Comparisons and Benchmarks

Measuring Performance Against a Benchmark

Absolute numbers mean nothing on their own. “€150,000 in revenue this month” tells you nothing until you know what to compare it to. Is that better than last month? Better than the target? Better than the competition? The same figure can be excellent news or a warning sign of a slump, depending on the benchmark you choose. This lesson teaches you how to structure rigorous comparisons with ChatGPT.

Types of Comparisons

Before writing a single prompt, identify the benchmark that gives your number meaning. Three families cover most business monitoring needs, and most solid analyses combine at least two: a region can grow year-over-year while still falling short of its budget.

Time-Based Comparison

Month-over-month comparison with the previous fiscal year is the most readable format for a committee because it neutralizes seasonality: January is compared to January, never to December.

Create a table comparing each month of 2026 with the same month in 2025.
Columns: Month, Revenue 2025, Revenue 2026, Absolute Change, % Change.
Highlight months with a decline.

However, a single month remains volatile: one large order signed on the 2nd of the following month instead of the 30th is enough to flip the sign of the variation. Year-to-date (YTD) cumulative totals correct this flaw by absorbing these billing shifts.

Compare cumulative revenue as of end-March 2026 with cumulative revenue
as of end-March 2025. Plot both cumulative curves on the same chart.

Cumulative data is more reliable than isolated months: it smooths out monthly fluctuations and shows the overall trajectory. On the chart, the vertical gap between the two curves reads directly as the lead or lag compared to last year, and its evolution tells you whether that gap is widening or narrowing.

Entity-to-Entity Comparison

Comparing five regions on a single metric produces a ranking that is immediately contestable, because the region gaining volume often loses average basket size—and its director will not fail to point that out. A multi-metric ranking with a global rank defuses the dispute.

Compare the 5 regions on the following metrics:
Total Revenue, Number of Customers, Average Basket Size, Conversion Rate.
Rank them from best to worst on each metric.
Add a global rank (average of the ranks).

When you are looking for a profile rather than an order, a radar chart becomes more meaningful than a ranking.

Create a radar chart comparing the Paris, Lyon, and Marseille regions
on 5 axes: Revenue, Volume, Average Basket, Retention Rate, Margin.
Normalize values between 0 and 1.

The radar chart shows at a glance what profile each entity has, even if the metrics have different units. A region whose polygon is small but regular is not managed like a region distorted on a single axis: the first needs volume, the second needs rebalancing.

Comparison Against Targets

The budgetary benchmark is provided directly in the prompt, since it usually does not appear in the sales file—it lives in a management control spreadsheet that no one has exported.

Here are the targets by region (in thousands of euros):
Paris: 500, Lyon: 300, Marseille: 250, Bordeaux: 200, Lille: 150.
Compare with actual revenue and calculate the achievement rate in %.
Grouped bar chart: Target vs Actual.

The global achievement rate answers the question “Where do we stand?” When the question becomes “Where does the annual gap come from?”, you need a view that attributes the shortfall to each month.

Calculate the gap between actuals and budget for each month.
Display a waterfall chart showing how each month contributes
to the total gap.

The waterfall chart is ideal for performance reviews: it shows precisely where gaps widened, and distinguishes a delay concentrated in one accident-prone month from a regular drift that calls for a completely different reaction.

Cohort Analysis

Cohorts compare groups formed at different times. The principle is simple: instead of mixing all your customers, you track those acquired in January separately from those in February, and you observe them at the same age rather than on the same date.

Group customers by acquisition month (cohort).
For each cohort, calculate the average revenue per customer
at 1 month, 3 months, 6 months, and 12 months after acquisition.
Display the result as a heatmap.

This analysis reveals whether your recent customers spend more or less than older ones at the same stage, a key indicator of your business health. A regular degradation from cohort to cohort signals poor acquisition quality, and this signal often appears while total revenue continues to grow quietly—hence its value.

Benchmarks and Percentiles

Positioning yourself within a distribution is better than comparing yourself to an average, which a few very large deals can easily distort. In a sales team, a single exceptional contract can place half the sellers “below average” without anyone underperforming. However, individual performance data is personal: replace names with identifiers before importing, or work on team aggregates.

For each salesperson, indicate which quartile they fall into
in terms of Revenue, Number of Deals, and Conversion Rate.
Who is above the 75th percentile on all 3 metrics?

When the entities being compared are not the same size, a base-100 index puts everyone on a common scale.

Create a performance index for each product:
100 = category average.
A product at 120 performs 20% better than average.
Rank products by descending index.

The index allows you to compare entities of different sizes on a common scale.

Visualizing Comparisons

Each type of comparison has its preferred chart. Grouped bars work well for confronting two or three series side by side; beyond that, reading becomes confusing and the message gets diluted. Diverging bars showcase positive and negative gaps around a benchmark, making them the natural tool for budget tracking. The bullet chart compares an actual value to its target with minimal clutter, a precious quality for a KPI dashboard. Finally, the heatmap displays cross-performance between entities and metrics, where a table of numbers would require a line-by-line read.

Create a diverging bar chart showing the budget variance
for each region. Green for regions above budget,
red for those below. Sort from best to worst.

Your Turn

Your first benchmark.

1. Choose a key indicator and its three benchmarks: last year, other entities, the target
2. Compare the isolated month and the year-to-date cumulative total
3. Rank your entities on multiple metrics, with a global rank
4. Calculate the target achievement rate and the month-by-month gap in a waterfall chart
5. Summarize in three sentences what each benchmark teaches you

Official sources: Extracting Insights with ChatGPT Data Analysis · Data analysis with ChatGPT (OpenAI Help Center)

Key Takeaways

• A number alone has no meaning—always compare it to a benchmark
• Year-to-date cumulative totals are more reliable than isolated months
• Radar charts compare multidimensional profiles at a glance
• Cohort analysis reveals the evolution of your customer quality over time
• Use indexes (base 100) to compare entities of different sizes