Level 9 · Boss: the marketing dashboard

A dashboard is a grid of Axes

plt.subplots(rows, cols) creates one Figure with a grid of Axes and returns them as a 2-D array:

fig, axs = plt.subplots(2, 2, figsize=(12, 8))
axs[0, 0].plot(...)     # top left
axs[0, 1].barh(...)     # top right
axs[1, 0].bar(...)      # bottom left
axs[1, 1].hist(...)     # bottom right
fig.suptitle("One title for the whole figure")
plt.tight_layout()      # stops titles and labels from overlapping

Rows and columns count from 0. pandas plots can go into a chosen cell with ax=: monthly.plot(ax=axs[0, 0]). When two charts should be compared by height, create the grid with sharey=True so they use the same scale.

The three tables

Spend per month. marketing_spend has one row per day and channel. A month column and a pivot give months down and channels across. In 2025 paid spend ran at about $21,000 to $28,000 a month from January to October and jumped to $47,475.61 in November.

Conversion by source. Every session has a source and a boolean converted. Summing a boolean counts the True values, and the mean of a boolean is the share that’s True. In 2025, 9.6% of email sessions converted, 5.4% of direct, 1.8% of organic, 1.5% of paid and 0.7% of social.

Cost per conversion. Divide each channel’s spend by the conversions its sessions produced. The channel names in marketing_spend (email, paid, social) match the source values in web_sessions, so after renaming source to channel the two small tables join one to one. Organic and direct traffic have no spend, and the inner merge leaves them out.

cpc = spent.merge(conv, on="channel")
cpc["cost_per_conversion"] = cpc["spend"] / cpc["conversions"]

Email: $38,322.22 for 187 conversions, $204.93 each. Paid: $340,778.33 for 55, $6,195.97 each. Social: $131,540.77 for 19, $6,923.20 each. Sessions aren’t linked to orders in this data, so a conversion here is a converted session, not revenue. A careful slide says that in a footnote.

Saving the figure

In a notebook you’d save the dashboard for the slide with fig.savefig("marketing_2025.png", dpi=150, bbox_inches="tight"). Call it before plt.show(): after show, the figure is closed and the saved file comes out blank. Here the page shows your figure automatically, so you don’t need either call.

Common mistakes

• Comparing spend and conversions from different periods (2025 spend against all-years sessions).
• Dividing spend by all sessions: that’s cost per visit, a different and much smaller number.
• Forgetting plt.tight_layout(), so titles of the lower charts overlap the upper axis labels.
• Writing axs[0][0] and axs[0, 0] interchangeably is fine, but axs[1] alone is a whole row.