Bow Tie Kreative SEO System

1. Measurement philosophy

SEO is a partially observed causal system. Search engines do not expose every internal variable, so the measurement model separates:

Never infer success from a proxy without checking the downstream outcome it is supposed to influence.

BUSINESS OUTCOMES
DIAGNOSTIC OUTCOMES
SYSTEM STATES
INTERVENTIONS
CONTEXT
UNCERTAINTY
ranking ↑
does not automatically imply
qualified value ↑

2. Outcome hierarchy

Level 1 — Business outcomes

qualified leads
sales
subscription starts
retained customers
expansion revenue
support value
contribution margin
customer lifetime value

Level 2 — Journey outcomes

task completion
next-step action
assessment completion
demo request
quote request
trial start
qualified form
return visit
renewal action
referral/review

Level 3 — Search outcomes

impressions
search appearances
average position / rank distribution
clicks
CTR
landing sessions
branded demand
non-branded demand

Level 4 — Asset and graph diagnostics

query-cluster coverage
canonical ownership
internal in-links
click depth
orphan state
qualified referring domains
content/evidence completeness
freshness

Level 5 — Technical states

discoverable
crawlable
renderable
canonicalized
indexable
indexed
retrievable
valid structured data
field performance

3. Primary formulas

3.1 Actual organic contribution margin

ACTUAL ORGANIC CONTRIBUTION
=
Σ(
  attributed organic conversions
  × contribution margin per conversion
)
− attributable SEO operating cost

3.2 Incremental lift

When a reasonable comparison group exists:

Apply separately to:

- impressions;

- clicks;

- qualified conversions;

- revenue;

- contribution margin;

- retention.

  • impressions;
  • clicks;
  • qualified conversions;
  • revenue;
  • contribution margin;
  • retention.
INCREMENTAL LIFT
=
(Post_treatment − Pre_treatment)
−
(Post_control − Pre_control)

3.3 Expected value

Every estimated input must be labeled and later replaced with observed data.

EXPECTED VALUE
=
Σ(
  demand estimate
  × probability of eligibility
  × probability of useful visibility
  × expected CTR
  × expected qualified conversion rate
  × contribution margin
)
− expected cost

3.4 Value-weighted visibility

Visibility weight may be based on impression share, rank distribution, or a calibrated CTR curve.

VALUE-WEIGHTED VISIBILITY
=
Σ(
  query opportunity value
  × observed visibility weight
)
/
Σ(query opportunity value)

3.5 CTR residual

This is more useful than a universal “good CTR” threshold.

CTR RESIDUAL
=
Observed CTR
− Expected CTR(
    position,
    device,
    country,
    brand class,
    query class,
    search appearance
  )

3.6 Qualified conversion value

QUALIFIED ORGANIC VALUE
=
organic value events
× qualification probability
× contribution margin

3.7 Opportunity score

OPPORTUNITY SCORE
=
(
  demand confidence
  × business fit
  × task value
  × information advantage
  × attainable response gap
  × conversion value
  × evidence confidence
)
/
(
  effort
  × risk
  × maintenance burden
  × time to learning
)

3.8 Eligible canonical coverage

ELIGIBLE CANONICAL COVERAGE
=
priority canonical URLs that are eligible and indexed
/
all priority canonical URLs intended for indexing

3.9 Canonical agreement

CANONICAL AGREEMENT
=
URLs where intended/declarative canonical agrees with observed selected canonical
/
evaluated duplicate or canonical URLs

3.10 Cannibalization indicators

High entropy is a diagnostic, not automatic failure. Confirm same-intent duplication, instability, divided links, or value loss.

DOMINANT URL SHARE
=
largest URL impression share for cluster
/
total cluster impressions
URL ENTROPY
=
−Σ(p_url × ln(p_url))

3.11 Internal graph diagnostics

PRIORITY ORPHAN RATE
=
priority pages with no useful crawlable internal in-link
/
all priority pages
WEIGHTED CLICK DEPTH
=
Σ(page business weight × page click depth)
/
Σ(page business weight)

3.12 Authority diagnostics

QUALIFIED LINK RATE
=
retained relevant editorial referring domains
/
all acquired referring domains

4. Data grain

Store search observations at the smallest useful grain:

Attach:

Store technical observations separately:

Store changes:

date
site_id
cluster_id
query
canonical_url
country
device
search_type
search_appearance
brand_class
change_id
impressions
clicks
ctr
average_position
organic_sessions
task_completions
next_actions
qualified_leads
sales
revenue
contribution_margin
observation_time
url
status_code
robots_allowed
render_success
declared_canonical
selected_canonical
index_state
sitemap_state
internal_inlinks
click_depth
field_lcp
field_inp
field_cls
structured_data_state
change_id
object_type
object_id
laka_level
hypothesis
mechanism
owner
start_time
end_time
rollback_state
primary_metric
guardrails
decision_rule

5. Event dictionary

A measurement plan must define every event.

Avoid meaningless event inflation. An event should correspond to a user state or business state.

Event, Definition, Required properties — 12 rows
EventDefinitionRequired properties
organic_landingSession begins on an organic search landing pagelanding URL, cluster, query class
task_startUser initiates the primary page tasktask ID
task_completeUser reaches a valid completion conditiontask ID, method
next_actionUser takes the intended adjacent stepaction ID
qualified_formForm meets qualification rulesoffer, segment, qualification
sales_acceptSales accepts the leadCRM status
closed_valueRevenue or margin is confirmedvalue, margin, offer
retention_valueRenewal, expansion, or support outcomecustomer, event value
evidence_interactionUser opens data, sources, demo, comparison, or case proofevidence ID
internal_pathUser follows a meaningful graph edgeedge type, target
calculator_completeTool returns a valid resultinputs band, result band
download_useUser downloads or uses a templateasset ID

6. Baseline protocol

Before changing a priority object:

If historical data is unavailable, create a prospective baseline and explicitly reduce causal confidence.

1. Verify data collection.
2. Define treatment object and exposure.
3. Record at least the available stable pre-period.
4. Record query mix, device, country, brand class, and seasonality.
5. Record technical eligibility.
6. Record current page and graph state.
7. Record current business value.
8. Assign change ID.
9. Predefine primary metric and guardrails.
10. State minimum exposure and review rule.

7. Hypothesis grammar

Example:

IF [INTERVENTION]
is applied to [OBJECT]
under [CONDITIONS],
THEN [PRIMARY METRIC]
will change in [DIRECTION]
because [MECHANISM],
while [GUARDRAILS]
remain within [THRESHOLDS].
IF the title proposition for the Calgary web design cost guide
is rewritten to state the year, location, and estimator benefit,
THEN position-adjusted organic CTR will increase
because relevance and expected utility become clearer,
while qualified assessment rate does not decline by more than 10%.

8. LAKA experiment card

Experiment ID:
Object:
Conditions:
Action:
Tools:
Resources:
Expected outcome:
Feedback source:
Constraints:
Value:
Failure mode:

LAKA level:
Magnitude:
Rate:
Direction:
Scope:
Depth:
Duration:
Frequency:
Acceleration:
Variability:
Detectability:
Reversibility:
Propagation:
Amplification:
Accumulation:

Primary metric:
Baseline:
Treatment scope:
Comparison:
Minimum exposure:
Decision rule:
Guardrails:
Rollback:
Owner:

9. Decision outcomes

At every review, choose one:

CONTINUE
EXPAND
REVISE
REVERSE
ESCALATE
RETIRE
MONITOR

Continue

IF evidence is positive
AND guardrails pass
AND additional exposure is needed
THEN continue.

Expand

IF positive effect is credible
AND mechanism is transferable
AND capacity permits
THEN expand to a defined cohort.

Revise

IF mechanism appears valid
BUT implementation is weak or guardrails fail mildly
THEN revise.

Reverse

IF primary outcome declines materially
OR a critical guardrail fails
AND rollback is feasible
THEN reverse.

Escalate

IF the diagnosed failure is deeper than the current LAKA level
AND expected value still exceeds cost
THEN escalate.

Retire

IF remaining value is low
OR maintenance/risk is too high
OR the asset has no distinct task
THEN consolidate, redirect, noindex, archive, or remove appropriately.

10. Experiment types

10.1 Minor element test

Examples:

- title proposition;

- snippet description;

- introductory answer;

- CTA wording;

- anchor text;

- small media change.

Best for high exposure and reversible mechanisms.

  • title proposition;
  • snippet description;
  • introductory answer;
  • CTA wording;
  • anchor text;
  • small media change.

10.2 Major asset test

Examples:

- full rewrite;

- new comparison;

- original data;

- tool or video;

- new evidence architecture.

Evaluate as an intervention package unless components can be separated.

  • full rewrite;
  • new comparison;
  • original data;
  • tool or video;
  • new evidence architecture.

10.3 Structural test

Examples:

- page merge;

- template change;

- internal graph redesign;

- facet control;

- CRM linkage;

- site architecture.

Use cohort or time-series analysis and monitor propagation.

  • page merge;
  • template change;
  • internal graph redesign;
  • facet control;
  • CRM linkage;
  • site architecture.

10.4 Paradigm test

Examples:

- calculator instead of article;

- benchmark instead of generic guide;

- verified directory instead of location-page set;

- interactive diagnostic instead of lead form;

- data API or agent action instead of static copy.

Primary value may include new demand, citations, brand searches, product adoption, and conversion.

  • calculator instead of article;
  • benchmark instead of generic guide;
  • verified directory instead of location-page set;
  • interactive diagnostic instead of lead form;
  • data API or agent action instead of static copy.

11. Comparison designs

Use the strongest feasible design:

SEO often prevents perfect randomization. State the design’s limits rather than implying certainty.

Randomized controlled test
> matched page/cluster control
> staggered rollout
> difference-in-differences
> interrupted time series
> before/after with context controls
> directional observation

12. Observation windows

Do not use one fixed calendar window for every intervention. Define exposure criteria:

A review may occur when one or more exposure thresholds are met, with a maximum stop date to prevent endless observation.

minimum crawl confirmation
minimum index confirmation
minimum impressions
minimum qualified sessions
minimum conversions
minimum full business cycle

13. Dashboard architecture

Executive dashboard

incremental organic contribution margin
qualified organic conversions
value-weighted non-brand clicks
value-weighted target-cluster visibility
eligible canonical coverage

Demand dashboard

cluster impressions
new query families
branded vs non-branded
journey state
audience/offer
country/device
value-weighted demand

Asset dashboard

canonical owner
rank distribution
CTR residual
task completion
next-action conversion
freshness
information/evidence score

Technical dashboard

status correctness
render parity
canonical agreement
index eligibility
sitemap coverage
orphan rate
click depth
field CWV exposure
structured data validity

Authority dashboard

qualified new/lost referring domains
relevance
target page
placement type
retention
brand-demand lift
assisted value

Experiment dashboard

change ID
LAKA level
status
primary metric
effect estimate
uncertainty
guardrails
decision
learning

14. Metric anti-patterns

DO NOT:
count indexed pages as success without value;
treat average position as a precise rank;
use a single sitewide CTR benchmark;
equate bounce with failure for every task;
equate tool authority scores with Google internals;
claim revenue from a forecast;
attribute all movement to the most recent change;
ignore branded demand created by other channels;
optimize event count rather than user state;
continue an experiment without a stop rule.

15. Monthly learning review

Ask:

The output is an updated rule prior, portfolio, and backlog—not merely a report.

Which clusters created the most qualified value?
Which assets lost demand or eligibility?
Which query families are emerging?
Where is the intended canonical not dominant?
Which formats improve task completion?
Which evidence earns links or branded demand?
Which next actions produce quality, not merely volume?
Which technical failures block the most value?
Which LAKA level produced the best value per effort?
Which assumptions were disproved?
What should the generator weight differently next month?