Home > Technical & Creative Skills > Tracking Campaigns

---

What kind of features you need for optimizer service ? (5)

---

Hello STMers,

I was a silent member of STM than years. My development team & I was working on an auto optimizer which works with API support of traffic sources & tracking software. I would like to ask you what kind of features you would like to see that kind of saas service?

I hope to open beta for STM users soon. Also if you would like to test the alpha stage and any recommendations for traffic sources & custom features you can always pm me.

Thanks.

---

Hi Wonnick! Sounds like a useful tool!

For traffic sources: Automatic cutting of publisher sites / placements according to a set of criteria, such as spend, profit/loss, ROI, and number of impressions.

For trackers: Automatic pausing of offers or landing pages when statistical significance is reached.

A great start would be to spy on your competition, i.e. other optimization tools, to see what features they have. But I'm sure you must have done that already.

Best of luck with your tool!



Amy

---

Markov chain Monte Carlo naive bayes automated ml, based on predictive conversion indicators such as time on site, click depth,
with progressive excluding / including of placements based on their probability to convert and spend.

The ad variations should be paused / restarted based on their probability also
(eg not allowing all to consume cpa at same time, but first allowing most probable and upon their failure activating less probable,
untill all tested)

The pre-indiciator snippet should be called inline; not on external domain.

---

Lots of great ideas here.

@johner911, could you elaborate on what you mean by this:

Thanks!

---

OK so we know that some events are linked to bigger likely hood of conversion...
or we hypothesise that they are..

Such events are:
- time on site
- page scroll event / distance / pattern
- field focus event (begins to type something in field)
- clickthru cta clicks
- multi-page steps completion events
(Ignoring here the technical specifics of the users device (browser fingerprint, IP etc..);
since these are classic segmentations and already solved; + they are not events.. they are specs..)

IF data is provided on those datapoints .. and linked to actual conversion counts,
its possible to assign a weight (indicating increased probability of conversion event happening) to each event,
a probability bias that if X event occurs, it increases chance of conversion completion by Y points.

Eg:

Time on site: 20-40 seconds
Page scroll: 2x page width
Clickthru click: true
field focus: true
Multi page steps: 3 steps within 20 seconds

Is more likely to convert than

time on site: 10seconds
page scroll: false
clickthru click: false..
,...

You can actually attach value to each of this event, akin to a eCPC,

So you end up killing bad placements .. with an ok probability (for affiliate purposes lol; its totally statistically invalid otherwise)
on 2 clicks worth of data per placement.. or in case of no cpc .. on 100-200 impression worth of data in the case of a typical display campaingn.

Also since .. its tied to a "route optimization" system, so it kills crap fast.. and then when it runs out of crap to kill (eg if everything is shit),
it revives things that are less crap than other.. to spend the 2-3x CPA and verify campaign.

---

Home > Technical & Creative Skills > Tracking Campaigns