Data Analytics and Business Intelligence

Most disappointing analysis fails in the first conversation, when someone asks you to "have a look at the numbers" and you start pulling data before anyone has said what decision the work is for.

This chapter teaches three habits. Turn a vague request into a decision question with an owner, real options and a deadline. Define each metric so precisely that two analysts working apart would get the same number. Agree the scope, data sources and shape of the answer before you start. These habits save days of wasted work and make your findings far more likely to be used. They work in Excel, SQL or Power BI, whether you are a student on a placement or a manager briefing an analyst.

"Can you look into customer complaints?" is a request. It names a topic but not what anyone will do with the answer. A decision question names a choice a specific person must make, the options and the date. "Should the head of customer service add four agents to the evening shift from next quarter?" tells you what to measure and when the work stops being useful.

Google's Ask Questions to Make Data-Driven Decisions course teaches SMART questions: specific, measurable, action-oriented, relevant and time-bound. If no answer would change what someone does, it is not yet a good question.

Step five gives you your key metric and threshold before you open any data. In a 2019 McKinsey podcast, former partner Charles Conn describes forming an early "day-one answer" and testing it, rather than collecting everything and hoping a pattern appears. If the owner says "I would add agents if evening wait times are above ten minutes", you know exactly which figure to calculate and what counts as a yes.

Emma, an operations analyst at a food delivery start-up in Pune, is told: "Complaints are up. Find out why customers are unhappy." Complaints rose from 1,200 in July to 1,850 in August. That is 650 more, and 650 divided by 1,200 is 0.54, a 54% rise. Real, but not yet useful.

A 15-minute call reveals that the head of operations must decide by 15 October whether to fund 40 extra riders or a tool for restaurants to report preparation times. He would fund riders if most late orders had no rider nearby.

Before: "Why are customers unhappy?"

After: "Should the head of operations fund 40 extra riders before 15 October, or are late deliveries mainly caused by restaurant preparation time?"

Emma now compares two shares of late orders and ignores everything else.

Common mistake: starting with the data you already have. The data you can reach easily shapes the question you answer, and it is often not the one the business needs. Frame first, then look for data that fits the frame.

Ask three teams how many active customers there are and you may get three answers. Sales counts anyone who bought in 12 months, marketing anyone who opened an email in 90 days, finance anyone with a paid invoice this quarter. Each is correct by its own rule, and the meeting ends up arguing about data instead of the decision.

The test: if two analysts working apart with the same data would produce the same number, the definition is precise enough.

Tom, a commercial analyst at a gym chain in Leeds, is asked for October's churn rate. The chain started with 4,000 members, 500 joined, and there were 240 cancellations, 40 of them by new joiners. Another 60 starting members froze their membership.

A, starting members only: 200 divided by 4,000 is 5.0%. B, everyone who was a member: 240 divided by 4,500 is 5.3%. C, starting members with freezes counted as losses: 200 plus 60 is 260, and 260 divided by 4,000 is 6.5%.

None is wrong. The retention manager wants to know whether long-standing members are leaving faster than last year, so Tom agrees Definition A, reports freezes separately and checks last year used the same rule.

Common mistake: reusing a metric from an existing dashboard without reading its definition. A field called "revenue" might include tax, exclude refunds, or be recorded on order date rather than payment date. If there is no definition, write one and get it confirmed.

Scope is the boundary of your analysis: what is in, what is out, and why. Without it, every new idea becomes extra work and a two-day task turns into two weeks. This is scope creep. Agree the boundary in writing so any addition becomes a conscious choice with a cost.

Cover the population, time period and geography, and list what you will not answer. That list protects you when someone asks later why you did not look at something.

For each data source, check access, freshness and whether it holds the fields your definition needs. Then agree what the answer will look like: a yes or no with conditions, a number with a range, or a one-page recommendation. Sketch the final table with empty values and show the owner. You learn "I need it by region" on day one, not day ten.

Agree how precise the answer must be as well. A decision with a ten-minute threshold rarely needs an answer accurate to the second, and chasing extra precision costs days you may not have. If the deadline is tight, agree what a good-enough answer looks like now.

Together these make a one-page analysis brief. Get it approved before the main work starts.

Laura, a business analyst at a microfinance bank in Lagos, is asked to "analyse loan defaults". The loan book holds 12,400 loans across five products and nine years, roughly three weeks of work.

She learns the credit committee meets in eight days to decide whether to cap first-time small business loans at 500,000 naira. She scopes the work to those loans from the last two years, 2,150 in total, and sketches a table of 90-day default rate by four loan size bands. The chair agrees. Three weeks becomes four days. The spare days go on checking the data properly, which makes the final answer easier to trust.

Common mistake: saying yes to "while you are at it" without resetting the deadline. When a request grows, say what it costs in time and let the owner choose between the addition and the date.

A strong analyst treats the first message as the start of a conversation, not a set of instructions. Before touching data, they find the decision and its owner, confirm the deadline, and agree which definition of each metric counts. They sketch the answer and get it approved. This takes less than an hour and often saves days. It also means the final number is trusted, because everyone agreed the rules before seeing the result.