is to focus on benefits + education.
Disclaimer: Am an AI startup founder (https://optimusprice.ai)
But I can assure you that customers do appreciate the AI component, especially when explained to them. I do a LOT of education in commercial meetings and demos. Some of our customers we invited to our offices for a more in-depth explanation of our technology.
My experience is that customers' interest is piqued by buzzwords, but if you are able to explain the tech in plain words, they are astonished and the sale is much easier.
To summarize, if you are raising money it may be worth it to buzzword the hell of your deck, but when selling the product the key is to focus on benefits + education.
Disclaimer: Am an AI startup founder (https://optimusprice.ai)
You are incredibly correct. If you are pitching the tech vs the product it's game over.
In the first startup I created, we made ML-based underwriting and credit scoring models for several banks and insurance companies. The results were astonishing. For instance, one insurance company improved its profit by ~70M$/year, thanks to our model (according to our client's report from actual data, not simulation).
But, it was actually tough to sell, because of several factors. 1) The internal risk team did not really like a technology that they could not understand. 2) It was hard to give an explanation of why it would work to the executives.
In the end, we were getting 2-4 big clients per year.
But, it was actually tough to sell, because of several factors. 1) The internal risk team did not really like a technology that they could not understand. 2) It was hard to give an explanation of why it would work to the executives.
In the end, we were getting 2-4 big clients per year.
How do you price a product like that before it has a track record? I'd imagine a high up-front sticker price is a deal killer, and it's probably not practical to tie it to profit (which may have too many other factors in it anyway).
And then what happened to the start-up?
The AI fired all the humans and became an Etherium contract.
I am not sure that a credit scoring system would become the first AI to reach consciousness :P
It is just an ML model, it just reads some inputs, then outputs some score.
It is just an ML model, it just reads some inputs, then outputs some score.
It is still operating with new owners and a new CEO. Since it accumulates track-records, it is slowly becoming easier to get projects. The first clients are still renewing their license.
AI is not a business model. It's usually just a component in a larger solution. You need to get a lot of other things right for AI to produce value, notably around the data you gather and pre-process. For most startups, that data is hard to access, and without it, their algorithms won't have much to train on.
While there are a few horizontal ML/AI/data science vendors, most AI is invisible, a component in some vertical solution, and consumed by people and businesses that don't really care how the results are produced, as long as it works.
Very few companies are good at applying AI across the board, like we see at Google, but the number of businesses using predictive models at least in some capacity is growing.
While there are a few horizontal ML/AI/data science vendors, most AI is invisible, a component in some vertical solution, and consumed by people and businesses that don't really care how the results are produced, as long as it works.
Very few companies are good at applying AI across the board, like we see at Google, but the number of businesses using predictive models at least in some capacity is growing.
Hi There - Could you please clarify what you mean by "horizontal vs. vertical" solutions/vendors? Maybe you could provide an example of each of those?
Horizontal means platforms, frameworks and tooling that are agnostic to the industry they're applied to.
Skymind is a machine-learning operations company (one of several): we help businesses train, deploy, monitor and update AI models. The same software is used by telecoms, finance, e-commerce and automotive companies -- horizontal.
There are countless vertical specific companies: e.g. Merlon applies ML to anti-money laundering.
Skymind is a machine-learning operations company (one of several): we help businesses train, deploy, monitor and update AI models. The same software is used by telecoms, finance, e-commerce and automotive companies -- horizontal.
There are countless vertical specific companies: e.g. Merlon applies ML to anti-money laundering.
We use DL for our grammar checker in Zoho Writer (https://zohowriter.com). We don't advertise our product as "The next generation intelligent word processor".
It's just one of many features that's best done with AI. I like it that way. Pitching utility is so much better than pitching how we built something with blockchain, AI and IOT combined ;)
It's just one of many features that's best done with AI. I like it that way. Pitching utility is so much better than pitching how we built something with blockchain, AI and IOT combined ;)
This is more a feature of your product than being a integral part of your business model. Would you agree on the distinction?
I hate to see that I need to login just to take a demo of your product. I find it a bad approach. First let the user checkout your product without intrusion.
We do have a trial editor without sign-in. Use this https://writer.zoho.com/writer/publiceditor.im
Right now the whole AI/ML is basically just a buzz word everyone uses because it is popular to use.
In reality, a true AI/ML company is basically just using AI/ML to create services/solutions for existing issues.
Case in point: google is using ai/ml to improve their speech to text and text to speech services.
In reality, a true AI/ML company is basically just using AI/ML to create services/solutions for existing issues.
Case in point: google is using ai/ml to improve their speech to text and text to speech services.
What do you mean exactly ? I've been running a machine learning / reinforcement / optimization service company since 2015 and our customers range from the largest EU corps to small and mid size startups. If you mean how many customers use AI/DL/RL in production right now, it's a smaller percentage of our customers, around 25% I'd say, but growing, slowly. Most is about research roadmaps and testing this testing that, though these can be yezr long projects.
DL integration into or as a replacement to pieces of software stacks, plus long term model updates, measurements, testing and avoiding regressions (very hard, think of ensuring all past samples are similarly predicted as before after a model update, that's beyond simply improving the accuracy) is difficult and standard practice is building up slowly.
DL integration into or as a replacement to pieces of software stacks, plus long term model updates, measurements, testing and avoiding regressions (very hard, think of ensuring all past samples are similarly predicted as before after a model update, that's beyond simply improving the accuracy) is difficult and standard practice is building up slowly.
Hi, what services do you use for data annotation? We've been using DL for some projects of ours and had to set up a data annotation team (remote) to get more control over the labeling process. We are looking to see if others might be interested in utilizing a service like ours, which will also help the people we've acquired gainfully employed. At present we are mainly looking at taking on small projects. If this is of interest, please do let me know! I posted about this here, but hasn't surfaced enough to generate conversation https://news.ycombinator.com/item?id=17874184
I know Seldon have real paying customers:
https://www.seldon.io
I can’t speak for them specifically or any other AI company but from cursory chats with folk in the industry suggests a lot of work is consultancy and proof of concept type stuff with bigger companies which they use to partially fund product dev on their own internal projects - some of which see the light of day.
I can’t speak for them specifically or any other AI company but from cursory chats with folk in the industry suggests a lot of work is consultancy and proof of concept type stuff with bigger companies which they use to partially fund product dev on their own internal projects - some of which see the light of day.
We use DL at my company. It's used to help and suggest the people in data input what some of the inputs should be. It's been getting better over time as it collects more data points.
Apart from that, I'm also personally responsible for implementing DL for data anomaly detection.
So yes, ML/DL is very much over-hyped right now, but it has real world tangible benefits. It is being used in multiple fields and I believe will grow quite fast.
Apart from that, I'm also personally responsible for implementing DL for data anomaly detection.
So yes, ML/DL is very much over-hyped right now, but it has real world tangible benefits. It is being used in multiple fields and I believe will grow quite fast.
Hi, what services do you use for data annotation? We've been using DL for some projects of ours and had to set up a data annotation team (remote) to get more control over the labeling process. We are looking to see if others might be interested in utilizing a service like ours, which will also help the people we've acquired gainfully employed. At present we are mainly looking at taking on small projects. If this is of interest, please do let me know! I posted about this here, but hasn't surfaced enough to generate conversation https://news.ycombinator.com/item?id=17874184
Hi Chaitanya,
That's quite an interesting project you have. Unfortunately probably not really relevant to my use case since two skills are necessary:
1. (good) control of the native language (not english) 2. Understanding and experience at the subject field
I don't want to give too many details but the workflow basically consists of taking a poorly labeled entry and expanding/filling out the missing fields. To do this you both need to speak the language and have some experience in the field.
As for the data anomaly detection I don't do any data annotation
Anyways, sorry it's not really relevant to my case but good luck, I think it's a great project that certainly has a lot of potential. Maybe in a future stage you might be interested with partnering up with a platform like coursera to form your workers to do slightly more complex data annotation jobs (like mine).
Regards
That's quite an interesting project you have. Unfortunately probably not really relevant to my use case since two skills are necessary:
1. (good) control of the native language (not english) 2. Understanding and experience at the subject field
I don't want to give too many details but the workflow basically consists of taking a poorly labeled entry and expanding/filling out the missing fields. To do this you both need to speak the language and have some experience in the field.
As for the data anomaly detection I don't do any data annotation
Anyways, sorry it's not really relevant to my case but good luck, I think it's a great project that certainly has a lot of potential. Maybe in a future stage you might be interested with partnering up with a platform like coursera to form your workers to do slightly more complex data annotation jobs (like mine).
Regards
Thanks for your response! Partnering with Coursera is an intriguing idea; will look into that. Thank you
I run a company that does financial portfolio optimizations and runs regressions on financial time-series data. These tools are in some sense the very first widespread use of "machine learning" techniques (at least in finance). I've noticed the same effects that the OP mentions: a handful of companies in this sector using the same tools may slap "ML and AI" all over their product and marketing. They're not incorrect - it just feels like a bit of a stretch to do so.
Sounds like your company is doing something similar to what my company was doing before I left. You're right - we weren't technically wrong, but it did feel disingenuous. I was never really sure if our customers actually derived value from our product in the way we wanted them to, or if it was simply a wildly speculative investment made by middle managers feeling FOMO after reading the Wall Street Journal.
I was at an AI company at my last job. They had customers, but it was largely speculative. I cannot say the product was worth what was being paid for it. That being said, there is certainly a problem domain that AI could be good for, if it ever lives up to the hype. Much more effective is human-in-the-loop AI, which is what Google, Facebook, etc. use and make lots of money from.
We are an AI company working out of Hyderabad / India, We have a very fast Classification algorithm as a product. and currently been used by 3 paying customers.
http://www.alpes.ai/
We are releasing our API next week. The idea being you can quickly try out the algorithm and see Results and Training time on your own datasets.
http://www.alpes.ai/
We are releasing our API next week. The idea being you can quickly try out the algorithm and see Results and Training time on your own datasets.
I think legal contract analysis is an area where machine learning is actually needed for core features to work.
Some products like https://www.contractstandards.com/ are beginning to appear. They need document classification, text sentence level classification, entailment, clustering are all necessary for making contract analysis easier for the user.
And these products need good domain experts to identify pain points where ml can help.
In ml as API side, one potentially valuable product suite I have seen is https://aylien.com/text-api/ . Even they are building additional products like news analysis.
But I am little sceptical on whether ml as API can scale in terms of addressing specific task-specific nuances needed for products building upon them.
Some products like https://www.contractstandards.com/ are beginning to appear. They need document classification, text sentence level classification, entailment, clustering are all necessary for making contract analysis easier for the user.
And these products need good domain experts to identify pain points where ml can help.
In ml as API side, one potentially valuable product suite I have seen is https://aylien.com/text-api/ . Even they are building additional products like news analysis.
But I am little sceptical on whether ml as API can scale in terms of addressing specific task-specific nuances needed for products building upon them.
I have seen some specialized image classifiers find a market in analyzing scientific instrument data in academia, outside of that literally nothing else.
But there sure are a lot of stories about adversarial bayesian multi-model deep net retention markov cycle consistent custom asic demystified robust perturbations.
The speech recognition and predictive text on my phone is still awful.
But there sure are a lot of stories about adversarial bayesian multi-model deep net retention markov cycle consistent custom asic demystified robust perturbations.
The speech recognition and predictive text on my phone is still awful.
My general rule of thumb is that any AI/ML is a result of it being a solution to a pain point within existing product(s). Not having a product with a necessary scale or require the technical sophistication of AI/ML but pitching it as if it has it is imo a red flag. This I call premature optimization.
Another point is marketing. A ride sharing service like Uber or Lyft could easily market themselves as "using AI/ML to solve last mile transportation". There's reason to believe they have some ML teams that do amazing work to support the ride sharing product. Both above have paying customers.
Another point is marketing. A ride sharing service like Uber or Lyft could easily market themselves as "using AI/ML to solve last mile transportation". There's reason to believe they have some ML teams that do amazing work to support the ride sharing product. Both above have paying customers.
A lot of compagnies do basically an externalisation of ml: they are providing the team the compagny would build if they had to do the job themselves.
Usually their website is buzzwordy but they are not necearily too much vocal because they find their customers through direct contact. And they do have customers... theres a real market
In healthcare there are not ao much killer application yet in ai/ml and its true that a lot are running on investor funds, but in my opinion this will change soon.
Usually their website is buzzwordy but they are not necearily too much vocal because they find their customers through direct contact. And they do have customers... theres a real market
In healthcare there are not ao much killer application yet in ai/ml and its true that a lot are running on investor funds, but in my opinion this will change soon.
At my company we try to sell a product that solves a customers need. Some of the things we’re doing as part of the solution we chose for the customers product could be called AI. But honestly, our customers don’t really care _how_ we solve their problem. They would be happy buying magic wands if that would solve their problem.
So sibling comments mentioned that some companies struggle with selling products not technologies. I agree.
So sibling comments mentioned that some companies struggle with selling products not technologies. I agree.
Facebook and Google use ML as a core part of their products? and they're some of the biggest companies in the world, so...
Right, but the ML they use is largely developed in-house. It's not part of a prepackaged ML "product" provided by an outside group.
My experience is that applying the ML to a domain is 99/100ths of the job; once you've got some traction with well known methods it's time to develop techniques that address their failings. But, any company that's done the hard yards of getting something working is definitely adding value. A good example is NLP, there's no rocket science in building embeddings, but any company that can build a specialist embedding for an important domain (medicine, engineering, finance) has an interesting and useful artefact that might be worth paying for (probably not as much as they would like though).
Right, and it's only reasonable to expect that if ML is a core differentiator for them, it wouldn't be provided by an outside group.
Customers (mostly businesses) are pretty much queueing to use our product, Datavoyant by Amplyfi (https://www.amplyfigroup.com/). We have real paying customers.
AI and ML techniques are great for both our product, as well for our marketing.
Yes they do, it's generally smoke and mirrors. But by being buzzword complient, party like it is 1999. Ai/ml is a cute way to fleece clients for 99.999 percent of the time. Ml/ai from the consumer side is a joke.
Uptake has real (paying) customers specifically for their ML based solutions.
Caterpillar (who initially was their biggest client and went as far as to buy a stake in the company) dumped them within the last year or two. Which doesn't exactly reassure me that they're creating real value.
Aktana has real big entreprise customers. They market themselves as AI while their algorithms are certainly not matching up to the buzz.
What kind of products were you're prior company offering before then? What kind of clients did they have?
Some of us not only have real customers, but are profitable also...
Most people that balk at the term A.I. seem to presume that intelligence is some capacity unique to humans.
Intelligence is merely pattern matching and goal oriented planning, exactly what machine learning is doing today.
Stop worrying and learn to love the bomb.
Intelligence is merely pattern matching and goal oriented planning, exactly what machine learning is doing today.
Stop worrying and learn to love the bomb.
While for the general population it might be true, I disagree about this statement refering to Hacker News. For devs I think the main issue with this buzzwords is that it's seen as overengineering for SMB.
"Intelligence is merely pattern matching and goal oriented planning" > ok you can talk about (parts) of cognition in that way, but it boils down to "we're all Turing machines" which is fine fine fine apart from the issue that what kind of turing machine doing what we do can fit into the soggy stuff in between our ears (as opposed to being able to model a conversation and yet the size of Jupiter with a run time of 2/3 the history of the universe). And there are parts that aren't covered by pattern matching and goal orientated planning such as goal generation and selection (accepting that if a pattern matcher is doing that -somehow - we have to explain how the pattern matcher "knows" to do it)
To be really precise when talking to potential customers, I call it artificial classification. Even artificial learning is a hazy term that doesn't really explain much. Classification has many real world examples that customers can relate to. "Intelligence" has too many connotations and isn't really useful in getting your point through when consulting people with non-technical domain expertise.
The term "Intelligence" does not have a definition everyone agrees on but it surely is not "pattern matching". Machine learning today does little more than pattern matching and it is far from intelligence at a human level.
> Machine learning today does little more than pattern matching and it is far from intelligence at a human level.
I'm rather less confident about this than you may be. When I'm doing something that I could plausibly claim requires "intelligence", then it certainly doesn't really feel like I'm just doing pattern matching. But I'm increasingly of the view that our perception of our own mental processes is a poor guide to their reality. The likelihood that my intelligence really is just large scale pattern matching, even though it doesn't subjectively feel like that, seems quite high to me.
I'm rather less confident about this than you may be. When I'm doing something that I could plausibly claim requires "intelligence", then it certainly doesn't really feel like I'm just doing pattern matching. But I'm increasingly of the view that our perception of our own mental processes is a poor guide to their reality. The likelihood that my intelligence really is just large scale pattern matching, even though it doesn't subjectively feel like that, seems quite high to me.
I’ve meditated a lot and come to the clear conclusion that this whole thing is just pattern matching and pattern generation. We’re not that special.
You've definitely bought the hype. There is no convincing definition of what (human) intelligence is.
All you've done is taken what machines do today and claimed that is what intelligence is.
All you've done is taken what machines do today and claimed that is what intelligence is.
I think the buzzword problem is rooted in sales mentality. Selling to VCs and selling to customers. Humans have to make decisions with incomplete information so we "pattern match" and "gut instinct" and "FOMO". So blame it on capitalism for systemically ensuring that entrepreneurs behave like this, or whatever, but frankly they are the only rational actors in this system.
Personally speaking, I left my own job earlier this year due in part to deep disagreements about how my previous company was utilizing and marketing itself as AI-based, even changing its name to include "AI".
But when I left fintech to work in the healthcare industry (totally different role), people started coming to me to pitch their products/solutions. I started getting pitched AI products (harness the power of AI & predictive analytics!), and the use of buzzwords was pretty outrageous. I started wondering...do these companies have any real customers? Who are buying these products, and are they seeing a sufficient-enough ROI to justify these purchases?