Tyler Gargula - Decision Intelligence for Enterprise SEO

Transcript

1. Decision Intelligence for Enterprise SEO
2. Tyler Gargula Tech SEO Director Web Dev Data Science Musician Chicago, IL
3. Meet Toby
4. Toby is an SEO Consultant
5. Toby’s disappointed with the results from his SEO recommendations
6. Toby feels stuck, what will move the needle?
7. and once he decides…how can he be sure?
8. decision intelligence…
11. Is your organization ready for Decision Intelligence? Business Willingness to change: can they pivot when data contradicts assumptions? Current process: are decisions made by instinct, committee, or data? Decision Tracking: are outcomes actually being measured? Development Priorities: can dev resources respond to insights? Timing?
12. Survey: “50% of people said a good decision is defined by its outcome and 50% said it’s defined by its process…” Source: Gartner
13. DECIDE 01 | DEFINE Primary Goal 02 EXTRACT & Engineer Data 03 CLASSIFY* Priority Levels 04 INTEGRATE* Multiple Data Sources 05 DISTRIBUTE Insights, Not Raw Data 06 EXECUTE Against Your Insights (Test) *=optional
14. 01 | DEFINE Your Primary Goal The most important step.
15. 01 | DEFINE "What does success look like for this analysis?" Process Inputs Data Outputs Goal x f(x) y
16. Outputs Significance: Do you have access to the right data? And enough of it? Connected: Is this tied to a business goal, not vanity? Clear: Can you tell a story about the outcome? Actionable: Will this analysis enable a decision? 01 | DEFINE
17. 01 | DEFINE Use common analysis types to help define goals: Answers Descriptive Diagnostic Prescriptive Predictive “What happened?” “Why x happened?” “What we should we do?” “What will happen if…?”
18. Identifying anomalies and outliers in data Correlation vs Causation (cautiously) Understanding performance across segments Identifying performance gaps Isolating signal from noise Quantifying contribution of different factors Identifying white-space opportunities Mapping competitive overlaps and gaps Understanding how metrics change over time Identifying leading vs. lagging indicators Mapping customer journey through data Identifying behavioral segments Understanding which variables drive outcomes • Mapping competitive overlaps and gaps Understanding data completeness and reliability Assessing representativeness of sample Recognizing inflection points • Identifying data collection issues 01 | DEFINE OUTPUTS
19. AVOID BIAS, SERIOUSLY
20. Bias in = Bias Out Even the best, most well-intentioned analysis is susceptible to unconscious bias. CHECK YOURSELF Dishonest or biased reports harm our industry reputation. Explain what's actually happening with all available data. 01 | DEFINE AVOID BIAS
21. Confirmation Bias: Seeking data that supports your hypothesis while ignoring contradictory evidence. Cherry-picking SEO data or dishonestly using filters to exclude inconvenient results 01 | DEFINE AVOID BIAS
22. Selection Bias: Analyzing only successful/visible data, missing the complete picture Exporting 1,000 rows from GSC… Using SEO tools for 1st party research… Sampled GA4 data… 01 | DEFINE AVOID BIAS
23. Outcome Bias: Defining success criteria after the results Adjusting thresholds until data looks favorable 01 | DEFINE AVOID BIAS
24. Recency Bias Over-weighting recent data while ignoring patterns Making strategy shifts based on last week's drops/spikes 01 | DEFINE AVOID BIAS
25. Just start with a solid hypothesis If “X” then “Y” We want to measure success by X, Y, or Z We want to analyze the relationship between “X” and “Y” 01 | DEFINE AVOID BIAS
26. Examples: DEFINING GOALS for complex SEO problems
27. Problem: How do we address“Crawled not Indexed” Define Goal: Analyze correlations of content signals and indexation Output: Attributes that matter: unique images vs placeholders, stock status, availability Site size: 40M URL site ⭐ ⭐ ⭐ 01 | DEFINE EXAMPLES
28. Problem: Why is our traffic eroding and where? Define Goal: Create keyword intent groupings and analyze trends Output: Categorized 408k kws by intents. Decay analysis of keyword intent groups Site size: 200M URL site 01 | DEFINE EXAMPLES Affected Intents Stable Intents Trend Analysis of 20 KW Intents
29. 02 | EXTRACT & Engineer Data
30. The ETL Process
31. Respect the ETL workflow EXTRACT TRANSFORM LOAD Pull raw data from various data sources Clean,label, enrich, score, join Deliver actionable insights via data viz, dashboards, etc 02 | EXTRACT ETL
32. EXTRACT - TRANSFORM - LOAD EXTRACT & Engineer Data
33. NOT ALL DATA SOURCES ARE CREATED EQUAL
34. Godly Solid Fine Meh Never API API BQ Exports CMS DB UI Export UI Exports Log Files SERP APIs Web Scraper …(all other SEO tool platforms) 02 | EXTRACT DATA SOURCE TIERS
35. Godly Solid Fine Meh Never API API BQ Exports CMS DB UI Export UI Exports Log Files SERP APIs Web Scraper …(all other SEO tool platforms) 02 | EXTRACT DATA SOURCE TIERS
36. Godly Solid Fine Meh Never API API BQ Exports CMS DB UI Export UI Exports Log Files SERP APIs Web Scraper …(all other SEO tool platforms) 02 | EXTRACT DATA SOURCE TIERS
37. Godly Solid Fine Meh Never API API BQ Exports CMS DB UI Export UI Exports Log Files SERP APIs Web Scraper …(all other SEO tool platforms) SEO tools are great for research, but not as reliable for analyzing 1st party traffic & performance 02 | EXTRACT DATA SOURCE TIERS
38. Godly Solid Fine Meh Never API API BQ Exports CMS DB UI Export UI Exports Log Files SERP APIs Web Scraper …(all other SEO tool platforms) 02 | EXTRACT DATA SOURCE TIERS
39. Godly Solid Meh Never API API BQ Exports CMS DB UI Export UI Exports Log Files SERP APIs Web Scraper …(all other SEO tool platforms) 02 | EXTRACT DATA SOURCE TIERS Fine
40. FOR LARGE SITES, USE RANDOM SAMPLING
41. You don’t need to analyze 100% of large datasets Stratified Sampling
42. EXTRACT - TRANSFORM - LOAD TRANSFORM (ENGINEER) Your Data
43. Cleaning: Handling duplicates,nulls, filtering Enriching: Categorizing, labeling, scoring Joining: Merging various data sources by URL, Keyword, Date, etc. Performance based filtering, removing rows with missing or incomplete metrics Categorize kw’s or pages into groups, performance buckets based on metric distributions Combining data that are siloed but report against the same dimensions 02 | EXTRACT TRANSFORM
44. Transforming data can help you better understand your data EXTRACT - TRANSFORM - LOAD
45. DISTRIBUTIONS Understand the distribution, or skew of your data. Identify outliers, and anomalies.
46. Histograms Understand the skew of your data. Greater concentration of pages indexed (y) when Crawl date is more recent (x). 02 | EXTRACT DISTRIBUTIONS
47. Lorenz Curve (cumulative distribution) Use for gathering a prioritized sample based on performance (pages, queries, etc) Sampling the 70th percentile of pages because they account for 85% of traffic 02 | EXTRACT DISTRIBUTIONS
48. PERCENTILES Identify natural breakpoints in your dataset.
49. Quartile Bins Group data based on their performance segment. Help understand dataset weights. Performance Bins Likely noise 02 | EXTRACT PERCENTILES
50. Violin Plots Visualize data spread and variance across various metrics to identify opportunities. The median internal links to Seg 1 are lower than other segments 02 | EXTRACT DISTRIBUTIONS Other Seg 1 Seg 2 Seg 3 Number Internal Links Internal Linking Distribution by Segment
51. DATA LABELING Assigning meaningful categories to unstructured data. Transform raw data into analyzable segments.
52. Labeling data aggregates similarities and shows more meaningful insights
53. Performance Labels Dividing pages into segments based on performance Example: Indexation vs page click bands More low value pages are increasing in indexation, high value pages are being lost. 02 | EXTRACT DATA LABELING 500−1000+
54. Rule-Based Category Groups Use metric groups & category groups to understand performance buckets across categories Crawl buckets by site categories, discover categories that need more internal linking 02 | EXTRACT DATA LABELING Product Categories by Last Crawled: Sorted by Recent Crawl
55. CORRELATIONS (cautiously) Understanding relationships between variables. Identify which metrics move together.
56. Multi-Dimensional Correlations Positive correlation: coefficient closer to 1 Negative correlation: closer to -1 P-value: < 0.05 suggests statistical significance, >0.05 suggests likely due to chance. Correlation coefficient: 0.91 P-value: 9^(-10) 02 | EXTRACT CORRELATIONS Discovery Crawls vs Refresh Crawls vs Indexed
57. Be skeptical of correlations, but understand that they sometimes occur. they’re good for hypothesis testing!
58. Some just make sense
59. Indexed vs Clicks vs Impressions Weak positive correlation between Clicks and Indexed Correlation coefficient: 0.25 P-value: 0.0178 Strong positive correlation between impressions and indexed Correlation coefficient: 0.70 P-value: 1 ^ -10 02 | EXTRACT CORRELATIONS Impressions vs Clicks vs Indexation
60. Some don’t make sense, but are interesting
61. Sitemap Index Order vs URL Order vs Crawl Frequency Inconclusive… But you never know 02 | EXTRACT CORRELATIONS Crawl Frequency by Sitemap Index Order & URL Order Sitemap Index Number (order) Days Since Last Crawled URL order in Sitemap
62. 03 | CLASSIFY* Priority Levels
63. EXTRACT - TRANSFORM - LOAD CLASSIFY
64. Connect business value with data & work towards what matters
65. Priority can be Risks or Opportunities: Risks Impact of loss (high value pages/queries declining) Decay of metric (negative slope, downward trend) Volatility (instability) Threshold breach (falling below critical percentile) Impact of inefficiency (crawl budget) 03 | CLASSIFY
66. Priority can be Risks or Opportunities: Opportunities Potential for gain (underperforming asset, crawl optimizations) Growth trajectory (positive slope indicating momentum). Efficiency gaps (high visibility, low ctr) Underutilized assets (good position low clicks) 03 | CLASSIFY
67. Potential for gains Indexation = Visibility, leads to brand exposure (their business value) Crawl-delay to bingbot to improve crawlability for Googlebot (higher value indexation) 03 | CLASSIFY OPPORTUNITY LEVELS
68. Risk Analysis Risk in this case, removing a nav item with high user engagement, link equity, and revenue attribution. 03 | CLASSIFY RISK LEVELS Cat_1 Cat_2 Cat_3 Cat_4 Cat_5 Cat_6 Cat_7 Cat_8 Cat_9
69. Content Decay (slope) PoP, MoM, YoY comparisons are missing all of the context between dates. Metric slopes are a more meaningful metric. 03 | CLASSIFY RISK LEVELS Page B Page A Page C Page D Page E Page F Clicks & Avg Position Over Time (Decaying Pages)
70. 04 | INTEGRATE* Multiple Data Sources
71. EXTRACT - TRANSFORM - LOAD INTEGRATE
72. You might be solving the wrong problem, perfectly
73. Merging various data sources allows priority to be even more clear
74. Single Source can leave you limited. GSC Alone: “High Impressions” -> Might be a priority? GA4 Alone: “This page has engagement” -> Is it valuable? Revenue data alone: “This product sells well” -> Is SEO contributing? 04 | INTEGRATE
75. Integrated view: GSC + GA4 + Conversions: "High impressions + Low traffic + High conversion potential = Clear opportunity" Crawl logs + GA4 (AI Referrals): “Low Bot Crawling + High Conversions = Optimize internal linking” GA4 + Conversions + GSC: "High traffic + Low conversions + Good position = Optimize for conversions" 04 | INTEGRATE
76. Efficiency Gaps GSC + GA4 +Conversions This is a sign that SEO can play a larger role in revenue. 04 | INTEGRATE OPPORTUNITY LEVELS
77. Use weighted averaged or composite scoring to understand impact & priority
78. Impact Score = ((W₁ × Clicks) +( W₂ × Conversions) + (W₃ × Weighted_Issues) / (W₁ + W₂ + W₃)) / Max_Value × 100
79. PageRank Share + Conversion Data + GSC Clicks + Backlink Analytics Understand which sections of your site need priority based on a gap between various metrics. 04 | INTEGRATE COMPOSITE SCORING
80. GA4 Conversions + GSC + Issue Severity + Issue Volume By incorporation multi-source data, audits can have direction, no more checklists. 04 | INTEGRATE IMPACT SCORE
81. Priority Score = [(SEO Impact × 3) + (Business Value × 4) + (Capacity × 2) + (Urgency × 1)] / 10
82. ç Simple Priority Matrix SEO Impact: Low, Medium High (1−3) Business Value: Low, Medium, High (1−3) Capacity (dev): Low, Medium, High (1−3) Urgency: Not Urgent, Urgent (1−2) Priority Score: 1−3 Create your own weights! 04 | INTEGRATE PRIORITY SCORE
83. Track Outcomes Start Event Trigger Analysis Type Define Primary Goal Extract & Engineer Data Classify Priority Integrate Data Sources Distribute Insights Execute & Test Backlog False True (External Request, External Event, Monitoring) Priority Score >= n?
84. 05 | DISTRIBUTE Insights, Not Raw Data
85. EXTRACT - TRANSFORM - LOAD DISTRIBUTE
86. Stakeholders likely don’t want to draw conclusions from raw data they may misinterpret your data, give up, or have biases.
87. Convert your raw data into actionable insights Data Processing (grouping,segmenting, tagging, etc) Raw Data Insightful Data 05 | DISTRIBUTE Organized Data Post Processing (conditionals, filtering, sorting, etc)
88. Summarize patterns in your data, programmatically
89. Condition-Based Summaries Create rule-based systems to convert patterns into summaries Much better than showing complex data patterns, or unnecessary examples. 05 | DISTRIBUTE INSIGHTS
90. Condition-Based Summaries + Filtering Example: Content performance slope summaries, filtered for negative trends Target Dataset, prioritize content refresh 05 | DISTRIBUTE INSIGHTS
91. Condition-Based Summaries + Intent Labeling Isolate keywords per decaying page, label keyword intents and show trends. Overlay with Google algo updates We know our declining intent, pages, and keywords! 05 | DISTRIBUTE INSIGHTS Click Movement by Intent Cluster
92. TIP: since we need to continue using spreadsheets, always have a summary column
93. 05 | DISTRIBUTE INSIGHTS
94. Speak towards the interests of stakeholders as best you can Data + Analytics knowhow + Storytelling + Domain Expertise
95. 06 | EXECUTE Against Your Insights (Test)
96. YOU MADE IT! EXECUTE
97. If you don’t have next steps, what’s the point? Action leads to results
98. Sharing new opportunities? Updating content? Removing pages? Requesting technical changes? Sharing a test/hypothesis? Clarifying a concern? Escalating a request for resources? Reprioritizing strategy? Sharing knowledge for further discussions? Based on your insights… Next Steps 06 | EXECUTE ARE YOU? Insights
99. Execution Establish your benchmark & expected outcome (estimate) Implement, or test Annotate & Measure for at least 3 months Report on the outcome, noting other site changes & external factors 06 | EXECUTE
100. Keep a running log of site changes & annotate everything
101. Start feeling more confident about your decisions! DECIDE 01 | DEFINE Primary Goal 02 EXTRACT & Engineer Data 03 CLASSIFY* Priority Levels 04 INTEGRATE* Multiple Data Sources 05 DISTRIBUTE Insights, Not Raw Data 06 EXECUTE Against Your Insights (Test)
102. Thank You For you attention. https://www.linkedin.com/in/tylergargula/ https://github.com/tylergargula 🔗 tylergargula.dev