A friend and I built a similar service back in 2006-2007. It was actually a rev share program based on affiliate aggregators like LinkShare and Commission Junction -- that's how we populated our product list. A quick look at some of the sample vendors mentioned and the affiliate programs they use, I suspect you are doing something similar.
Problems we ran into included: trying to normalize the product data from the vendors into a format usable to make actionable and accurate ads--the aggregators practiced very little data hygiene and did not enforce much consistency, oddly, some vendors couldn't even spell the names of their products correctly which made for some pretty crappy looking ads; at one point we had over 4 million products listed, primarily because we started importing Amazon products and vendors, so the product selection was difficult, some things have general appeal and would show up all the time and others wouldn't match much content at all; we had trouble signing up sites to display the ads, CPA was relatively new at the time for the scale we were going for, we might have been better off selling our tech to the top affiliate aggregators as a better, easier method for sites to integrate affiliate program membership; reporting was a real bitch because of the delay from click through to actual purchase to the payment showing up, which depending on the affiliate program could be upwards of 90 days.
In retrospect, we would have better off setting it up for passive income and not try to do active sales to integrator websites and instead shored up the pitch copy and made it largely self service for the integrators (it mostly was already, but I spent a lot of time tweaking the matching algorithm rather than on the sales/pitch side), and pitched it as an easier way to join affiliate networks. That would have set expectations a little better than calling it an "ad network" like AdSense, which we also considered our main competitor.
We are built on top of the feeds from the affiliate networks - and yes, there are issues with some of the data :-)
We have about 9 million products in our system (although not all are active) and we regularly inspect it and blacklist products (particularly adult products).
The system also tracks which products attract clicks and more importantly convert well, and if they aren't performing, they are disabled. So crappy ads don't get shown, once the system has noticed that they are crappy.
Unfortunately, we never reached the scale where we could exercise product blacklisting, and even 50k products were difficult to vet manually. And I had written significant code to help seek and destroy affiliate feed data problems which helped a lot but still ate up a lot of time.
The major data hygiene issue was that the affiliate networks tried to pigeon hole every product into the same schema. I think this was largely due to their major users being non-technical and their feed format being CSV based (so people could manipulate it in Excel)--this was the only time in my life that I wished for an XML set up, one that had different schemas for different products (this is why Amazon was attractive, they do this, but it's still not 100%. While Amazon has a large product category tree, a lot stuff still ends up being pigeon holed, especially if it's a specialty product). The worst part was the "extra" fields that the vendor could use for anything, some used them for keywords, others for deeper hierarchical categorization. I think my favorite was the "category" fields which were chevron delimited phrases within a single CSV field.
Issues? I find that Commission Junction affiliates can't even get their links right all the time which is basically their entire business. Talk about a "shotgun" approach...
Thanks for the advice, we are doing something similar yeah but we haven't normalised the products so much, we care more about that specific product from that retailer.
We are currently talking to some of the networks and affiliates about using our technology directly as well, as they seem to like our matching
Problems we ran into included: trying to normalize the product data from the vendors into a format usable to make actionable and accurate ads--the aggregators practiced very little data hygiene and did not enforce much consistency, oddly, some vendors couldn't even spell the names of their products correctly which made for some pretty crappy looking ads; at one point we had over 4 million products listed, primarily because we started importing Amazon products and vendors, so the product selection was difficult, some things have general appeal and would show up all the time and others wouldn't match much content at all; we had trouble signing up sites to display the ads, CPA was relatively new at the time for the scale we were going for, we might have been better off selling our tech to the top affiliate aggregators as a better, easier method for sites to integrate affiliate program membership; reporting was a real bitch because of the delay from click through to actual purchase to the payment showing up, which depending on the affiliate program could be upwards of 90 days.
In retrospect, we would have better off setting it up for passive income and not try to do active sales to integrator websites and instead shored up the pitch copy and made it largely self service for the integrators (it mostly was already, but I spent a lot of time tweaking the matching algorithm rather than on the sales/pitch side), and pitched it as an easier way to join affiliate networks. That would have set expectations a little better than calling it an "ad network" like AdSense, which we also considered our main competitor.