Originally published along with a blog post explaining cohort analysis for eCommerce here: http://www.decalmarketing.com/page/The_simplest_explanation_of_cohort_analysis_for_determining_lifetime_customer_value_in_an_eCommerce_business

PHP code for doing cohort analysis on an eCommerce website

<?php
    /****
    * This script presents an "algorithmic" way of doing a cohort analysis
    * on your eCommerce customer database.
    *
    * This can be useful if you want to automate a lifetime value calculation
    * based on some data available from a CSV file, database or API.
    *
    * The specifics of how your eCommerce system stores this information is
    * abstracted out to a couple of objects with methods such as 
    * "firstSuccessfulOrder".
    * 
    * For your reference, 2 abstract class definitions appear at the end
    * of this file that describe the methods and return values required
    * by the Person and Order class.
    *
    * Also, where variables represent these "abstraction points", I have noted
    * it in the source code. That way, you can relatively simply implement
    * this for your own eCommerce system by just creating objects that
    * implement these methods. The bulk of the work here is done in 
    * "plain old procedural" PHP. Even though in a lot of cases I find it
    * much simpler to work with objects (because keeping track of lots of 
    * arrays and indexes and whatnot can be a strain on the brain!) I wanted
    * to just have this laid out as "purely" as possible so you can easily see
    * the "algorithm" behind it.
    * 
    * This code was first published on the Decal Marketing blog at:
    * 
    * http://www.decalmarketing.com/blog/
    *
    * and written by Iain Dooley in February 2014.
    *
    * You're welcome to re-use or redistribute this code however you see fit
    * however I'd appreciate it if you could leave this message in-tact 
    * at the top. Any questions, please don't hesitate to email me:
    *
    * iain@workingsoftware.com.au
    * 
    * Enjoy!
    */

    //First, place each customer into a cohort based on month of first purchase
    $cohorts = array();
    
    //Here, we assume that we have an array of Person objects that has
    //been generated by some method specific to your eCommerce system.
    //Each Person object will have some methods, which have been noted
    //in the comments below. In order to re-use the code below *verbatim*
    //all you should need to do is implement the Person class and Order
    //class as described below, then put them into an array called 
    //$people anywhere above the following line of code.
    array_filter($people,function($person) use(&$cohorts)
    {
        //The Person object needs to implement a method:
        //
        //Person::firstSuccessfulOrder()
        //
        //This method should return an object of class Order
        //which has methods as described below
        if(is_object($person->firstSuccessfulOrder()))
        {
            //The return value of firstSuccessfulOrder
            //is an object which has the method orderDate() on it,
            //for example you might return an object of class
            //Order which has a method orderDate(). The orderDate()
            //method needs to return a date/time string in a format
            //that can be correctly parsed by php.net/strtotime
            //
            //NB: we use a unix timestamp for the purchase month
            //in the array index so we can sort based on 
            //array key
            $purchase_month = strtotime(date('Y-m-01',strtotime($person->firstSuccessfulOrder()->orderDate())));
            
            if(!isset($cohorts[$purchase_month]))
                $cohorts[$purchase_month] = array();
            
            $cohorts[$purchase_month][] = $person;
        }
    });

    //Get into descending order
    krsort($cohorts);

    //Now for each cohort create an array with one element per month since the
    //month of first purchase, and find the average purchase in that month for
    //the given cohort
    $avg_month_purchase_by_cohort = array();
    $max_months = 0;

    array_walk($cohorts,function($cohort,$purchase_month) use(&$max_months,&$avg_month_purchase_by_cohort)
    {
        //Start by finding the number of months since first purchase
        $first_purchase_month_start = date('Y-m-01 00:00:00',$purchase_month);
        $date_format = 'Y-m-d H:i:s';
        $diff = date_diff(date_create_from_format($date_format, $first_purchase_month_start), date_create());
        //The date_diff months value here resets as the year ticks over, so need to
        //add years * 12 to the month value in order to work for more than 1 year ago
        $months = $diff->m + ($diff->y * 12);
            
        //Now define our array to hold averages per month, noting
        //that our months will start from 0
        $current_row = range(0,$months);
        
        //If this is the oldest cohort, record that this is the maximum number of months we have
        //history for. Later we will use this to "pad out" incomplete cohorts with averages
        //based on past purchase data
        if(count($current_row) > $max_months)
            $max_months = count($current_row);

        array_walk($current_row,function(&$average,$index) use($cohort)
        {
            $current_total = 0;
            
            foreach($cohort as $person)
                //Find total purchases in this month number.
                //The Person class will need to implement the
                //method totalPurchasesInMonth() such that it 
                //accepts the number of months since first purchase
                //as a sort of "offset"
                $current_total += $person->totalPurchasesInMonth($index);
            
            //Average is the total purchases in month $index
            //divided by number of people in this cohort
            $average = $current_total/count($cohort);
        });
        
        //We're building a "matrix" where each row represents a cohort
        //and each column represents the number of months since first purchase.
        //Each "cell" is therefore the average amount spent by people in that
        //cohort in the given month number.
        $avg_month_purchase_by_cohort[date('Y-m',$purchase_month)] = $current_row;
    });

    //Now we have a "matrix" but the right hand end is "jagged", that is, for  those
    //who first purchased 2 months ago, we don't have a value for any months
    //beyond the 2nd month, so fill up any "empty" cells with the average for
    //that month number across all cohorts for which we have data in that month number
    array_walk($avg_month_purchase_by_cohort,function(&$months,$cohort) use($max_months,&$avg_month_purchase_by_cohort)
    {
        //If there are months for which we have data that aren't recorded for this
        //cohort
        if(count($months) < $max_months)
        {
            //Get a range of months for which we don't have data
            $range = range(count($months),$max_months-1);
            //Create a copy of our array so we can loop over it below
            //without modifying the array pointer
            $averages_orig = $avg_month_purchase_by_cohort;
            //Now loop over all cohorts for each month number for which 
            //we don't have data, and enter in the average we have for
            //all cohorts that do have sales data for this month number
            array_filter($range,function($index) use($averages_orig,&$avg_month_purchase_by_cohort,$cohort)
            {
                $cohort_totals = array();
                
                array_walk($averages_orig,function($months) use(&$cohort_totals,$index)
                {
                    if(isset($months[$index]))
                        $cohort_totals[] = $months[$index];
                });
                
                if(count($cohort_totals))
                    $cohort_average_for_month = array_sum($cohort_totals)/count($cohort_totals);
                else
                    $cohort_average_for_month = 0;
                
                //This is the bit where we set the value for future months with
                //the average based on past months
                $avg_month_purchase_by_cohort[$cohort][$index] = $cohort_average_for_month;
            });
        }
    });
    
    //Format everything into a CSV file so we can easily open it in a spreadsheet application
    //to make a pretty graph or sum totals etc.
    $header = array('Cohort');
    
    array_walk(range(1,$max_months),function($i) use(&$header)
    {
        $header[] = 'Month '.$i;
    });
    
    $header[] = '# in Cohort';
    $out = array();
    $out[] = '"'.implode('","',$header).'"';
    
    array_walk($avg_month_purchase_by_cohort,function($months,$cohort) use(&$out,$cohorts)
    {
        array_unshift($months,$cohort);
        $months[] = count($cohorts[strtotime($cohort.'-01')]);
        $out[] = '"'.implode('","',$months).'"';
    });

    print implode(PHP_EOL,$out);

    /***
    * Below are interfaces for Person and Order, describing what
    * methods need to be implemented, what arguments they accept
    * and what the return values are
    */
    interface Person
    {
        /**
        * Returns an object of the class Order that represents
        * the first time this person successfully completed an
        * order
        */
        public function firstSuccessfulOrder();
        
        /**
        * Accepts the "month since first purchase" as an 
        * argument and returns total amount purchased in
        * that month by this Person. For example, if they
        * first purchased in January, and the parameter
        * passed in is 0, you would return the total
        * spent in January. If the parameter was 1,
        * you would return the total spent in February.
        */
        public function totalPurchasesInMonth($index);
    }

    interface Order
    {
        /**
        * Return a string that can be parsed by
        * php.net/strtotime representing the date
        * and time of this order
        */
        public function orderDate();
    }