I was really interested in your product.

Unfortunately your website was less than informative. Any chance you could provide some more details around your approach and methodology?

You're right, the landing pages are basically just teasers to get people interested. It's very much in beta.

Happy to provide more details. The platform works by tracking the visits to your site and determining where they came from (via referrer + url params / utm tags). Then when an event is fired, I look back at the history of that user's visits and run attribution models to distribute credit for the event back to the sources.

Technically, I'm using a few rails sites, sidekiq workers, redis and postgres to keep it all humming. Not much fancy on the front-end, just plain ol javascript.

Is there something specific you're interested in?

Edit: feel free to mail me if you want more personalize info / discussion: hello @ attribution.io (added spaces for spam robots)

Are you doing anything with dynamic/data driven models? Or do you just focus on the same static ones as GA plus your static content models?

Also, how do you handle view through and other harder to measure branding touch points? Do you integrate with things like DCM?

My beef with most non-enterprise solutions (which I'd classify yours as based on pricing) is they are very limited, and typically don't do much of anything beyond what is available for free in GA.

Fwiw I'm a senior digital media guy who goes deep on attribution. It is what keeps me up at night as we work on display,video and paid social. The fact that we have a longer sales cycle due to a free trial further complicates things.

All of my models are 'data-driven' in that they are computed directly from the visit & event data. I don't currently have models that change based on the data, but I could easily add them if you're interested in helping me define how they work.

I find one of the big problems with GA is that it doesn't show you the actual people that are included in the final attributed credit per channel/campaign, does it? Without this, it's quite a leap of faith.

I'd love to chat with you about your current practices; I still think there's still tons of room for improvement past the current models. For example, I'm experimenting with a model that gives credit to pages in addition to traffic source to help justify the cost of content production / optimization of which content is working and which is not.

I'm using "data-driven" synonymously with "dynamic" in this case, which is indeed the changing model based on the data using machine learning.

I haven't played with it to this extent, but can't you get a GUID of some sort using the GA API[1] or custom dimensions? I haven't tried it personally and I know they restrict some of the data so this may not be possible.

I do like that you are factoring in content into this. The regular way of doing that is to create some custom channel groupings in GA (probably some other ways) but content really deserves a place in the funnel when looking at attribution. And you are right that content marketing is hardly free and has a high cost associated with it that is hard to prove ROI on.

Definitely would be open to chatting. I work in Mountain View if you're in the area and that would be easier.

[1] https://developers.google.com/analytics/devguides/reporting/...

I do have a few questions...

Is there a way to do TV attribution beyond multivariate statistical learning models attempting to gauge the impact on other channels based on broadcast period? (i.e. treat the offline channel as a categorical variable)

What is "dynamic" attribution modelling? Does this mean the attribution model changes depending on who you perceive the customer to be?

What is the best way to do multi-device tracking if you don't have a login?

Check out what Adometry and other dedicated attribution providers are doing if you are interested in TV. You need a hefty budget and volume to use them though.

Basically it is a combination of art and science with the statistical modeling being more on the art side, and then the science being tagging the ever-loving crap out of everything you can through a single platform to record as many touchpoints as possible.

TV is increasingly becoming measurable through things like Hulu, etc. You can also do vanity URLs for direct response TV ads.

Dynamic attribution modeling uses machine learning to continually evaluate the effectiveness of touch points, and then changes those weights over time. There are soooo many variables to factor in (decay rate, creative, placement, etc.) and every user has their own unique conversion path where each touch point might have a different level of impact. It is in theory the holy grail of marketing, but totally unique to each company and VERY hard to implement decently.

Multi-device tracking has come a long way thanks to lots of data partnerships, but honestly a lot of that feels icky to me. They are effective though. A lot of them rely on cookies and using data brokers and cookie onboarding services to have cookies continually re-added and matched across various devices. So you may not login on my site, but if you logged-in to another site that partners with a data broker (or is Google/FB who do this on their own), they know who you are, and can then map that data to existing audience data records.

I'm not who you asked, but maybe I can help with my viewpoint.

TV/Radio/Offline marketing can only be attributed via these statistical models (eg in the zip codes where the ads ran, did a bump in traffic or store visits happen). Although old fashioned, I think coupons per channel are a great way to get confirmation of the link between ad and purchase, it's less precise than digital tracking, but at least makes some concrete connections.

Multi-device tracking without a login can only be done by using a huge network of 3rd party data. E.g. if xyz company has cookies on everyone and they see a login on one site, then they equate that cookie as the same user across devices, effectively sharing the information with their clients. I think this is very shady and I do not use any data like this because I think it violates people expectation of privacy across interactions.

Have you tried correlating visits from different devices from the same IP? E.g. if you get 2 devices from 3 St John St within an hour, then the same 2 from The IBM Building again within an hour, you can make a fair assumption that's the same person (you need the second site to isolate family members).

I've never implemented this - never had the time nor clients who wanted it enough - but have always wondered whether it might work. Maybe later this month. In declarative SQL, naturally.