Auto-Optimization Based on Offers' EPV (7)

Have you ever wanted to have a smart rotator sending traffic to your best path?

That's exactly what we're going to do in this tutorial!

First, let's define what we call the best path for a funnel such as the one below:



The best path, is the one giving us the highest revenue per view. If on average you earn $0.37 for each view of offer 1, while you earn $0.45 for each view of offer 3, you will want to show offer 3 to your visitors more often than offer 1. Granted you have reached statistical significance.

You can see the earning per view (EPV) by enabling the following heatmap:



And you can see how confident you are that an offer will bring more revenue than others by enabling the EPv Winners calculator:



In this example, we are only 49% confident that offer 3 will be more profitable than the other offers.

What we will do is take this EPv winner rate to automatically split the traffic to these offers.

In the example above, since we are about 26% confident that offer 1 and offer 2 will outperform the others, and about 49% confident that offer 3 will be the best performer, we will route 26% of traffic to offer 1, 26% to offer 2 and the rest to offer 3.

The more confident FunnelFlux's Bayes calculator is that an offer is going to outperform others, the more percentage of traffic it will send toward it.

So here are the steps to put this in place.

First, you have to split the traffic via a PHP node instead of a regular rotator. Let's call this PHP Node "EPV Weight Auto-Routing"



Here is the code that you will use in that PHP node:

<?php

require_once __DIR__ . './../admin/api/v2/toolbox.php';

$cacheExpirationInMinutes = 5;
$checkLastHours = '[[epv-auto-routing-lasthours]]';
$checkLastHours = empty($checkLastHours) ? 24 : $checkLastHours;

$cacheKey = '{node-id}-{trafficsource-id}';
$cache = Cache::getInstance( Cache::TYPE_FILE );
$nodeWeights = $cache->get( $cacheKey );
if( $nodeWeights === Cache::NOT_FOUND )
{
  $nodeWeights = [];
  $nodeConnections = DBTableCampaignFunnelConnections::getNodeExitConnections( '{node-id}' );
  if( !empty( $nodeConnections ) )
  {
    $nodeIds = [];
    foreach( $nodeConnections as $connection )
    {
      $targetNodeId = $connection->getTargetNodeId();
      $nodeIds[] = $targetNodeId;
      $nodeWeights[ $targetNodeId ] = -1;
    }

    $to = time();
    $from = $to - ($checkLastHours * 60 * 60);

    $to = (new DateTime())->setTimestamp( $to );
    $from = (new DateTime())->setTimestamp( $from );

    $drilldownRequest = [
      'timeRange' => [ 
        'start' => [ 
          'date' => [ 'year' => $from->format('Y'), 'month' => $from->format('m'), 'day' => $from->format('d') ], 
          'time' => [ 'hour' => $from->format('H'), 'minutes' => $from->format('i') ] 
        ],
        'end' => [ 
          'date' => [ 'year' => $to->format('Y'), 'month' => $to->format('m'), 'day' => $to->format('d') ], 
          'time' => [ 'hour' => $to->format('H'), 'minutes' => $to->format('i') ] 
        ],
      ],
      'timeZone'  => [ 'name' => 'UTC' ],
      'options'   => [ 
        'idFunnelFilter' => '{funnel-id}', 
        'idTrafficSourceFilter' => '{trafficsource-id}',
        'computeEPVConfidenceRate' => true,
        'confidenceRateIncludeAll' => true
      ],
      'groupings' => [
        [ 'groupBy' => 'Element: Node ID', 'whitelistFilters' => $nodeIds ]
      ]
    ];

    $report = \FluxAPI\v2\Toolbox::statsDrilldownRequest( $drilldownRequest );
    $nodesInReport = 0;
    foreach( $report['rows'] as $row )
    {
      $nodeId = $row['cells'][0]['raw'];
      $epvWinnerRate = $row['epvConfidenceRate']['rate'];
      if( $epvWinnerRate != -1 )
      {
        $nodeWeights[ $nodeId ] = $epvWinnerRate;
        $nodesInReport++;
      }
    }

    // If some nodes have no traffic yet, we need to re-scale the weights of the nodes that did get traffic
    if( $nodesInReport < count($nodeIds) )
    {
      $ratio = $nodesInReport / count($nodeIds);
      $assignedWeight = 0;

      foreach( $nodeWeights as $k => $w )
      {
        if( $nodeWeights[ $k ] != -1 )
        {
          $nodeWeights[ $k ] = $w * $ratio;
          $assignedWeight += $nodeWeights[ $k ];
        }
      }

      $missingWeight = 1.0 - $assignedWeight;
      $missingNodesCount = count($nodeIds) - $nodesInReport;
      if( $missingNodesCount > 0 )
      {
        $missingWeightPerNode = $missingWeight / $missingNodesCount;

        foreach( $nodeIds as $nodeId )
        {
          if( $nodeWeights[ $nodeId ] == -1 )
          {
            $nodeWeights[ $nodeId ] = $missingWeightPerNode;
          }
        }
      }
    }

    $nodeWeights = array_values( $nodeWeights );
  }

  $cache->set( $cacheKey, $nodeWeights, $cacheExpirationInMinutes * 60 );
}

$choice = mt_rand() / mt_getrandmax();
if( empty($nodeWeights) )
{
  $iRoute = -1;
}
else
{
  foreach( $nodeWeights as $iRoute => $weight )
  {
    if( $choice < $weight )
    {
      break;
    }

    $choice -= $weight;
  }
}

return( $iRoute + 1 );

?>

• The connections going out from this PHP node must be sequential and start with "On Done 1". Do not skip numbers. "On Done 1", then "On Done 3" would not be valid.
• There is some added caching to avoid fetching stats from the API at each visit. By default, it will check the stats and re-calculate the path weights every 5 minutes. You can change this timing by editing this line:
  $cacheExpirationInMinutes = 5;
• You can put any node type after this PHP Node, it doesn't have to be only offers. If you want to place landers after, or a mix of node types, it's perfectly fine. It will still be able to calculate the best performing paths.

That's pretty neat!

It could have saved me a lot of money when I wasn't around for the weekend and an offers suddenly stopped converting.

awesome - thanks!

You Sir are quick, looking forward to put this into practice. Sending you good energy to Keep Making Badass Developmental Progress!

Isn't it amazing how quickly you can make stuff like this happen with a small, dedicated team lead by an absolute genius developer?

Vitavee you're always full of surprises! How long did this take, like 2 days? My god.

Also, worth noting that there are some cool flexibilities here:

• You can specify the time range on which it analyses stats - so you have decent transparency there about how it calculates things
• It's based on the nodes the PHP node connects to, so could be anything.
• You can use these many times in a funnel so could separately optimise lander and offer nodes
• There is a "cacheKey" parameter that lets you specify the segmentation it does for optimisation. By default it is $cacheKey = '{node-id}-{trafficsource-id}'. You could change this to e.g. $cacheKey = '{node-id}-{country}-{trafficsource-id}' to treat different countries separately, and likewise for tracking fields something like $cacheKey = '{node-id}-{trackingfield-agegroup}-{trafficsource-id}' to further split optimisation based on data you are passing from ads. If this data is N/A since its not present or known, it also won't break the node or tracking (lol good to know!).

Originally Posted by zeno

Vitavee you're always full of surprises! How long did this take, like 2 days?

      'groupings' => [
        [ 'groupBy' => 'Element: Node ID', 'whitelistFilters' => $nodeIds ]
      ]

      'groupings' => [
        [ 'groupBy' => 'Element: Node ID', 'whitelistFilters' => $nodeIds ],
        [ 'groupBy' => 'Location: Country Code', 'whitelistFilters' => [ '{location-countrycode}' ] ]
      ]

Ahh, right. Have updated our documentation to suit. That's still pretty easy to do since its just a basic reporting API call in the code and subsequent change to cacheKey.