Estibot (old version) to be taken offline in a week

[Bob Hawkes]

For those that might not know, the regular version of Estibot is being taken down in a week, and only the beta version will be available. Here is the statement:

IMPORTANT NOTICE: EstiBot will be switching to the new interface (currently accessible to all users at https://beta.estibot.com) on 20180221. We strongly encourage all customers to start using the new interface and submit bug reports and suggestions for improvements.

I give this heads up as the information provided is somewhat different, so some might want to do valuations before the existing one disappears.

I am not sure if anyone from Estibot is on NPs, and may want to add information. I certainly hope the problem of the beta saying extensions are available when they are not (in certain names, not just a case or two) will be resolved, as it causes unrealistically low valuations in some cases.

ps Not sure which was the best thread, so hope that an administrator will move this if this is the wrong place.

 

[OmarVG]

I don't use Estibot anyways... I have my own Auto-Appraisal Machine AKA ME!

 

[equity78]

Interesting, thanks for posting here.

 

[Jv1999]

I was relieved and happy when I thought this meant estibot was being taken down for good. Instead, there's the beta version still :/. Siiigh.

U guys know estibot basically includes the domain prices that people list as BIN or Make offer for "related" domains, right? I could put one of my very crappy domains that's related, up for $1,000,000... and estibot would take that into account!

 

[Bob Hawkes]

"U guys know estibot basically includes the domain prices that people list as BIN or Make offer for "related" domains, right? I could put one of my very crappy domains that's related, up for $1,000,000... and estibot would take that into account!"​

I agree it may be true @Jv1999 , but do you have evidence for your statement that listed prices influence Estibot? Anyone from Estibot on NPs wish to comment?

From the Estibot manual that is available at this link there is no mention of that as one of the 30 variables.

https://www.estibot.com/guide.html

"EstiBot valuates domains based on over thirty internal and external domain attributes.

Internal attributes include such things as domain length, extension, word count and other characteristics directly linked to a specific domain. External attributes refer to third party data such as a domain's search popularity or type-in rank.

The characteristics of a specific domain name are then compared to those of previously sold domain names and the valuation is based on that comparison.

More than anything, EstiBot is here to help you valuate your domain name. It will provide you with an array of information and a ballpark valuation to give you a starting point for your own, critical appraisal of the domain name.

While EstiBot often hits the right ballpark, some estimates may be quite a bit off. Statistically, on an average, EstiBot is quite accurate, but individual appraisals can vary. An automatic valuator will never be able to account for everything, because a domain's value is not a scientific absolute, but depends on countless variables, the most important being the motivation of both the seller and the buyer."

 

[RobM]

I've never seen it hit the right ballpark *before* a sale. Basically the last sentence translates as 'we can be right or we can be wrong there is no way of knowing.' Um...so why use it or give it credibility? Someone on this forum has posted 7 pages of bumps for a domain estibot values at three figures. Buyers know this is rubbish. The only person this random number generator appeals to is a (naive) seller.

 

[Bob Hawkes]

ICYMI Andrew Allemann in December 2017 posted on DomainNameWire re use of automated domain evaluation tools. It starts and ends with the quotes below, but I urge those open to different viewpoints to read the entire post that provides sound advice on different ways to use automated evaluation.

"Automated domain appraisals aren’t perfect but smart domainers can use them to their advantage."

"I firmly believe that automated domain name appraisals have value. Are they spot on? Absolutely not. Are they built on great data that can inform your buying and selling decisions? Absolutely."​

Read Andrew's entire post here: https://domainnamewire.com/2017/12/04/use-automated-domain-appraisals/

Can GoValue and Estibot be hugely better? Definitely!

Should we totally ignore them because they are not as good as they could be? Not in my opinion.

Artificial Intelligence is growing in leaps and bounds, but already it can outperform doctors on many complex medical tasks, situations that in my mind are more challenging than domain name valuation. IBM Watson significantly outperforms skilled physicians in lung cancer diagnosis, a Chinese AI implementation recently passed a medical school exam through deep learning, and a team from Google last month issued a preprint showing that AI - deep learning with direct input of emergency room notes and health measures could predict outcomes, including probability of death and length of hospital stay, better than existing methodologies.

Are these things easier than evaluating the worth of a domain name? An automated tool can in microseconds consult a huge database of prior sales, consult every dictionary in the world for the words, look at news sources for changes in the frequency of the word, examine what advertisers are offering to pay for related search terms, how often are Google searches made on the term and how is that changing, how many relevant extensions are available for registration if any, what similar names are available for sale, etc. As the domain tools themselves say, this is a starting point for your evaluation, not a precise value to be blindly used.

We ignore applying AI to domain name valuation at our peril.

 

[RobM]

No we don't at all. We *listen* to them at our peril. I have experience in coding genetic algorithms and neural networks and they can't simply be applied to everything with the intelligent sounding phrase 'Artificial Intelligence is growing in leaps and bounds'. AI can *never* work out a formula or pattern because there simply isn't one. Simply looking through data of previous sales is *not* Artificial Intelligence or even very clever. Try applying one (as I have for many years) to n-queens solutions and you will see that not everything can be solved that way. Go and look up 'expert system'. That is what you are doing (badly) - zero artificial intelligence involved. As long as all buyers and sellers behave differently there will never be anything at all for them to learn from even if you put in data for every sale from now forever.

If you need a tool as a 'starting point' for your evaluation and, by your own admission, that tool is as correct as often as chance what you're saying to people is that they would be absolutely no better off using or paying for a tool/certificate. Listen people - if you need something to TELL you what your domain is worth you shouldn't be selling one. If you don't know what a domain is worth *to you* you shouldn't be registering one letalone selling it. If you are a buyer and you need a tool to TELL you how much you should pay then you seriously should not be in charge of the money to spend. When seller expectation meets buyer expectation a sale occurs. That's it. You provide info as the seller and the buyer adds in his motivation. There is no magic 'formula'.

Your logic is seriously flawed. I pity anyone that loses a single cent on any tool/certificate advice or paying the people behind them.

 

[Bob Hawkes]

This may surprise you (and others) @RobM but I agree with much more than I disagree with in your last post on this thread.

I agree that even seemingly simple tasks are hard to automate. I spent a good chunk of 10 years of research trying to automate systems that were as good as humans in perceiving certain kinds of transient linear (but changing intensity) patterns in a noisy background. It took many years, and the diligent efforts of various scientists and programmers way brighter than me, to get it about 90% as good as a trained human. But once it was perfected, it now is the base of a system that can automatically, inexpensively observe and analyze in a way no human could. But that was way simpler than domain names.

I used the medical examples, that were not algorithmic implementations, but rather used deep learning AI to seek patterns, since I think that is closer to the domain valuation case. And just to make clear, I am not saying current Estibot or GoValue are deep learning algorithms, just that they or a competitor could be.

I also agree with your important point that why automated will never predict perfectly or even near perfectly is because of the human dimension as you say different "buyers and sellers behave differently". This year's Nobel Prize in Economics to Thaler for Nudge theory and his work in general on behavioural economics was recognizing that in day to day economic decisions people don't make logically rational decisions, and the same is true for domain sales. I agree that an AI based system will never be perfect because of this (although disagree with the view that we should never use them at all therefore).

But a human predicting a reasonable price has exactly the same problems, and in fact much more so because of our biases and incomplete knowledge bases. I won't pick on anyone in particular but it is incredibly easy to find on NPs discussions people who claim to be experienced who say with complete confidence a domain name is worthless, and then the name sells for many thousands within months. Not all purchasers, but some, will find comfort in knowing that the price they are considering paying is not out of line with values suggested by a tool that is not part of the transaction. I suspect that is why NameJet include Estibot valuations in their auction listings, and GoDaddy now GoValue.

I am not sure what "by your own admission that tool is as correct as often as chance" - I don't recall saying that (maybe my senior brain is having a memory lapse!). I certainly don't believe that! I have been critical of Estibot and GoValue that could be made so much better. They make big mistakes sometimes. But I definitely don't believe they are no better than chance!

But if you feel Estibot is totally useless, which you have clearly expressed on numerous posts over several threads, why bother wasting your time even reading and replying to threads with it in the title? It is a serious question, and I am not trying to be snide. Time is precious, if we feel no argument could ever change our view on something because we are sure we are right, surely it is best is to simply ignore it?

Anyway, thank you for your contributions, and have a good day! Your points are valuable reminders that one limitation on automated (and human) appraisal is the human dimensions of both buyer and seller. And on that we agree!

 

[RobM]

I answer because I think it is important that new entrants to the business see appraisal tools for what they are and don't spend a single penny on them. If more people had been more vocal a few years ago (I remember when estibot first appeared and how they were received) maybe this kind of parasitic activity wouldn't be happening. This is no different to the email appraisal scam (which we all claim to hate) - getting an air of credibility doesn't change it's substance. All appraisal tools and certificates (free and paid) have done is give sellers false hope whilst making it harder for sales to occur as sellers sit until they go broke on a domain that they have been told is worth far outside of reality. They are just a blight on the industry. This is why NO appraisal tools have ever been responsible for a sale price. Anyway point taken - I won't mention it again I think people have got the idea. If they are stupid enough to give people money to tell them what their own domain is worth they're probably not going to listen.

 

[Bob Hawkes]

Re the original post, I see today that Estibot have delayed by a week taking down the 'old' Estibot. It is now scheduled on Feb 28.

Also, I note that the 'new' Estibot now has a button near top with a link if the domain name you searched is currently for sale on Afternic.

I pointed out to Estibot some problems the beta version has saying that domain name extensions are available when they are not. They provided a couple of responses, not too specific, but did point out that they go by whether a domain name is in the zone, and not whether it is unregistered. That does not explain all the discrepancies, however.

 

[Bob Hawkes]

On the general question of automated appraisals, there is a similar debate happening in the real estate business, driven by Zestimate tool by Zillow in the US. Here is an article (from UPenn) that includes some points with relevance to the Estibot/GoValue etc. debate as well (note however the involvement of a Zillow related participant, so it is not really a neutral article).

In particular I note:

• Zillow do tests of the accuracy of their estimates. They claim that initially it was about 14%, but now is about 4%. I think that it would be great if similar estimates of accuracy were done in automated domain name valuation, e.g. using Namebio data, on how precisely Estibot and GoValue predict (but somehow before any chance they adjust the estimate to reflect the price).
• One role of appraisals (both actual person ones and automated) is for others involved in the approval - those providing mortgage funds in the real estate business (or sometimes taxation authorities). I think that there is a similar case with some domain names. Sometimes a business needs to convince its CEO or board that a domain price is reasonable, and automated estimates can be helpful (as can of course those of an uninvolved domain name broker with experience in the extension and topic).
• I am perhaps extrapolating what was said in the article, but really automated appraisals realistically provide a valuation range and should indicate that. I think that GoValue, Estibot and the like should give their estimates as a range, that incorporates the uncertainty in the value, rather than a precise value. GoValue in particular should not state values down to the dollar (it goes against all my science researcher significant figures instincts!). i.e. it would be more correct to give an estimate as $300-$800 rather than saying $451.

http://knowledge.wharton.upenn.edu/article/home-appraisers/