Skip to content

What should you measure first, second and third in marketing?

You have Google Analytics, maybe Search Console and a tag manager, and a dashboard full of numbers that nobody trusts or acts on.

By Adnan GourijaHead of Search Marketing at Webtec · Reviewed 2026-10-17

Add us as a preferred source on Google

Decision · 12 min read

The route: 5 moves, in this order

  1. 1Define one business outcome and verify its proof event
  2. 2Find out where those outcomes come from, at channel level
  3. 3Measure the cost and the quality of an outcome
  4. 4Add leading indicators for each channel
  5. 5Only then look at diagnostics

The short answer

Measure in five layers, in order. First, one business outcome and a proof event that fires correctly. Second, which channels produce that outcome. Third, what an outcome costs and how many become customers. Fourth, leading indicators per channel. Fifth, diagnostics like speed and engagement. Ignore everything that cannot change a decision this month. Each layer only means something once the one before it is trustworthy.

The rule

Outcome before activity

Never measure an activity until you can measure the outcome it is supposed to produce. A metric earns a place on your dashboard only if you can name the decision it would change. Everything else is reference data, not reporting.

The order

5 moves. Do them in this order.

  1. 1

    Move 1 of 5

    Define one business outcome and verify its proof event

    You are here when: You cannot finish the sentence "marketing worked this month because we got N of ___" without hesitating, or you have more than three events marked as important.

    Why now

    Every later metric is a ratio or a breakdown of this one. If the outcome is wrong, duplicated or missing, all of them inherit the error. Analytics treats a key event as an action especially important to the business. Any event you collect can become one, so the tool will not warn you when you pick badly.[1] Choosing is your job, and verifying is the second half of it. A proof event that fires twice or not at all is worse than none, because it looks like data.

    Do this

    • Write one sentence naming the outcome: a booked call, a paid order, a qualified enquiry. Pick the closest event to revenue that you can capture reliably.
    • Name the proof event: the exact page, button or confirmation that means the outcome happened. A thank-you page you can reach without submitting a form is not proof.
    • Mark only that event, and at most two supporting ones, as key events in Analytics.[1]
    • Test the event in your tag manager's preview mode.[7] It shows which tags fired, in what order and which trigger caused them. Submit a real test and confirm one event, not zero or two.
    • Compare the count with your source of truth, such as the CRM, the booking tool or the order system, for the same seven days.
    Done whenFor a full week, proof events in Analytics are within about 10 percent of real outcomes in your CRM or order system. You can explain the gap.
  2. 2

    Move 2 of 5

    Find out where those outcomes come from, at channel level

    You are here when: The proof event is trustworthy, and your reports lead with sessions, users or pageviews instead of outcomes by source.

    Why now

    Volume without origin cannot guide spending. Once the outcome is reliable, the next question is which channels produce it. Analytics attributes key events to touchpoints using a chosen model.[8] All models exclude direct visits from credit unless the whole path is direct.[8] That means your channel numbers are an allocation rule, not a fact, so you need to know the rule before you compare channels.

    Do this

    • Open the acquisition report and filter to the proof event. Read outcomes by default channel group, not sessions.
    • Tag every paid, email and social link you control with consistent campaign parameters, so they do not collapse into direct or referral.
    • Check how AI assistants appear. Analytics now reports visits from assistants such as ChatGPT, Gemini and Claude in their own channel, so confirm yours is not hidden in referral.[10]
    • Record the attribution model in your Analytics settings and write it at the top of your report.[8]
    • Ask every new lead "how did you hear about us?" in the form or on the call, and compare the answers with the report once a quarter.
    Done whenYou can state, for the last 90 days, the share of outcomes by channel, the attribution rule behind it, and how large the "direct" and "unknown" slice is.
  3. 3

    Move 3 of 5

    Measure the cost and the quality of an outcome

    You are here when: You know which channels produce outcomes, but you cannot say which ones produce customers or what a customer costs.

    Why now

    Outcomes are not equal. A channel that sends many cheap, unqualified leads looks excellent until the sales team ignores them. Cost and quality turn a count into a decision. Our rule: if you cannot compute cost per qualified outcome and outcome-to-customer rate by channel, you are still reporting activity. The formulas are in conversion rate math and marketing KPIs, so this page does not repeat them.

    Do this

    • Define "qualified" in writing with sales: three or four criteria that a stranger could apply to a lead.
    • Pass the lead source into the CRM when the lead is created, so the channel travels with the person.
    • Calculate cost per qualified lead and lead-to-customer rate for each channel, monthly. Include all costs: media, tools, agency fees and your own time.
    • Rank channels by cost per customer once you have at least 20 customers in the period. Below that, rank by cost per qualified lead and mark the ranking provisional.
    Done whenFor each channel you can write one line: "N qualified leads at a cost of X, Y percent became customers." Sales agrees with the qualified count.
  4. 4

    Move 4 of 5

    Add leading indicators for each channel

    You are here when: Outcome, source, cost and quality are stable, and you now need to know whether next quarter will be better or worse before the revenue shows up.

    Why now

    Outcomes lag. Search and AI visibility can take months to compound, so waiting for the outcome means reacting late. A leading indicator is a signal that moves before the outcome and has a plausible causal link to it. Search Console impressions count when a link was seen or potentially seen, which is why they lead clicks.[4] The report also supplies clicks, click-through rate and average position.[4] Pick one or two per channel, and check once a quarter that they still predict outcomes.

    Do this

    • Search: track non-brand impressions and clicks for the pages that lead to your proof event, from Search Console.[4]
    • AI answers: track whether you are cited or mentioned for a fixed list of 20 to 30 buyer questions, run on a schedule. See measuring AI visibility for the method.
    • Paid: track qualified-lead cost trend and impression share, not clicks.
    • Email and outbound: track reply and booking rate per 100 delivered messages.
    • Review each indicator against outcomes after two quarters. Drop any that did not move before outcomes did.
    Done whenEach active channel has one or two leading indicators, each with a baseline, and you have written down why you believe each one precedes outcomes.
  5. 5

    Move 5 of 5

    Only then look at diagnostics

    You are here when: The first four layers are in place, and an outcome rate or a leading indicator looks weak without an obvious cause.

    Why now

    Diagnostics explain a problem you have already seen in an outcome. Used earlier, they send you off fixing things that are not the constraint. Core Web Vitals are real thresholds. Good means a largest contentful paint of 2.5 seconds or less and an interaction to next paint of 200 milliseconds or less.[5] It also means a layout shift of 0.1 or less, at the 75th percentile.[5] They matter when a page that should convert is failing them. They are not a goal in themselves.

    Do this

    • Start from the weakest step in your funnel and open only the diagnostics that bear on that step.
    • Use field data from real visitors first. Lab tests run on one device and network and cannot measure interaction delay, so they are a debugging aid, not a verdict.[6]
    • Read engagement rate as a hint: a session is engaged if it lasts over 10 seconds, has a key event or has two or more page views.[2]
    • Change one thing, wait for enough outcomes to compare, and record the result.
    Done whenEach diagnostic you review is attached to a named outcome problem, and you can say what you will change if it is bad.

Decision finder

What to measure first, by business type

Business type

Local service business

Proof event

Booked appointment or tracked call

Channel view that matters

Map listing, search, paid, referrals

Quality measure

Calls that became jobs

Skip for now

Followers, bounce rate, keyword rankings by town

Evidence

Why the order matters more than the tools

Most teams measure in the order the tools present data. Analytics opens on users and sessions, Search Console opens on clicks and impressions, and the speed tool opens on a score. So the first numbers a team learns to read are the ones furthest from revenue. That is how a business ends up with a dashboard it cannot act on.

The order we recommend runs the other way, from the business outward. Layer one is the outcome. Layer two splits it by source. Layer three prices it and checks its quality. Layers four and five explain and predict. Each layer is a ratio or a breakdown of the one before, which is why a mistake at layer one corrupts everything above it.

This is a judgment about sequencing, not a study. The pieces are all documented, but we know of no controlled research that proves one measurement order beats another. What we do see, repeatedly, is that teams who skip layer one spend months arguing about numbers that were never reliable.

Evidence

Why vanity metrics are not harmless

A vanity metric is a number that rises easily and predicts nothing you care about. Followers, pageviews and raw impressions fit. They are not always useless, since an impression is a leading signal when it is tied to a channel that produces outcomes. They become a problem when they are the target.

The reason is a well-known pattern called Goodhart's law. When a measure becomes a target, it stops being a good measure.[11] Economist Charles Goodhart stated the original idea in 1975, and anthropologist Marilyn Strathern later gave it the form people quote today.[11] In marketing it looks like a team optimizing for pageviews with content that attracts readers who never buy. The number goes up and the business does not.

The practical defense is the rule on this page. Anything you target must be the outcome or a metric that has been shown to precede it, and everything else is watched but not rewarded. That keeps the incentive pointed at revenue.

  • Safe to watch, unsafe to target: pageviews, sessions, followers, impressions without a channel link.
  • Safe to target: qualified outcomes, cost per qualified outcome, outcome-to-customer rate.
  • Test for any metric: "If this doubled and revenue stayed flat, would anyone be surprised?" If not, it is a vanity metric for you.

Evidence

What your tools cannot tell you

Every number in this order has limits, and a good reader knows them. Analytics withholds some data under thresholds designed to protect individual users, mainly in reports that include demographics or search query information.[3] You cannot change the thresholds. You can sometimes avoid them by widening the date range.[3]

Attribution is the larger limit. Analytics offers data-driven and last-click models.[8] Direct visits get no credit unless the whole path is direct, and data-driven credit can be reassigned for up to 7 days after the outcome.[8] A channel that introduces people but never gets the last click will look weak. This is why we advise reading channel results alongside self-reported source, and why the reports guide explains why two tools disagree.

Search Console has its own definitions. An impression means a link was seen or potentially seen. A click generally means a visit to a page outside Google, and average position is the topmost position averaged across queries.[4] Position varies by user, so it is a trend line and not a promise.[4] Pages that appear in AI Overviews and AI Mode are included in the overall search traffic in the Performance report under the web search type.[9] It is not possible to separate them using that page alone, so treat any "AI Overview traffic" figure from a third party with care.

Evidence

Hard cases: low volume, long cycles and tracking you cannot trust

Low volume changes layers three and four. With fewer than about 30 outcomes a month, rates are noisy, and a single deal can swing cost per customer by half. In that case, read every lead by hand, record the real source in a spreadsheet, and treat ratios as direction, not proof. Our rule: below 20 customers in a period, do not rank channels on customer cost.

Long sales cycles break the link between this month's spend and this month's revenue. Use the earliest reliable proof event that sales will honor, usually a qualified opportunity in the CRM, and report it by first touch and last touch side by side. If the two disagree, the gap is itself the finding.

If you cannot make layer one reliable, stop and fix it before anything else. Common causes are duplicate tags, a thank-you page that loads without a submission, and forms that bypass the tag. A week of verified data beats a year of unverified data. The conversion definition guide and conversion tracking and attribution cover the setup.

Evidence

What to ignore for now, and the reason for each

Ignoring a metric is a decision, not a denial that it exists. Each item below can matter later. Each one is a distraction until the layers above it are in place.

  • Total traffic: it says nothing about which visitors matter. Reintroduce it as a denominator once outcomes by channel are reliable.
  • Bounce rate and engagement rate: engagement rate is the share of sessions that last over 10 seconds, include a key event or have two or more page views.[2] Bounce rate is its inverse.[2] A one-page visit that ends in a phone call counts as engaged only if you tracked the call. Use it as a diagnostic, never a target.
  • Average position and keyword rankings: position is a relative, averaged and user-dependent figure.[4] Watch non-brand impressions and clicks to outcome pages instead.
  • Core Web Vitals scores in isolation: they have pass thresholds,[5] but a green score on a page that gets no outcomes is not progress. Look at them when a high-intent page is failing, and use field data first.[6]
  • Social followers and reach: they rarely map to a proof event. Track only if social is a channel you can tie to outcomes.
  • Multi-touch attribution models and incrementality tests: valuable at scale, wasteful at 20 outcomes a month. Revisit when volume allows.

Evidence

The monthly review routine: 45 minutes, same order every time

The routine follows the five layers, which keeps the meeting short and stops it drifting into whichever chart looks interesting. Run it in the first week of the month on the previous full month, and keep the same one-page layout so trends are visible. The monthly report guide shows how to read the output.

Spend the first five minutes on layer one, because a broken proof event invalidates the rest. Compare the month's proof event count with the CRM or order system, and note any gap above 10 percent. If the gap is larger, stop the review and fix tracking. Never present results from a month you cannot reconcile.

Then move down the layers. Outcomes by channel take 10 minutes. Cost per qualified outcome and outcome-to-customer rate take 15, with sales in the room. Leading indicators against baseline take 10, and diagnostics get 5 only if a rate or indicator fell. End by writing three lines: what we learned, what we will change, and what we will check next month. A review that ends without a change is a status update.

  • Minute 0 to 5: reconcile the proof event with the source of truth.
  • Minute 5 to 15: outcomes by channel, with the attribution model and the unknown share stated.
  • Minute 15 to 30: cost per qualified outcome and lead-to-customer rate by channel, agreed with sales.
  • Minute 30 to 40: leading indicators against baseline; quarterly, test whether they still lead.
  • Minute 40 to 45: diagnostics if needed, then the three-line decision record.

Scroll through

6 mistakes that cost the most

  • 01

    Marking everything as a key event

    When ten events are marked as key events, every report shows a big, flattering number and none of them matches a real sale. It costs you the ability to say whether marketing worked. Mark one primary outcome and at most two supporting events, and keep the rest as ordinary events.[1]

  • 02

    Trusting a proof event you never tested

    A tag that fires on page load instead of on submission, or that fires twice, inflates results for months. It costs you every ratio downstream. Test in preview mode, submit a real test, and reconcile one week of counts against the CRM.[7]

  • 03

    Reading channel numbers as facts

    Channel results are an allocation under a stated model.[8] Direct visits earn no credit unless the path is all direct.[8] Cutting a channel because it loses on last click can remove the one that introduced your buyers. Record the model, add self-reported source, and compare both.

  • 04

    Optimizing a leading indicator after it stopped leading

    Impressions or rankings can rise while outcomes stay flat, especially if you chase terms that never convert. You pay for the activity and get nothing. Review each indicator against outcomes every two quarters and drop the ones that do not move first.

  • 05

    Starting with speed and engagement

    A team spends a quarter lifting a performance score on pages that were never the constraint. The cost is the quarter. Start from the weakest funnel step, and use diagnostics only on the page that step involves.

  • 06

    Building the report before agreeing the definition of qualified

    If marketing counts a lead and sales rejects half of them, cost per lead is a fiction and the two teams argue instead of deciding. Agree three or four qualification criteria in writing, and report on the qualified count.

Worked example

The rule applied

A services firm that cannot say what worked

1. The situation

Illustrative: a 12-person professional services company spends 6,000 a month across search, paid social and email. Analytics shows 4,000 sessions and 11 events marked as key events. The monthly report is eight pages long. Nobody can say how many customers marketing produced. These numbers are invented to show the logic.

2. The decision

The owner follows the order. Layer one: the only key events kept are a booked call, with a form submission and a tracked phone click as supporting events. A test submission fires twice, which a tag fix corrects. Layer two: outcomes by channel show search 60 leads, paid social 40, email 20. Layer three: sales agrees a qualified lead needs budget, authority and a four-month timeline. Qualified counts are search 18, paid social 4, email 8.

3. What it shows

Cost per lead looked similar across channels. Cost per qualified lead did not: the 6,000 spend split roughly evenly gives about 111 per qualified lead for search, 500 for paid social and 250 for email. With 6 customers from 30 qualified leads, the overall rate is 20 percent of qualified, 5 percent of all leads. The sample is small, so the firm marks the ranking provisional and keeps testing paid social with a narrower audience. It adds non-brand search impressions as a leading indicator for next quarter and ignores total traffic. The decision is clear without a single speed score.

Questions people ask

  1. 01

    What is the single most important metric to start with?

    The business outcome you can verify, such as a booked call, a paid order or a qualified enquiry. It is the metric every other number is a ratio of. Choose the event closest to revenue that your tools capture reliably, then test that it fires once per real outcome. If you only fix one thing in measurement this quarter, fix that. Everything else on a dashboard is a breakdown or an explanation of it.

  2. 02

    How many key events should I mark in Google Analytics?

    Keep to one primary outcome and at most two supporting events. Analytics lets any event become a key event, so the tool will not limit you.[1] More events dilute reports and make totals unlike any real count. A typical set is a booked call, a form submission and a tracked phone click. Everything else stays as a regular event you can still analyze when you need it.

  3. 03

    Should I ignore traffic completely?

    No, but stop reporting it as success. Traffic is the denominator for rates and the raw material for channel analysis, so you still collect it. What changes is its position: it sits below outcomes, source, cost and quality. A traffic rise with flat outcomes is a warning, not a win. Report traffic only next to the outcome it should have produced.

  4. 04

    Why do Analytics and Search Console show different numbers?

    They measure different things. Search Console counts impressions and clicks on Google results, credited to the canonical URL,[4] while Analytics counts sessions and events on your site.[3][8] It applies its own attribution rules and thresholds. Clicks and sessions will not match one to one. Use Search Console for visibility trends and Analytics for what happened after the visit, and never subtract one from the other.

  5. 05

    How often should I review marketing numbers?

    Monthly for decisions, weekly only for tracking health and quarterly for strategy. A monthly review gives enough outcomes to compare and enough time to see leading indicators move. Check that the proof event still fires every week, because tracking breaks silently after site changes. Review leading indicators against outcomes every quarter and drop any that stopped predicting them.

  6. 06

    What if I have too few conversions to measure anything?

    Count by hand and treat ratios as direction. With fewer than about 30 outcomes a month, read each lead, record its real source in a spreadsheet and ask customers how they found you. Do not rank channels on customer cost until you have at least 20 customers in a period. Use qualified-lead cost for early comparisons and call the ranking provisional.

The evidence behind this page

Sources

Evidence freshness

11 sources, 4 dated in the last 3 years

Each bar counts sources by the date the source itself shows. 6 show no date.

2223242526

Cite or share this page

Quote the rule or the order with credit and a link. Suggested citation:

Adnan Gourija (2026). What should you measure first, second and third in marketing?. The Marketing Field Manual, Webtec. https://www.trywebtec.com/resources/field-manual/what-to-measure-first/

  1. 1Google Analytics Help: About key eventsNo date shown
  2. 2Google Analytics Help: Engagement rate and engaged sessionsNo date shown
  3. 3Google Analytics Help: About data thresholdsNo date shown
  4. 4Google Search Console Help: Performance report (Search results)No date shown
  5. 5web.dev: Web VitalsPublished May 4, 2020. Last updated October 31, 2024
  6. 6web.dev: Why lab and field data can be differentLast updated 2022-07-18 UTC
  7. 7Google Tag Manager Help: Preview and debug containersNo date shown
  8. 8Google Analytics Help: Attribution in AnalyticsNo date shown
  9. 9Google Search Central: AI features and your websiteLast updated 2025-12-10 UTC
  10. 10Google Analytics Help: Release notes (AI Assistant channel entry)No date shown (entry dated May 13, 2026)
  11. 11Wikipedia: Goodhart's law (secondary source, for the 1975 and 1997 attributions)Last edited 9 October 2026, 05:33 UTC

Where to go next

Rated by our clients

Don't take our word for it.

Some of the businesses we have worked with

Your cookies, your call

We use only the cookies this site needs to work. If you say yes, we also turn on Google Analytics to see what works, and our live chat, which records how you use the site so we can help you faster. Nothing optional runs until you choose. See our cookie policy.