Marketing KPIs: What to Measure and How

What's the difference between a marketing KPI and an ordinary metric?

A metric is any measurable number. A KPI is a metric tied to a priority goal, with an owner, a calculation rule, a time window and a decision threshold. Every KPI is a metric. Not every metric is a KPI.

Digital marketing looks easy because it's measurable. In truth, piling everything measurable onto a panel makes deciding harder. KPI selection should start from "which decision must we make?", not "which data do we have?"

How are marketing KPIs chosen?

A working KPI set can have four layers: the business outcome, the marketing outcome closest to it, the signal that shows the process early, and the threshold protecting quality. Don't give them all the same weight. The core outcome is one or two; the rest are for diagnostics.

1. Write the business goal and the period

"Growing the brand" is not a measure. For example: "Increase fitting new customers in 90 days, keeping acquisition cost within what one customer's margin and payback calculation allows." If the goal has no timeframe, you can compare monthly clicks with annual revenue in the same table and comfortably draw the wrong conclusion.

2. Split the goal across the sales funnel's stages

Revenue is a lagging outcome. Until you reach it, track fitting leads, meetings, proposals and sales transitions separately. But to make a funnel stage a KPI, there must be a real decision at that stage. If the team changes nothing, keep it as a diagnostic metric.

3. Prepare a passport for each KPI

The name can stay the same while the calculation changes. "Conversion" can be forms / sessions on one panel and new customers / inquiries on another. The KPI passport must state the definition, formula, data source, scope, exclusions, owner, refresh frequency and decision threshold.

4. Take the baseline from your own data

An "average CTR" found online is not your target. When the channel, device, offer, price, region and measurement rules change, the average's meaning changes too. First calculate the recent period with the same definition. Choose the target with the marketing strategy, the margin and the team's capacity.

5. Write a stop threshold beside the target

Writing only "12 customers" doesn't say when the budget will be held. For example: pause the audience trial when the fit rate falls below a set threshold; freeze sends if complaints rise; don't raise the budget if CAC exceeds the maximum the business approved. Thresholds must be confirmed by the business's own economics and risk, not arbitrarily.

6. Match the reporting frequency to the decision's speed

Ad spend can be visible daily, while a B2B contract can lag by months. Staring at the revenue panel every morning creates no information. Keep daily monitoring, weekly optimisation and the monthly business decision separate.

How are the core marketing KPI formulas calculated?

The formula can be simple; the scope can't. Use the same currency, the same period and the same customer definition. Otherwise a correct division will give a wrong business answer.

In a CAC calculation you can't divide only ad spend and call the result "customer acquisition cost." If the salesperson's time, the tools used and agency fees are in scope, keep that the same across all periods. If you measure only media spend, name the metric accordingly.

ROAS can show more revenue while harming the business. If the product margin is low, refunds are high or the revenue hasn't been collected yet, a 4:1 ratio isn't automatically a good result. Alone, the number defends the ads; joined with finance data, it defends the business.

Why do channel and attribution numbers differ?

A customer can see the ad, search the brand name on Google two weeks later, then return directly to the site and buy. Which channel "brought the sale"? The answer depends on the attribution rule you use.

Google Analytics' current attribution documentation shows credit is split across touchpoints by model, with several models available in reports. A model is an accounting rule; it doesn't prove the sole cause of the customer's decision.

Even within the same platform the scope changes. According to Google's User acquisition vs Traffic acquisition comparison, the first report focuses on a new user's first source and the second on a new session's source. Metrics from reports with different scopes can't be compared directly. When connecting site events to real sales outcomes through the CRM, both systems must keep the same date rules and a permitted identifier.

That connection does not mean sending emails and phone numbers to Google Analytics. Google's official policy on personally identifiable information prohibits passing data like names, emails and personal phone numbers into Analytics. Such data must not remain in URLs and campaign parameters, and personal data from form fields must not flow into analytics events.

Search KPIs carry an illusion of completeness too. Search Console's current explanation of Performance data notes some queries are anonymised for privacy and the table doesn't show every row. Don't treat the query list as full market demand. Read it together with content, sales language and site behaviour.

A marketing KPI example: an Azerbaijani B2B company

Suppose a hypothetical Azerbaijani B2B services company spent 18,000 AZN on sales and marketing in a 90-day trial. During the period, 120 inquiries, 72 fitting leads, 36 real sales opportunities and 12 new customers were recorded. These are not market norms; they're examples of how a formula becomes a decision.

If the company had pre-approved a maximum CAC of 1,800 AZN from its own economics, 1,500 AZN is within bounds. That's not an immediate order to raise the budget. The twelve customers' payments, margins, cancellations and service capacity must still be checked. The target too comes from the business's financial model, not from an article.

Showing the 120 inquiries as the headline victory on the panel is comfortable. The more useful reading is investigating the source of the 48 unfit inquiries and the reason only 12 of 36 opportunities became sales. If the audience selection, the sales response and the offer are changed in the same week, nobody will know which fix delivered the result.

How do AI, Google and the legal boundary affect KPIs?

AI can summarise the panel, flag anomalous changes and sort sales notes into initial themes. It cannot invent unknown revenue, choose the loss reason in the customer's place, or make two numbers with different scopes comparable. Beside an automatic summary, the formula, the source and a final human check must remain.

If AI is used to produce content for email, forms and ads, Google's generative AI guidance doesn't count the tool as an automatic penalty cause. Creating many pages without extra value for users, though, can violate the scaled content abuse rule. The people-first guidance foregrounds original information, complete answers, accuracy and authorship. A click target doesn't cancel those quality boundaries.

If you join form, phone, email and behavioural data at person level, Azerbaijan's Law "On Personal Data" requires specific review for collection, processing, protection, third-party disclosure and cross-border transfer. "More precise attribution" is not a legal basis for limitless collection of personal data.

When turning a KPI result into an advertising claim, check Articles 6–8 of the Law "On Advertising" on unfair, inaccurate and hidden advertising. Don't present a hypothetical calculation as a real customer result, or attributed revenue as proof of causation. This article is not legal or financial advice; get legal and financial review for your specific data flow, campaign and economic model.