My first failed software product was in about 2003 or so.  I had been developing and maintaining my own compiler language for over a decade, since about 1985.  It had some unique feautures and I had made some money selling/using it pre-internet.  Now with the internet putting me before a world-wide audience, perhaps i could make a go of it. 

For about 2 years, I tried selling over the internet and had zero sales.  I ended up killing the product after attending a conference of about 300 software developers.  Among all the people I met, there was only one person who I thought was a suitable match for my language ... and he was selling his own language product!  I realized that the total world-wide market for this product could be me.

I pulled the product shortly after the conference and set about developing something new.  At the time, spam was emerging as a real bother and SpamAssasin was the premier solution.  At that time, I switched from using my proprietary compiler to using Python for most of my development.  I developed a SpamAssasin inspired solution but one not based on Bayesian inferencing, but, rather, on rules entered by the user as regular expressions. 

My product, SpamAI, enjoyed modest success but, unfortunately, many others thought anti-spam was an emerging field and the competition numbered in the hundreds.  After a good run of a few years, I pulled the product rather than go through the support costs of getting a new user running.  I still use it myself on a daily basis but have had the program off the market for years now.  Indeed, I let the domain registration lapse and the domain is in the hands of someone who is trying to sell it.

SpamAI is not a failed product.  It was successful but I pulled it off the market when the cost of supporting it exceeded any possible income from it.

For one product, DAIRGram Server, I tried changing my development methodology.  Rather than develop the specs myself, I would work hand-in-hand with a marketing partner who would guide the development.  Indeed, there were two partner, one after the other.  The partnerships did not last past the development stage but the product design benefited by by having several sets of eyes looking at the product over many months. 

Both partners had real needs but did not have the money to fund development.  It turned out that this turned out a good predictor for the market.  I had a few nibbles from people with needs and very little money.  Looking back, I don't think I could have predicted that.  One of the partnerships was absolutely dependent on both of us working as a team. The program was of no real value without him and his plans would never go anywhere without my software.

The DAIRGram Server product was meant to install on a user's server or could be run as a service by me.  I think that part of the problem with the product was, indeed, the service model.  I think the customer I was able to reach wanted desktop software they bought once.  They did not want to pay recurring charges for a service.

I did another server program, Syncation Server.  The idea was that it was a shared in-out board with (eventual) geo-location features.  There was simply no interest in a server program.  As part of the in-out board, I developed a telephone message taking component.  I have since extracted that component at TelephoneMessagePad.com and am selling it now as a client-side only component.  The server program failed but the extracted client program is doing fine.

The program which just failed is a website log file utility.  With millions of web sites and robotic visitors accounting for about half of all traffic, you would think there is a need to do a good job of separating which visitors are robotic from human?  So did I but many weeks of advertising have convinced me otherwise.

First, the number of searches for log analysis topics in a month on Google is in the few hundreds to a few thousand.  There is no tens or hundreds of searches per month term.  This volume is low for the markets I have competed in.  At the same time, the competition in this market is 10-15.  This is a saturated market and competition is high.

Many people satisfy their analysis needs by using Google Analytics.  A fraction now are using log analysis.  I know that before I brought my program to market.  Much of the reason I developed it was to meet my own internal needs.  So, I guessed that that market was not as large as the "millions of websites" might imply.

I was not prepared for having my ads ignored.  Even when my ad was delivered for relevant search terms, out of 1,500-2000 ad impressions spread over several weeks, I only got 3 click-throughs.  This is simply not high enough to run a business and Googles algorithms tend to drive up the cost of low click-though accounts and campaigns.  So, I stopped advertising the log analysis program.

I had plenty of advance warnings that there was low interest in log analysis.  What took me by surprise that I could not elicit interest even among those searching for log analysis topics.

I am working on a new product which early results say there is considerable interest.  That would be nice if that early promise delivers a success.