The Indicator No Longer Shows the Outcome: A 2026 Guide

Why Did They Separate?

Three mechanisms produce this split.

The unit of measurement moved

Revenue is measured in a currency and the currency itself is moving. The same figure meant one thing a year ago and something else now. Where inflation ran low this could be ignored; above thirty per cent it grows large enough to break the indicator. An indicator that looks fixed is being measured with a moving ruler.

The group outside the definition grew

The unemployment rate counts only active jobseekers; whoever stops looking falls outside it. As the excluded group grows, the rate appears to improve. The same logic runs inside a business: survey only your remaining customers and the departed never enter the measurement.

A layer moved in between

Summaries, models and agents now sit between your ranking and your visitor. Ranking remains measurable but no longer determines the outcome alone. As the layer thickens the indicator loses explanatory power — the measurement is intact, the link to the result has weakened.

What Happened

Six stories sit behind what follows. Each was taken apart separately; these paragraphs show where they touch.

Revenue rose, purchasing power fell

Turnover rose 25.8 per cent year on year in June while inflation stood at 31.75. Nominal growth can conceal real contraction, and the growth rate slowed seven points between May and June.

Unemployment fell, so did participation

The unemployment rate dropped to 7.9 per cent while participation fell to 52.7 and underutilisation held at 29.9. The denominator is shrinking too.

Impressions rose, clicks fell

A randomized field experiment measured AI summaries cutting organic clicks by roughly 38 per cent. Sponsored clicks and user satisfaction, meanwhile, did not move.

Ranking held, citation changed

After one model change, roughly 42 per cent of previously cited domains dropped off the list. The link between top-ten placement and citation has weakened.

Present on one engine, absent on another

An analysis of 680 million citations found only 11 per cent of domains appearing on both engines. Measurement on a single platform is a small slice of the picture.

The page is visible but unread

Agents view pages and inspect their structure; outdated prices and browser-rendered content get skipped. A human visitor asks; software skips.

The Common Thread: The Failure Is Silent

What these six share is the most awkward feature of all — nothing raises an alarm.

The indicator keeps looking healthy

Revenue up, unemployment down, ranking steady. Anyone watching the dashboard sees no problem; the problem forms outside the dashboard. That is more insidious than bad news arriving late: the good news keeps arriving and generates no signal that anything has deteriorated.

The loss surfaces late

Real contraction appears when the year closes, citation loss when traffic falls, and an agent skipping your page never appears at all. By the time it registers, the window for recovery has closed.

Misdiagnosis is expensive

Falling traffic gets blamed on content quality, shrinking revenue on the sales team, hiring difficulty on the market. Intervention applied while the cause sits elsewhere burns resources.

The Number Is Right, the Question Is Old

The second thread: none of these indicators is lying.

The question stopped fitting

“Did revenue grow?” was once sufficient. It now needs to become “what happened in real terms” — same indicator, different question.

Pairs rather than singles

Revenue with units, unemployment with participation, impressions with clicks, ranking with citation. Every indicator in this period has a partner, and read alone each one misleads.

The headline percentage does not decide

The same study gets reported at different rates in different places. Working from the figures themselves is safer than working from someone else’s percentage.

The Source Deserves Scrutiny Too

The third thread: in this period where data comes from matters as much as the data.

Method determines the result

What makes the click-loss study valuable is not its figure but its design: because users were randomly assigned, causal inference becomes possible. Earlier observational estimates could not do that.

A vendor’s data gets read carefully

Much of the measurement in the visibility field comes from companies selling tools in that field. The same figure being confirmed by several independent methods is what makes it credible.

Faults enter the data as well

A malfunction during one model transition distorted the earliest measurements; researchers could not separate the real effect from the fault for a period. When you see a sharp drop, waiting a few weeks is the cheapest insurance available.

Decision Map: Four Profiles, Four Routes

Which indicator to examine depends on where your business measures from. Start with whichever route sits closest; each ends with the indicator that most often misleads on that route.

Profile 1 · Reading revenue as growth

Your priority is real measurement. Compare revenue growth against inflation, track unit sales separately, use your own sector’s rate and test your price pass-through. Most often misleading: annual revenue growth. Read alone it shows growth; set beside inflation it says the opposite.

Profile 2 · Hiring

Your priority is your own application data. Compare applications to the same advert year on year, do not set wages by the national rate, measure your existing team’s hours, and reach the underutilised pool with flexible terms. Most often misleading: the national unemployment rate. It describes nothing about the market for your particular skill.

Profile 3 · Traffic arriving through search

Your priority is channel separation. Split traffic by question type, separate brand searches from general ones, read impressions alongside clicks, and track ranking apart from citation. Most often misleading: average position. It can sit perfectly still while your visibility has changed.

Profile 4 · Selling from your site

Your priority is data hygiene. Audit price accuracy, move critical information out of browser-rendered code, test forms on mobile, and clear out discontinued services and closed locations. Most often misleading: the site appearing to work. A page that works for a person can be incomplete for automated systems.

Thirty Days, One List

By the end of the month you want one page naming the indicators you can no longer trust alone, and beside each the second figure that verifies it. The aim is not more measurement but paired measurement.

Week 1 · Inventory

Write down the figures you check regularly. The list probably stops at five: revenue, traffic, applications, ranking, order count. A short list is fine; not knowing it is short is not.

Week 2 · Find the partners

Beside each indicator write the second figure that would verify it: units for revenue, citation for ranking, clicks for impressions, interview rate for applications. That matching exercise is the whole job.

Week 3 · Run the first pair

Take one pair only and set twelve months of data side by side. Moving the same way means the indicator holds; diverging means the real story sits inside the divergence.

Week 4 · Make it routine

Write the remaining pairs down with who checks them and how often. Unwritten measurement is forgotten by the second month; this is the easiest and most skipped step in the plan.

What to Track

Nine indicators suffice, all read in pairs.

Three economic pairs

Revenue and unit sales, your price increases and inflation, order count and average basket. All three expose the gap between nominal and real.

Three visibility pairs

Impressions and clicks, ranking and citation, brand search and general search. All three catch the break on the search side.

Three structural checks

Price accuracy, mobile form functionality, and whether critical information exists server-side. These three are audited rather than measured; annually is enough.

Six Reasonable-Sounding Errors

Careful people make all of these. That is what makes them worth listing.

Deciding on one indicator

Every indicator in this period has a partner. Read alone, an accurate figure leads to an inaccurate conclusion.

Taking the headline percentage

The same study appears at different rates in different places. Working from the figures themselves is always safer.

Applying the national average to yourself

National unemployment or total turnover growth describes neither your sector nor your skill requirement. Your own data is always more useful.

Reacting immediately to a sharp drop

Part of the early measurement in one model transition came from a fault later corrected. Waiting a few weeks protects against a wrong diagnosis.

Generalising from one platform

Overlap between two engines sits around 11 per cent. Appearing in one does not mean appearing in the others.

Treating a working site as sufficient

A page that runs smoothly for a human visitor can be incomplete for automated systems. The failure forms silently and appears in no report.

How Does This Period Develop?

No firm forecast; three directions can be argued.

The intervening layer keeps thickening

Summary, model and agent layers are growing. The distance between ranking and visitor is not expected to shorten.

Measurement fragments further

Divergence between platforms is widening rather than narrowing. Producing a single visibility figure gets harder over time.

Whoever holds their own data pulls ahead

As sector averages lose usefulness, the advantage of running your own measurement grows. That advantage costs almost nothing.

A Solid Digital Foundation

Paired reading depends on the data having been separated properly.

Channels need reading apart

Brand search, general search and direct traffic move differently; merged, the real change disappears. The Google Search Central documentation describes how search performance data can be split by query and channel. Merged data hides the divergence.

Choose the comparison period carefully

Comparing against the same month a year earlier, rather than the previous month, strips out seasonality and prevents inventing trends.

Preserve the history

This month’s figure means nothing on its own. Paired reading works only with a time series; storing the record is half the job.

Built once, runs every period

Channel separation, period comparison and indicator pairs, defined once, generate data by themselves. We cover the setup in our approach to digital consulting, with implementation in process and e-commerce consulting.

Frequently Asked Questions