multi_touch.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "d9d74b4f",
   "metadata": {},
   "source": [
    "# Multi-Touch Attribution Models\n",
    "Multi-touch models assume that all touchpoints play some role in driving a conversion. We look at the most popular multi-touch attribution models which include Linear, Position decay, and U-shaped models. These models have become important for marketers, particularly those looking to measure the impact of digital campaigns, since they provide a more granular and person-level view than traditional aggregation methods. These models should be used when the number of channels for a campaign is 5 to 10, the most preferable of which are position-based, like the U-shaped Attribution Model.\n",
    "\n",
    "## Import Dependencies and Load Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "f49c43ae",
   "metadata": {},
   "outputs": [],
   "source": [
    "from collections import Counter\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "\n",
    "%matplotlib inline\n",
    "pd.options.mode.chained_assignment = None\n",
    "sns.set_palette('summer')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "ce64689e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>cookie</th>\n",
       "      <th>time</th>\n",
       "      <th>interaction</th>\n",
       "      <th>conversion</th>\n",
       "      <th>conversion_value</th>\n",
       "      <th>channel</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>00000FkCnDfDDf0iC97iC703B</td>\n",
       "      <td>2018-07-03T13:02:11Z</td>\n",
       "      <td>impression</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>Instagram</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>00000FkCnDfDDf0iC97iC703B</td>\n",
       "      <td>2018-07-17T19:15:07Z</td>\n",
       "      <td>impression</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>Online Display</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>00000FkCnDfDDf0iC97iC703B</td>\n",
       "      <td>2018-07-24T15:51:46Z</td>\n",
       "      <td>impression</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>Online Display</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>00000FkCnDfDDf0iC97iC703B</td>\n",
       "      <td>2018-07-29T07:44:51Z</td>\n",
       "      <td>impression</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>Online Display</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0000nACkD9nFkBBDECD3ki00E</td>\n",
       "      <td>2018-07-03T09:44:57Z</td>\n",
       "      <td>impression</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>Paid Search</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                      cookie                  time interaction  conversion  \\\n",
       "0  00000FkCnDfDDf0iC97iC703B  2018-07-03T13:02:11Z  impression           0   \n",
       "1  00000FkCnDfDDf0iC97iC703B  2018-07-17T19:15:07Z  impression           0   \n",
       "2  00000FkCnDfDDf0iC97iC703B  2018-07-24T15:51:46Z  impression           0   \n",
       "3  00000FkCnDfDDf0iC97iC703B  2018-07-29T07:44:51Z  impression           0   \n",
       "4  0000nACkD9nFkBBDECD3ki00E  2018-07-03T09:44:57Z  impression           0   \n",
       "\n",
       "   conversion_value         channel  \n",
       "0               0.0       Instagram  \n",
       "1               0.0  Online Display  \n",
       "2               0.0  Online Display  \n",
       "3               0.0  Online Display  \n",
       "4               0.0     Paid Search  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv('data/attribution_data.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ea65e53b",
   "metadata": {},
   "source": [
    "## Linear Attribution Model\n",
    "The Linear Attribution Model gives each touchpoint across the buyer journey the same amount of credit toward driving a sale; it values every touchpoint evenly. This model is easy to implement and is better than all the single-touch attribution models. The *disadvantage* here is that in reality, consumers aren't equally impacted by every kind of channel."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "27d190a6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Weight (%)</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Paid Search</th>\n",
       "      <td>27.63</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Online Display</th>\n",
       "      <td>26.43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Instagram</th>\n",
       "      <td>16.56</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Facebook</th>\n",
       "      <td>16.28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Online Video</th>\n",
       "      <td>13.11</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                Weight (%)\n",
       "Paid Search          27.63\n",
       "Online Display       26.43\n",
       "Instagram            16.56\n",
       "Facebook             16.28\n",
       "Online Video         13.11"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def linear_model(df, conv_col, channel_col, user_id):\n",
    "    \"\"\"\n",
    "    Build a Linear Multi-Touch Attribution Model\n",
    "    \"\"\"\n",
    "    # Keep data of only those users who are getting converted\n",
    "    tmp = df[df[conv_col] == 1]\n",
    "    cookie_index = list(tmp[user_id])\n",
    "    df['new'] = df[user_id].isin(cookie_index)\n",
    "    y = df['new'].isin([True])\n",
    "    df_conv = df[y]\n",
    "    tmp = pd.DataFrame(df_conv.groupby(user_id).tail(1))\n",
    "    \n",
    "    # Get total amount of channels visited by user\n",
    "    x = Counter(df_conv[user_id])\n",
    "    tmp['user_click_count'] = x.values()\n",
    "    tmp.set_index(user_id, inplace=True)\n",
    "    df_conv['clicks'] = df_conv[user_id].map(x)\n",
    "    \n",
    "    # Assign weights\n",
    "    df_conv = df_conv.assign(clicks_per=lambda x: round(\n",
    "        100 / df_conv['clicks'], 2))\n",
    "    \n",
    "    # Get the mean weight of every channel\n",
    "    res_linear = df_conv.groupby(\n",
    "        channel_col, as_index=False)['clicks_per'].mean()\n",
    "    _sum = res_linear['clicks_per'].sum()\n",
    "    \n",
    "    # Get as a percentage\n",
    "    res_linear['Weight (%)'] = res_linear.apply(lambda x: round(\n",
    "        (x['clicks_per'] / _sum) * 100, 2), axis=1)\n",
    "    res_linear.drop(['clicks_per'], inplace=True, axis=1)\n",
    "    res_linear = res_linear.set_index(channel_col)\n",
    "    res_linear.index.name = None\n",
    "    return res_linear.sort_values(by='Weight (%)', ascending=False)\n",
    "\n",
    "linear = linear_model(df, 'conversion', 'channel', 'cookie')\n",
    "linear"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "9181432a",
   "metadata": {},
   "outputs": [],
   "source": [
    "linear.to_csv('results/linear.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "6136ae08",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.subplots(figsize=(10, 6))\n",
    "sns.barplot(data=linear, x=linear.index, y='Weight (%)')\n",
    "plt.xlabel('Channel')\n",
    "plt.title('Linear Model Channel Weights', fontsize=20)\n",
    "plt.savefig('plots/linear.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e10a42d8",
   "metadata": {},
   "source": [
    "From the linear model, we see that the leaders are Paid Search and Online Display. These results were au contraire to our single-touch models, where Facebook dominated, which demonstrates that we can't just rely on one model, since our results are dynamic.\n",
    "## Position-Based (U-Shaped) Attribution Model\n",
    "This model splits the credit for a sale between the first interaction and the moment it converts.\n",
    "> 40% of the credit is given to each of these points, with the remaining 20% spread out between any other interactions that happened in the middle."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "a0f67781",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Weight (%)</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Paid Search</th>\n",
       "      <td>27.48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Online Display</th>\n",
       "      <td>26.61</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Instagram</th>\n",
       "      <td>16.65</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Facebook</th>\n",
       "      <td>16.28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Online Video</th>\n",
       "      <td>12.99</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                Weight (%)\n",
       "Paid Search          27.48\n",
       "Online Display       26.61\n",
       "Instagram            16.65\n",
       "Facebook             16.28\n",
       "Online Video         12.99"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def calc_attribution_u(click_pos, total_clicks):\n",
    "    \"\"\"\n",
    "    Helper function to calculate attributions\n",
    "    \"\"\"\n",
    "    default_att = 0.5  # If a user has visited only two channels\n",
    "    extreme_touch_att_1 = 0.4  # Assigning weight to first and last channels\n",
    "    intermed_att_1 = 0.2  # Total weight for remaining channels\n",
    "    \n",
    "    if total_clicks == 2:\n",
    "        return default_att\n",
    "    elif total_clicks == 1:\n",
    "        return 1\n",
    "    else:\n",
    "        if click_pos == total_clicks or click_pos == 1:\n",
    "            return extreme_touch_att_1\n",
    "        else:\n",
    "            return intermed_att_1 / (total_clicks - 2)\n",
    "        \n",
    "\n",
    "def U_shaped_model(df, conv_col, channel_col, user_id):\n",
    "    \"\"\"\n",
    "    Build a U-Shaped Multi-Touch Attribution Model\n",
    "    \"\"\"\n",
    "    temp = df[df[conv_col] == 1]\n",
    "    cookie_index = list(temp[user_id])\n",
    "    df['new'] = df[user_id].isin(cookie_index)\n",
    "    y = df['new'].isin([True])\n",
    "    df_conv = df[y]\n",
    "    \n",
    "    count = Counter(df_conv[user_id])\n",
    "    df_conv['clicks'] = df_conv[user_id].map(count)\n",
    "    df_conv['click_pos'] = df_conv.groupby(user_id).cumcount() + 1\n",
    "    df_U_shaped = df_conv\n",
    "    df_U_shaped['U_shape'] = df_conv.apply(lambda x: round(\n",
    "        calc_attribution_u(x['click_pos'], x['clicks']) * 100, 2), axis=1)\n",
    "    \n",
    "    # Get mean weight of every channel, don't return group names as index\n",
    "    res_U_shaped = df_U_shaped.groupby(\n",
    "        channel_col, as_index=False)['U_shape'].mean()\n",
    "    _sum = res_U_shaped['U_shape'].sum()\n",
    "    res_U_shaped['Weight (%)'] = res_U_shaped.apply(\n",
    "        lambda x: round((x['U_shape'] / _sum) * 100, 2), axis=1)\n",
    "    res_U_shaped.drop(['U_shape'], inplace=True, axis=1)\n",
    "    res_U_shaped = res_U_shaped.set_index(channel_col)\n",
    "    res_U_shaped.index.name = None\n",
    "    return res_U_shaped.sort_values(by='Weight (%)', ascending=False)\n",
    "\n",
    "U_shaped = U_shaped_model(df, 'conversion', 'channel', 'cookie')\n",
    "U_shaped"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "e222c119",
   "metadata": {},
   "outputs": [],
   "source": [
    "U_shaped.to_csv('results/U_shaped.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "0a918b00",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.subplots(figsize=(10, 6))\n",
    "sns.barplot(data=U_shaped, x=U_shaped.index, y='Weight (%)')\n",
    "plt.xlabel('Channel')\n",
    "plt.title('U-Shaped Model Channel Weights', fontsize=20)\n",
    "plt.savefig('plots/U_shaped.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "495fb8fd",
   "metadata": {},
   "source": [
    "Our U-Shaped Model results are nearly identical to those of our Linear Model.\n",
    "## Position Decay Multi-Touch Attribution Model\n",
    "This algorithm allocates points to all the channels used to conert, giving them credit in ascending order up to the last interaction. This modle assumes that a user's interest in a product grows with every interaction, and the last one is the most important as it determines whether a customer converts."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "38be42dd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Weight (%)</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Paid Search</th>\n",
       "      <td>25.64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Online Display</th>\n",
       "      <td>25.44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Instagram</th>\n",
       "      <td>17.14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Facebook</th>\n",
       "      <td>17.10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Online Video</th>\n",
       "      <td>14.68</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                Weight (%)\n",
       "Paid Search          25.64\n",
       "Online Display       25.44\n",
       "Instagram            17.14\n",
       "Facebook             17.10\n",
       "Online Video         14.68"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def calc_attribution_decay(click_pos, total_clicks):\n",
    "    \"\"\"\n",
    "    Calculate attribution according to exponential decay of channel\n",
    "    \"\"\"\n",
    "    rel_pos = total_clicks - click_pos\n",
    "    attribution = pow(2, -rel_pos)  # Exponential decay\n",
    "    return attribution\n",
    "\n",
    "\n",
    "def pos_decay_model(df, conv_col, channel_col, user_id):\n",
    "    \"\"\"\n",
    "    Build a Position Decay Multi-Touch Attribution Model\n",
    "    \"\"\"\n",
    "    # Filter out non-converted users\n",
    "    tmp = df[df[conv_col] == 1]\n",
    "    cookie_index = list(tmp[user_id])\n",
    "    df['new'] = df[user_id].isin(cookie_index)\n",
    "    y = df['new'].isin([True])\n",
    "    df_conv = df[y]\n",
    "    \n",
    "    count = Counter(df_conv[user_id])\n",
    "    df_conv['clicks'] = df_conv[user_id].map(count)\n",
    "    df_conv = df_conv.assign(\n",
    "        clicks_per=lambda x: round(100 / df_conv['clicks'], 2))\n",
    "    df_conv['click_pos'] = df_conv.groupby(user_id).cumcount() + 1\n",
    "    df_conv['pos_decay'] = df_conv.apply(lambda x: round(\n",
    "        calc_attribution_decay(\n",
    "            x['click_pos'], x['clicks']) * 100, 2), axis=1)\n",
    "    df_pos_decay = df_conv\n",
    "    \n",
    "    # Get mean weight of every channel, don't return group names as index\n",
    "    res_pos_decay = df_pos_decay.groupby(\n",
    "        'channel', as_index=False)['pos_decay'].mean()\n",
    "    _sum = res_pos_decay['pos_decay'].sum()\n",
    "    res_pos_decay['Weight (%)'] = res_pos_decay.apply(\n",
    "        lambda x: round((x['pos_decay'] / _sum) * 100, 2), axis=1)\n",
    "    res_pos_decay.drop(['pos_decay'], axis=1, inplace=True)\n",
    "    res_pos_decay = res_pos_decay.set_index(channel_col)\n",
    "    res_pos_decay.index.name = None\n",
    "    return res_pos_decay.sort_values(by='Weight (%)', ascending=False)\n",
    "\n",
    "\n",
    "pos_decay = pos_decay_model(df, 'conversion', 'channel', 'cookie')\n",
    "pos_decay"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "9f0b5bfd",
   "metadata": {},
   "outputs": [],
   "source": [
    "pos_decay.to_csv('results/pos_decay.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "3aaa6699",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.subplots(figsize=(10, 6))\n",
    "sns.barplot(data=pos_decay, x=pos_decay.index, y='Weight (%)')\n",
    "plt.xlabel('Channel')\n",
    "plt.title('Position Decay Model Channel Weights', fontsize=20)\n",
    "plt.savefig('plots/pos_decay.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c8ea6186",
   "metadata": {},
   "source": [
    "All of our multi-touch attribution models preached Paid Search and Online Display, unlike our single-touch models, which suggested that these channels don't perform as well. With these different results, we haven't yet reached our goals of optimizing our budget; which channels will drive the most revenue?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "63d2c7bf",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.8.5 64-bit",
   "language": "python",
   "name": "python38564bit40f31465e77c45df95a515d1a253e522"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}