100 Exercises / Marketing Science / Marketing Science 100 Exercises

Introduction to Marketing Science in Manufacturing | Learning KPI, Causality, and Budget Optimization with Python

2020-08-27

From ‘Intuition’ to ‘Decision-Making’ in Manufacturing B2B Marketing: 10 Steps into Marketing Science

Title & Overview

This notebook is an introductory guide for sales and marketing departments in manufacturing to make data-driven decisions about investments in exhibitions, technical seminars, web advertising, and more. Through all 100 courses, we progressed step by step to statistics, customer analysis, demand forecasting, pricing, recommendations, optimization, and management decision-making. In the first 10 articles, we examine the underlying Question,KPIModels, causality, uncertainty, simulation using data from a fictional industrial equipment manufacturer.

The goal of learning is not to memorize the methods. It means being able to explain decisions like, “Which measures to allocate to what, what to measure, and how to change next” in reproducible ways.

[!NOTE] This material is a notebook previously used by Surikoubo (or personally by the representative, Kazuyama), and has been reconstructed, edited, and published with the company’s permission.
All data listed is fictional and has no relation whatsoever to real companies, factories, or figures.

Introduction: Practical Challenges in Manufacturing Covered in This Article

The fictional industrial equipment manufacturer ‘Kobo Tech’ is receiving inquiries from multiple channels. However, it takes a long time to secure an order, and the scale of the project varies. Simply relying on lead counts or order numbers is not enough to decide whether to strengthen trade shows or web advertising.

In this article, marketing is treated not as an “aggregation of ads,” but as a Decision-making that allocates limited management resources to market opportunities. We look at not only sales but also gross profit, sales load, certainty, and variation simultaneously.

Common situations on site

The definition of “results” varies by department
The number of leads has increased, but the number of deals that sales can pursue has not increased.
Last year’s results are being used directly for the next year’s budget
When orders increase after a policy, people tend to interpret it as the effect of the initiative.
Planning based solely on average values without determining how to respond to downside fluctuations

Why is this issue so difficult to judge?

In manufacturing B2B, there is a time lag between measures and order receipt, and business conditions, product strength, and sales activities all affect results. Observational data alone does not reveal false hypotheticals, that is, what would happen if no measures were taken. Also, correct predictions do not necessarily lead to correct actions. It is necessary to convert forecasted values into decision-making under the constraints of profit, capability, and risk.

Overview of Exercise covered this time

No.	Theme	Questions to answer on site
001	What is Marketing Science?	How to connect data to decision-making
002	Why is mathematics necessary?	How to compare measures with different unit prices and probabilities
003	Data-Driven Management	How to change meetings using common metrics
004	KPI	How to connect leading indicators and outcome indicators
005	Decision-Making and Models	How to use order probabilities to assist decision-making
006	Correlation and Causality	How to avoid superficial effects
007	Optimization	Where to allocate the budget
008	uncertainty	How to handle not only averages but also downsides
009	Simulation	How to visualize the distribution of decision-making
010	Python environment setup	How to create reproducible analytical environments

Preparing the Python environment

No external data is used. Generate random numbers with NumPy, aggregate them with pandas, and visualize them with matplotlib. Because the seed of the random number generator is fixed, the same result can be reproduced no matter how many times you run it.

import sys
import numpy as np
import pandas as pd
import matplotlib
import matplotlib.pyplot as plt
from IPython.display import display

rng = np.random.default_rng(42)
pd.set_option("display.max_columns", 20)
pd.set_option("display.float_format", lambda x: f"{x:,.2f}")
plt.rcParams["figure.figsize"] = (8, 4.5)
plt.rcParams["axes.unicode_minus"] = False

print("Python:", sys.version.split()[0])
print("NumPy:", np.__version__)
print("pandas:", pd.__version__)
print("matplotlib:", matplotlib.__version__)

Python: 3.11.9
NumPy: 1.26.4
pandas: 2.2.2
matplotlib: 3.9.2

Creation of Fictional Data

We will build a track record of 24-month, four-channel initiatives. In addition to costs, lead numbers, negotiation rates, order probability, deal unit price, and gross profit margin, we also provide market indices. The amount is in ten thousand yen. In practice, CRM, advertising management, exhibition directories, and core system IDs are connected to create similar analysis tables.

months = pd.date_range("2024-01-01", periods=24, freq="MS")
channels = ["Exhibition", "Technical Seminar", "WebAdvertisement", "Introducing Existing Customers"]
base = {
    "Exhibition": (300, 85, 0.24, 0.22, 850, 0.35),
    "Technical Seminar": (130, 48, 0.31, 0.25, 720, 0.37),
    "WebAdvertisement": (180, 125, 0.14, 0.16, 520, 0.32),
    "Introducing Existing Customers": (70, 24, 0.44, 0.38, 980, 0.40),
}
rows = []
for t, month in enumerate(months):
    market = 100 + 8*np.sin(2*np.pi*t/12) + rng.normal(0, 3)
    for ch in channels:
        cost, leads0, opp_rate, win0, price0, margin = base[ch]
        spend = max(20, cost * rng.normal(1, 0.12))
        leads = rng.poisson(leads0 * (spend/cost)**0.65 * market/100)
        opportunities = rng.binomial(leads, opp_rate)
        win_prob = np.clip(win0 + 0.0025*(market-100), 0.05, 0.70)
        orders = rng.binomial(opportunities, win_prob)
        avg_price = max(200, price0 * rng.lognormal(0, 0.10))
        revenue = orders * avg_price
        gross_profit = revenue * margin - spend
        rows.append([month, ch, market, spend, leads, opportunities, orders,
                     avg_price, revenue, margin, gross_profit])

df = pd.DataFrame(rows, columns=["month", "channel", "market_index", "spend",
    "leads", "opportunities", "orders", "avg_price", "revenue", "margin_rate", "gross_profit"])
df["lead_to_opportunity"] = df["opportunities"] / df["leads"].replace(0, np.nan)
df["win_rate"] = df["orders"] / df["opportunities"].replace(0, np.nan)
display(df.head(8))
print(f"Number of lines: {len(df):,}, Number of Missing Items: {df.isna().sum().sum():,}")

	month	channel	market_index	spend	leads	opportunities	orders	avg_price	revenue	margin_rate	gross_profit	lead_to_opportunity	win_rate
0	2024-01-01	Exhibition	100.91	262.56	90	16	7	860.94	6,026.55	0.35	1,846.73	0.18	0.44
1	2024-01-01	Technical Seminar	100.91	125.07	38	11	5	724.77	3,623.85	0.37	1,215.76	0.29	0.45
2	2024-01-01	WebAdvertisement	100.91	204.35	135	19	1	567.75	567.75	0.32	-22.67	0.14	0.05
3	2024-01-01	Introducing Existing Customers	100.91	69.58	28	17	9	938.91	8,450.19	0.40	3,310.50	0.61	0.53
4	2024-02-01	Exhibition	102.94	319.16	68	18	5	816.15	4,080.74	0.35	1,109.09	0.26	0.28
5	2024-02-01	Technical Seminar	102.94	122.01	45	11	1	661.98	661.98	0.37	122.92	0.24	0.09
6	2024-02-01	WebAdvertisement	102.94	162.19	126	21	4	549.03	2,196.10	0.32	540.56	0.17	0.19
7	2024-02-01	Introducing Existing Customers	102.94	64.41	28	11	5	1,002.16	5,010.80	0.40	1,939.91	0.39	0.45

Number of lines: 96, Number of missing items: 0

No.001: What is Marketing Science?

Meaning in Practice

Marketing science is the activity of quantifying customer understanding, forecasting, evaluating initiatives, and resource allocation to improve decision-making. The goal is not to create an analytical report itself, but to Clearly state options, evaluation criteria, and constraints when choosing actions.

Approach to Analysis and Modeling

Compare channel $c$ value by gross margin after investment expenses, not sales.

$\text{Contribution}_c = \text{Revenue}_c\times\text{MarginRate}_c-\text{Spend}_c$

Check with Python

channel_summary = df.groupby("channel").agg(
    spend=("spend", "sum"), leads=("leads", "sum"), orders=("orders", "sum"),
    revenue=("revenue", "sum"), gross_profit=("gross_profit", "sum")
)
channel_summary["profit_per_spend"] = channel_summary["gross_profit"] / channel_summary["spend"]
display(channel_summary.sort_values("gross_profit", ascending=False).round(2))

	spend	leads	orders	revenue	gross_profit	profit_per_spend
channel
Introducing Existing Customers	1,653.50	592	92	87,987.12	33,541.35	20.29
Exhibition	7,068.09	1996	123	103,732.18	29,238.17	4.14
Technical Seminar	3,131.49	1117	87	63,869.47	20,500.22	6.55
WebAdvertisement	4,317.58	3045	54	27,153.56	4,371.56	1.01

Reading the results

The rankings of sales, orders, and investment efficiency do not always match. The choice depends on whether management prioritizes “growth amount” or “capital efficiency.” The first step in analysis is not to come up with a single correct answer, but to agree on what to optimize.

No.002: Why Mathematics Is Necessary for Marketing

Meaning in Practice

If you choose strategies based solely on lead price, you may miss out on high-cost, high-probability deals. By converting probability and economic value to the same scale, you can compare different channels.

Approach to Analysis and Modeling

Expected gross profit margin for 1 lead

$EV=(Negotiation rate\timesOrder rate\timesAverage unit price\timesgross profit margin)-1Cost per lead$

That’s how it is defined. The expected value is a long-term average and does not guarantee a single-month outcome.

Check with Python

economics = df.groupby("channel").agg(
    spend=("spend", "sum"), leads=("leads", "sum"), opportunities=("opportunities", "sum"),
    orders=("orders", "sum"), revenue=("revenue", "sum"), margin_rate=("margin_rate", "mean")
)
economics["opportunity_rate"] = economics["opportunities"] / economics["leads"]
economics["win_rate"] = economics["orders"] / economics["opportunities"]
economics["avg_order_value"] = economics["revenue"] / economics["orders"]
economics["cost_per_lead"] = economics["spend"] / economics["leads"]
economics["expected_value_per_lead"] = (economics["opportunity_rate"] * economics["win_rate"] *
    economics["avg_order_value"] * economics["margin_rate"] - economics["cost_per_lead"])
display(economics[["cost_per_lead", "opportunity_rate", "win_rate", "expected_value_per_lead"]].round(2))

	cost_per_lead	opportunity_rate	win_rate	expected_value_per_lead
channel
WebAdvertisement	1.42	0.13	0.13	1.44
Exhibition	3.54	0.25	0.24	14.65
Technical Seminar	2.80	0.33	0.24	18.35
Introducing Existing Customers	2.79	0.44	0.35	56.66

Reading the results

Cheap leads are not necessarily the most valuable leads. Mathematics translates chains of probability and amounts into a single evaluation axis. However, if you lack the ability to pursue customers, you should also include sales work in the expenses.

No.003: What is Data-Driven Management?

Meaning in Practice

Data-driven management is not about automatically following numbers. It means making defined data a common language and continuously updating hypotheses, judgments, results, and learning.

Approach to Analysis and Modeling

In the monthly management view, activity volume (costs), funnel (leads and negotiations), and results (orders and gross profit) are arranged at the same granularity. A structure that can show differences from previous years and targets is important.

Check with Python

monthly = df.groupby("month").agg(spend=("spend", "sum"), leads=("leads", "sum"),
    opportunities=("opportunities", "sum"), orders=("orders", "sum"),
    gross_profit=("gross_profit", "sum")).reset_index()
monthly["rolling_3m_profit"] = monthly["gross_profit"].rolling(3).mean()
display(monthly.tail(6).round(1))

fig, ax = plt.subplots()
ax.plot(monthly["month"], monthly["gross_profit"], marker="o", alpha=.55, label="Monthly")
ax.plot(monthly["month"], monthly["rolling_3m_profit"], linewidth=2.5, label="3-month average")
ax.set_title("Monthly contribution profit and trend")
ax.set_xlabel("Month"); ax.set_ylabel("Contribution profit (10k JPY)")
ax.grid(True, alpha=.3); ax.legend(); fig.tight_layout(); plt.show()

	month	spend	leads	opportunities	orders	gross_profit	rolling_3m_profit
18	2025-07-01	653.50	279	56	16	3,956.50	3,571.70
19	2025-08-01	630.40	256	50	14	3,758.90	3,500.80
20	2025-09-01	684.80	244	60	8	1,835.10	3,183.50
21	2025-10-01	591.80	250	62	15	3,708.40	3,100.80
22	2025-11-01	652.70	250	55	9	1,389.40	2,310.90
23	2025-12-01	633.60	277	69	17	3,598.60	2,898.80

png

Reading the results

Monthly values fluctuate due to large contracts. Including the 3-month moving average makes it easier to read the trend. In meetings, it is important to record the “reasons for increases/decreases” and “next verification actions,” rather than just viewing the dashboard.

No.004: The Concept of KPIs

Meaning in Practice

Sales are important, but by the time the results are finalized, the move is delayed. KPIs are designed as indicators that the site can operate weekly and monthly, leading to future results.

Approach to Analysis and Modeling

The funnel can be represented by the following multiplicative factorization.

$Number of orders received=Number of leads\timesNegotiation rate\timesOrder rate$

Gross profit is the final KGI, the number of deals and order rates are KPIs, and the amount of work such as email submissions is PI.

Check with Python

kpi = df.groupby("channel").agg(leads=("leads", "sum"), opportunities=("opportunities", "sum"),
    orders=("orders", "sum"), gross_profit=("gross_profit", "sum"))
kpi["opportunity_rate"] = kpi["opportunities"] / kpi["leads"]
kpi["win_rate"] = kpi["orders"] / kpi["opportunities"]
kpi["orders_reconstructed"] = kpi["leads"] * kpi["opportunity_rate"] * kpi["win_rate"]
display(kpi.round(3))

fig, ax = plt.subplots()
x = np.arange(len(kpi)); width = .35
ax.bar(x-width/2, kpi["opportunity_rate"], width, label="Lead to opportunity")
ax.bar(x+width/2, kpi["win_rate"], width, label="Win rate")
ax.set_title("Conversion KPIs by channel"); ax.set_xlabel("Channel"); ax.set_ylabel("Rate")
ax.set_xticks(x, kpi.index, rotation=15); ax.grid(True, axis="y", alpha=.3); ax.legend()
fig.tight_layout(); plt.show()

	leads	opportunities	orders	gross_profit	opportunity_rate	win_rate	orders_reconstructed
channel
WebAdvertisement	3045	405	54	4,371.56	0.13	0.13	54.00
Exhibition	1996	508	123	29,238.17	0.26	0.24	123.00
Technical Seminar	1117	367	87	20,500.22	0.33	0.24	87.00
Introducing Existing Customers	592	262	92	33,541.35	0.44	0.35	92.00

/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 24195 (\N{CJK UNIFIED IDEOGRAPH-5E83}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 21578 (\N{CJK UNIFIED IDEOGRAPH-544A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 23637 (\N{CJK UNIFIED IDEOGRAPH-5C55}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 31034 (\N{CJK UNIFIED IDEOGRAPH-793A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 20250 (\N{CJK UNIFIED IDEOGRAPH-4F1A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 25216 (\N{CJK UNIFIED IDEOGRAPH-6280}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 34899 (\N{CJK UNIFIED IDEOGRAPH-8853}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 12475 (\N{KATAKANA LETTER SE}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 12511 (\N{KATAKANA LETTER MI}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 12490 (\N{KATAKANA LETTER NA}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 12540 (\N{KATAKANA-HIRAGANA PROLONGED SOUND MARK}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 26082 (\N{CJK UNIFIED IDEOGRAPH-65E2}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 23384 (\N{CJK UNIFIED IDEOGRAPH-5B58}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 39015 (\N{CJK UNIFIED IDEOGRAPH-9867}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 23458 (\N{CJK UNIFIED IDEOGRAPH-5BA2}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 32057 (\N{CJK UNIFIED IDEOGRAPH-7D39}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1667793017.py:14: UserWarning: Glyph 20171 (\N{CJK UNIFIED IDEOGRAPH-4ECB}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 24195 (\N{CJK UNIFIED IDEOGRAPH-5E83}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 21578 (\N{CJK UNIFIED IDEOGRAPH-544A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23637 (\N{CJK UNIFIED IDEOGRAPH-5C55}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 31034 (\N{CJK UNIFIED IDEOGRAPH-793A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 20250 (\N{CJK UNIFIED IDEOGRAPH-4F1A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 25216 (\N{CJK UNIFIED IDEOGRAPH-6280}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 34899 (\N{CJK UNIFIED IDEOGRAPH-8853}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12475 (\N{KATAKANA LETTER SE}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12511 (\N{KATAKANA LETTER MI}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12490 (\N{KATAKANA LETTER NA}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12540 (\N{KATAKANA-HIRAGANA PROLONGED SOUND MARK}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 26082 (\N{CJK UNIFIED IDEOGRAPH-65E2}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23384 (\N{CJK UNIFIED IDEOGRAPH-5B58}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 39015 (\N{CJK UNIFIED IDEOGRAPH-9867}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23458 (\N{CJK UNIFIED IDEOGRAPH-5BA2}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 32057 (\N{CJK UNIFIED IDEOGRAPH-7D39}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 20171 (\N{CJK UNIFIED IDEOGRAPH-4ECB}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)

png

Reading the results

Even if the number of leads is high, a low negotiation rate will only increase the sales burden. KPIs are not indicators that blame departments, but rather instruments that identify bottlenecks. Manage definitions, responsible persons, update frequency, and execution methods as a set.

No.005: Decision-Making and Models

Meaning in Practice

The model is a condensation of reality. Rather than a complete reproduction, it leaves the elements necessary for a certain judgment. Here, we support you with whether to pursue clients based on order probabilities and expected gross profit.

Approach to Analysis and Modeling

For probability $p$ , $G$ gross profit margin on orders, and $C$ customer acquisition costs,

$EV=pG-C$

If it’s correct, then it’s a potential follower. However, elements that are difficult to quantify, such as strategic customers and learning value, are supplemented in the final decision.

Check with Python

deals = pd.DataFrame({
    "deal": [f"D-{i:03d}" for i in range(1, 9)],
    "win_probability": [0.12, .25, .32, .45, .55, .62, .70, .82],
    "gross_profit_if_won": [600, 420, 900, 300, 520, 750, 280, 450],
    "follow_up_cost": [50, 45, 85, 70, 65, 90, 55, 80]
})
deals["expected_net_value"] = deals["win_probability"]*deals["gross_profit_if_won"]-deals["follow_up_cost"]
deals["model_recommendation"] = np.where(deals["expected_net_value"] > 0, "Prioritize", "Review")
display(deals.sort_values("expected_net_value", ascending=False).round(1))

	deal	win_probability	gross_profit_if_won	follow_up_cost	expected_net_value	model_recommendation
5	D-006	0.60	750	90	375.00	Prioritize
7	D-008	0.80	450	80	289.00	Prioritize
4	D-005	0.60	520	65	221.00	Prioritize
2	D-003	0.30	900	85	203.00	Prioritize
6	D-007	0.70	280	55	141.00	Prioritize
3	D-004	0.40	300	70	65.00	Prioritize
1	D-002	0.20	420	45	60.00	Prioritize
0	D-001	0.10	600	50	22.00	Prioritize

Reading the results

Even if the chances of getting orders are low, if the gross profit margin from the deal is high, it has value for attracting clients. Conversely, ranking based solely on accuracy overlooks economic value. Probability calibration, cost ranges, and exception approval rules must be regularly audited.

No.006: Correlation and Causality

Meaning in Practice

Even if advertising spend and sales increase simultaneously, it does not necessarily mean that advertising has increased sales. It may have been that advertising spending was increased during a boom period, and market conditions became a confounding factor.

Approach to Analysis and Modeling

Correlation refers to the strength with which variables move together, while causality refers to the relationship where outcomes vary depending on the intervention. Causality requires discriminative designs such as randomized comparisons, differences in differences, and regression discontinuities. Here, we also check the residual correlation after excluding market conditions, but this alone cannot prove causality.

Check with Python

web = df[df["channel"] == "WebAdvertisement"].copy()
raw_corr = web["spend"].corr(web["revenue"])
spend_coef = np.polyfit(web["market_index"], web["spend"], 1)
revenue_coef = np.polyfit(web["market_index"], web["revenue"], 1)
web["spend_residual"] = web["spend"] - np.polyval(spend_coef, web["market_index"])
web["revenue_residual"] = web["revenue"] - np.polyval(revenue_coef, web["market_index"])
adjusted_corr = web["spend_residual"].corr(web["revenue_residual"])
print(f"simple correlation: {raw_corr:.3f}")
print(f"Residual correlation excluding linear market effects: {adjusted_corr:.3f}")

fig, ax = plt.subplots()
sc = ax.scatter(web["spend"], web["revenue"], c=web["market_index"], cmap="viridis", s=65)
ax.set_title("Web spend and revenue colored by market conditions")
ax.set_xlabel("Spend (10k JPY)"); ax.set_ylabel("Revenue (10k JPY)")
ax.grid(True, alpha=.3); fig.colorbar(sc, ax=ax, label="Market index")
fig.tight_layout(); plt.show()

Simple correlation: -0.040
Residual correlation excluding linear market effects: 0.096

png

Reading the results

If the correlation changes before and after the correction, it becomes clear that interpreting the market without regard is dangerous. However, unobserved sales power and seasonality may also remain. In the next measure, we will consider experimental designs that randomly divide regions and customer groups.

No.007: The Concept of Optimization

Meaning in Practice

Even if the efficiency of each initiative is known, allocation is not self-evident due to budget limits, minimum spending requirements, and sales capability. Optimization explicitly explores candidates by clearly stating their objectives and constraints.

Approach to Analysis and Modeling

Maximize expected gross profit $f_c(x_c)$ , including diminishing returns, against channel-specific budget $x_c$ .

$\max_x \sum_c f_c(x_c)\quad \text{s.t.}\quad \sum_c x_c\leq B,\;L_c\leq x_c\leq U_c$

As an introduction, I will list all combinations in 500,000 yen increments.

Check with Python

budget = 800
grid = np.arange(50, 501, 50)
response = {"Exhibition": (12.0, 240), "Technical Seminar": (10.5, 150),
            "WebAdvertisement": (8.5, 180), "Introducing Existing Customers": (14.0, 90)}
plans = []
for a in grid:
    for b in grid:
        for c in grid:
            for d in grid:
                alloc = np.array([a, b, c, d])
                if alloc.sum() <= budget:
                    expected = sum(scale*np.sqrt(x) - x*0.15 for x, (scale, _) in zip(alloc, response.values()))
                    plans.append([*alloc, expected])
plans = pd.DataFrame(plans, columns=[*channels, "expected_profit"])
best = plans.nlargest(5, "expected_profit")
display(best.round(1))

fig, ax = plt.subplots()
ax.bar(channels, best.iloc[0][channels])
ax.set_title("Optimized marketing budget allocation")
ax.set_xlabel("Channel"); ax.set_ylabel("Budget (10k JPY)")
ax.grid(True, axis="y", alpha=.3); ax.tick_params(axis="x", rotation=15)
fig.tight_layout(); plt.show()

	Exhibition	Technical Seminar	WebAdvertisement	Introducing Existing Customers	expected_profit
1373	250	150	100	300	525.80
1208	200	200	100	300	525.70
1174	200	150	100	350	525.20
1180	200	150	150	300	524.90
1399	250	200	100	250	524.60

/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 23637 (\N{CJK UNIFIED IDEOGRAPH-5C55}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 31034 (\N{CJK UNIFIED IDEOGRAPH-793A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 20250 (\N{CJK UNIFIED IDEOGRAPH-4F1A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 25216 (\N{CJK UNIFIED IDEOGRAPH-6280}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 34899 (\N{CJK UNIFIED IDEOGRAPH-8853}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 12475 (\N{KATAKANA LETTER SE}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 12511 (\N{KATAKANA LETTER MI}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 12490 (\N{KATAKANA LETTER NA}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 12540 (\N{KATAKANA-HIRAGANA PROLONGED SOUND MARK}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 24195 (\N{CJK UNIFIED IDEOGRAPH-5E83}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 21578 (\N{CJK UNIFIED IDEOGRAPH-544A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 26082 (\N{CJK UNIFIED IDEOGRAPH-65E2}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 23384 (\N{CJK UNIFIED IDEOGRAPH-5B58}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 39015 (\N{CJK UNIFIED IDEOGRAPH-9867}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 23458 (\N{CJK UNIFIED IDEOGRAPH-5BA2}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 32057 (\N{CJK UNIFIED IDEOGRAPH-7D39}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/3625590457.py:23: UserWarning: Glyph 20171 (\N{CJK UNIFIED IDEOGRAPH-4ECB}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23637 (\N{CJK UNIFIED IDEOGRAPH-5C55}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 31034 (\N{CJK UNIFIED IDEOGRAPH-793A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 20250 (\N{CJK UNIFIED IDEOGRAPH-4F1A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 25216 (\N{CJK UNIFIED IDEOGRAPH-6280}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 34899 (\N{CJK UNIFIED IDEOGRAPH-8853}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12475 (\N{KATAKANA LETTER SE}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12511 (\N{KATAKANA LETTER MI}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12490 (\N{KATAKANA LETTER NA}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12540 (\N{KATAKANA-HIRAGANA PROLONGED SOUND MARK}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 24195 (\N{CJK UNIFIED IDEOGRAPH-5E83}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 21578 (\N{CJK UNIFIED IDEOGRAPH-544A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 26082 (\N{CJK UNIFIED IDEOGRAPH-65E2}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23384 (\N{CJK UNIFIED IDEOGRAPH-5B58}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 39015 (\N{CJK UNIFIED IDEOGRAPH-9867}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23458 (\N{CJK UNIFIED IDEOGRAPH-5BA2}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 32057 (\N{CJK UNIFIED IDEOGRAPH-7D39}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 20171 (\N{CJK UNIFIED IDEOGRAPH-4ECB}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)

png

Reading the results

The optimal solution is to avoid allocating the entire amount to the most efficient measures ever, but to spread out the diminishing returns. In practice, restrictions such as the number of negotiations that salespeople can handle, brand maintenance, and contracted limits are also added to the requirements. Since the coefficient is an estimate, sensitivity analysis is performed without treating the solution as absolute.

No.008: How to Face Uncertainty

Meaning in Practice

Even if the average order rate is the same, the reliability difference between 10 observations and 1,000 observations can be different. If you set your budget based solely on point estimates, you risk overestimating initiatives with limited data.

Approach to Analysis and Modeling

Orders are considered to be Bernoulli trials, and the standard error is

$SE(\hat p)=\sqrt{\frac{\hat p(1-\hat p)}{n}}$

This approximates the situation. The 95% interval is set to $\hat p\pm1.96SE$ (for small specimens, Wilson intervals are appropriate).

Check with Python

uncertainty = df.groupby("channel").agg(wins=("orders", "sum"), trials=("opportunities", "sum"))
uncertainty["rate"] = uncertainty["wins"] / uncertainty["trials"]
uncertainty["se"] = np.sqrt(uncertainty["rate"]*(1-uncertainty["rate"])/uncertainty["trials"])
uncertainty["lower95"] = (uncertainty["rate"]-1.96*uncertainty["se"]).clip(0, 1)
uncertainty["upper95"] = (uncertainty["rate"]+1.96*uncertainty["se"]).clip(0, 1)
display(uncertainty.round(3))

fig, ax = plt.subplots()
err = np.vstack([uncertainty["rate"]-uncertainty["lower95"], uncertainty["upper95"]-uncertainty["rate"]])
ax.errorbar(uncertainty.index, uncertainty["rate"], yerr=err, fmt="o", capsize=5)
ax.set_title("Win rate with approximate 95% intervals")
ax.set_xlabel("Channel"); ax.set_ylabel("Win rate")
ax.grid(True, alpha=.3); ax.tick_params(axis="x", rotation=15)
fig.tight_layout(); plt.show()

	wins	trials	rate	se	lower95	upper95
channel
WebAdvertisement	54	405	0.13	0.02	0.10	0.17
Exhibition	123	508	0.24	0.02	0.20	0.28
Technical Seminar	87	367	0.24	0.02	0.19	0.28
Introducing Existing Customers	92	262	0.35	0.03	0.29	0.41

/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 24195 (\N{CJK UNIFIED IDEOGRAPH-5E83}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 21578 (\N{CJK UNIFIED IDEOGRAPH-544A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 23637 (\N{CJK UNIFIED IDEOGRAPH-5C55}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 31034 (\N{CJK UNIFIED IDEOGRAPH-793A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 20250 (\N{CJK UNIFIED IDEOGRAPH-4F1A}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 25216 (\N{CJK UNIFIED IDEOGRAPH-6280}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 34899 (\N{CJK UNIFIED IDEOGRAPH-8853}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 12475 (\N{KATAKANA LETTER SE}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 12511 (\N{KATAKANA LETTER MI}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 12490 (\N{KATAKANA LETTER NA}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 12540 (\N{KATAKANA-HIRAGANA PROLONGED SOUND MARK}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 26082 (\N{CJK UNIFIED IDEOGRAPH-65E2}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 23384 (\N{CJK UNIFIED IDEOGRAPH-5B58}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 39015 (\N{CJK UNIFIED IDEOGRAPH-9867}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 23458 (\N{CJK UNIFIED IDEOGRAPH-5BA2}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 32057 (\N{CJK UNIFIED IDEOGRAPH-7D39}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/var/folders/3y/fmw40k0x78xblvb3gkcyvy1h0000gn/T/ipykernel_1522/1893485025.py:14: UserWarning: Glyph 20171 (\N{CJK UNIFIED IDEOGRAPH-4ECB}) missing from font(s) DejaVu Sans.
  fig.tight_layout(); plt.show()
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 24195 (\N{CJK UNIFIED IDEOGRAPH-5E83}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 21578 (\N{CJK UNIFIED IDEOGRAPH-544A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23637 (\N{CJK UNIFIED IDEOGRAPH-5C55}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 31034 (\N{CJK UNIFIED IDEOGRAPH-793A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 20250 (\N{CJK UNIFIED IDEOGRAPH-4F1A}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 25216 (\N{CJK UNIFIED IDEOGRAPH-6280}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 34899 (\N{CJK UNIFIED IDEOGRAPH-8853}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12475 (\N{KATAKANA LETTER SE}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12511 (\N{KATAKANA LETTER MI}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12490 (\N{KATAKANA LETTER NA}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 12540 (\N{KATAKANA-HIRAGANA PROLONGED SOUND MARK}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 26082 (\N{CJK UNIFIED IDEOGRAPH-65E2}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23384 (\N{CJK UNIFIED IDEOGRAPH-5B58}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 39015 (\N{CJK UNIFIED IDEOGRAPH-9867}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 23458 (\N{CJK UNIFIED IDEOGRAPH-5BA2}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 32057 (\N{CJK UNIFIED IDEOGRAPH-7D39}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)
/Users/hiroshi/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:170: UserWarning: Glyph 20171 (\N{CJK UNIFIED IDEOGRAPH-4ECB}) missing from font(s) DejaVu Sans.
  fig.canvas.print_figure(bytes_io, **kw)

png

Reading the results

We should not definitively talk about small ranking differences in channels where the intervals overlap. It is reasonable to set a training budget for new initiatives and redistribute data after accumulating data. Depending on risk tolerance, some judges prioritize the lower boundary of the range over expected values.

No.009: The Role of Simulation

Meaning in Practice

A business plan requires not only “how much sales are worth,” but also “what the probability of losses is” and “how far it could drop in bad cases.” Simulations combine multiple uncertainties and present the results as a distribution.

Approach to Analysis and Modeling

We use the Monte Carlo method, which represents the number of deals with the Poisson distribution, the number of orders as the binomial distribution, and the unit price of the deal as the log-normal distribution. Distribution assumptions are verified using historical data and business knowledge.

Check with Python

sim_rng = np.random.default_rng(42)
n_sim = 10_000
sim_opportunities = sim_rng.poisson(55, n_sim)
sim_orders = sim_rng.binomial(sim_opportunities, 0.24)
sim_prices = sim_rng.lognormal(np.log(720), 0.25, n_sim)
sim_profit = sim_orders * sim_prices * 0.36 - 650
q = np.quantile(sim_profit, [0.05, 0.50, 0.95])
print(f"5%point: {q[0]:,.0f} ten_thousand_yen / median: {q[1]:,.0f} ten_thousand_yen / 95%point: {q[2]:,.0f} ten_thousand_yen")
print(f"deficit probability: {(sim_profit < 0).mean():.2%}")

fig, ax = plt.subplots()
ax.hist(sim_profit, bins=45, color="steelblue", alpha=.8)
for value, label in zip(q, ["5%", "Median", "95%"]):
    ax.axvline(value, linestyle="--", label=label)
ax.set_title("Monte Carlo distribution of contribution profit")
ax.set_xlabel("Contribution profit (10k JPY)"); ax.set_ylabel("Frequency")
ax.grid(True, alpha=.3); ax.legend(); fig.tight_layout(); plt.show()

5% point: 10.44 million yen / Median: 26.83 million yen / 95% point: 54.04 million yen
Probability of deficit: 0.04%

png

Reading the results

By showing not only the median but also a 5% point and a probability of loss, contingency funds and withdrawal conditions can be decided in advance. Simulation is not a device for predicting the future, but rather a device that visualizes risks if assumptions are true.

No.010: Building a Python Environment

Meaning in Practice

If analysis only runs on the person in charge’s PC, it will not become established as a decision-making process. Record and reproduce the environment, dependency libraries, random numbers, data definitions, and execution order.

Approach to Analysis and Modeling

The Notebook is suitable for exploration and explanation. On the other hand, production operations require data validation, functionalization, testing, version control, and regular execution. The minimum condition is that cells can be redone in order from the top.

Check with Python

environment = pd.DataFrame({
    "item": ["Python", "NumPy", "pandas", "matplotlib", "random seed", "external data"],
    "value": [sys.version.split()[0], np.__version__, pd.__version__, matplotlib.__version__, "42", "None"]
})
display(environment)
assert len(df) == 24 * 4
assert df[["spend", "leads", "orders", "revenue"]].notna().all().all()
assert (df[["spend", "leads", "orders", "revenue"]] >= 0).all().all()
print("Data Quality Check: OK")

	item	value
0	Python	3.11.9
1	NumPy	1.26.4
2	pandas	2.2.2
3	matplotlib	3.9.2
4	random seed	42
5	external data	None

Data Quality Check: OK

Reading the results

The version and seed were clearly displayed, and the quality check was passed. In practice, dependencies are locked and the original data update date, column definitions, handling of personal information, and approval history are recorded.

Practical Implications Seen Through Target Exercise

What all 10 have in common is that Designing Decisions comes before the amount of data. Break down KGI into funnel KPIs, convert probabilities and amounts into expected value, suspect confounding, allocate under constraints, and report uncertainty as a breadth. This shifts the meeting focus from “which numbers are correct” to “which assumptions and which actions to choose.”

What is necessary for practical implementation

Defining management challenges and decision frequency
Connecting customers, projects, initiatives, and orders with a common ID
Turning KPI formulas, responsible persons, and update frequency into a data dictionary
Decide on control groups, duration, success criteria, and withdrawal conditions before implementing measures
Evaluate not only model accuracy but also gross profit, man-hours, and risk
Review assumptions and results monthly and redistribute them in small segments.

Conclusion

Marketing science is not about using complex methods, but about making questions measurable, creating comparable options, and learning from the results. In this notebook, we examined the process of making decisions through hypothetical data from manufacturing B2B to aggregation, KPIs, models, causality, optimization, uncertainty, and simulation.

Consultations for Corporations

At Suri Kobo, we support everything from problem organization to analysis, implementation, and human resource development in the manufacturing industry, covering marketing KPI design, customer and project data infrastructure, demand forecasting, measure effectiveness verification, and budget allocation optimization. Even when data is not yet fully prepared, you can consult us starting with decision-making and organizing collection design.

📩 Contact Us: surikobo.co.jp/contact
Please feel free to consult us first.

Introduction to Marketing Science in Manufacturing | Learning KPI, Causality, and Budget Optimization with Python

From ‘Intuition’ to ‘Decision-Making’ in Manufacturing B2B Marketing: 10 Steps into Marketing Science

Title & Overview

Introduction: Practical Challenges in Manufacturing Covered in This Article

Common situations on site

Why is this issue so difficult to judge?

Overview of Exercise covered this time

Preparing the Python environment

Creation of Fictional Data

No.001: What is Marketing Science?

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.002: Why Mathematics Is Necessary for Marketing

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.003: What is Data-Driven Management?

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.004: The Concept of KPIs

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.005: Decision-Making and Models

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.006: Correlation and Causality

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.007: The Concept of Optimization

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.008: How to Face Uncertainty

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.009: The Role of Simulation

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

No.010: Building a Python Environment

Meaning in Practice

Approach to Analysis and Modeling

Check with Python

Reading the results

Practical Implications Seen Through Target Exercise

What is necessary for practical implementation

Conclusion

Consultations for Corporations

Related Posts

Changing Manufacturing Decision-Making with Data and AI | 10 Marketing Science Practices

Optimizing Marketing Investment in Manufacturing | From Attribution to AI and Digital Twins

Recommendation Systems and Mathematical Optimization in Manufacturing | Practicing A/B Testing, LTV, Inventory, and Production Planning with Python

Introduction to Recommendation Systems in Manufacturing | From Collaborative Filtering to Online Learning: Hands-on Use of Python

Practicing Price Optimization in Manufacturing with Python | From Price Elasticity to Dynamic Pricing and Reinforcement Learning