The HBB
Process.
How we run client marketing day to day, how we measure whether we did the work well, and how outcomes are reviewed on a slower clock.
- Why a process
- Three kinds of measures
- The weekly build cycle
- Daily, weekly, monthly
- Channel processes
- AEO: reading intent
- Tailoring per client
- What clients see
- What to build first
Scroll, or use the arrow keys
01 Why a process
Nobody judges a house by its appraisal while it's being built.
Our work runs the same way.
The full reasoning
Nobody judges a house by its appraisal while it's being built. Each day's work is checked against the blueprint and the spec, and inspections happen at set stages. The appraisal comes later, and it reflects the work that went in.
Our work runs the same way. Traffic, leads, and sales move slowly, lag the work by months, and depend on things we don't control: the client's sales team, seasonality, competitors, and a buyer who reads an AI answer and calls without ever visiting the site. When we check outcomes every day, we spend time looking for results that haven't had time to arrive. We spend less time improving the work that produces them.
01 Why a process
We already say this to clients.
The Search and GEO Assessment splits the work in two. This process turns that split into how we run and measure our own work.
What You Can Control
- The core website
- Content freshness and expansion
- Ad execution
What You Can't Control, But Can Monitor
- Signs of life
- AI visibility
- Leads and sales
01 Why a process
The goal: get a little better at the work every week, and be able to show that we did.
The process runs through the same layers as everything else we do.
02 Three kinds of measures
Every channel gets measured at three levels.
Each level has its own review clock. This color code repeats through the rest of the deck.
Did we do the work, on time, to spec?
- Search terms reviewed this week
- Pages published against plan
- Foundation updated after each call
Early signals that the work is landing
- Impressions, rankings
- AI mentions
- Impression share on core terms
- Assistant fetches of client pages
Matched in the client's CRM where we can
- Leads, qualified leads
- Opportunities, revenue
- Cost per qualified lead
02 Three kinds of measures
Judge the week by the work. Judge the strategy by outcomes over months.
The rule, in full
The rule: we judge our week by the workmanship measures, and judge the strategy by the outcomes over months. If outcomes stay flat while the workmanship measures are strong for a full quarter, the strategy is wrong, and that's a Blueprint conversation. If the workmanship measures are weak, we fix the work first, before we draw any conclusion about the strategy.
A missed workmanship measure is useful information. It tells us exactly what to fix. A missed outcome usually doesn't tell us why it was missed.
03 The weekly build cycle
Every channel on every account runs the same five steps.
This is the brand-level template. The channel slides fill in the details.
03 The weekly build cycle
The build log is the record of getting 1% better.
Anatomy of a build log entry
Two ways to tell whether we're actually getting better
Process changes logged per client per month. If it's zero, we aren't reviewing.
Workmanship measures trend the right way over a quarter. Client corrections per draft go down. Time from a found question to a published answer gets shorter. (Line shows direction, not data.)
HBB-PRACTICE.md notes that the editorial and foundation logs never had a review cadence and stayed empty. The weekly walkthrough is that cadence.
The build log, in full
The build log is the record of getting 1% better. Each client gets one running log of process changes. It can live in the existing places: the client's editorial log for content, and the agent-ops learnings files for ads. Each entry records the date, the channel, what we changed, why, and the process measure we expect to move.
When a change would help every client, it moves into the HBB playbook for that channel. That's how one account's lesson becomes the agency's standard.
04 Daily, weekly, monthly
Each cadence answers a different question.
Main rule: no outcome analysis in the daily or weekly review, except when something is broken.
What the daily check looks for
If nothing is wrong, the daily is empty. Accounts with low lead volume can review ads every two weeks instead of weekly; there isn't enough data in a week to act on.
04 Daily, weekly, monthly
The quarterly review reads outcomes against where we expected to be, including the dip.
Full cadence table
| Cadence | Question it answers | What's in it | What stays out |
|---|---|---|---|
| Daily site check | Is anything broken or off pace? |
|
Trend commentary. If nothing is wrong, the daily is empty. |
| Weekly walkthrough | Did we do the work to spec, and what are we improving? |
|
Judging the strategy on one week of outcomes. |
| Monthly Control Room review | Is the work landing, and what did it produce? |
|
New strategy, unless a quarter of evidence supports it. |
| Quarterly Blueprint review | Is the strategy right? |
|
Day-to-day execution detail. |
Accounts with low lead volume can review ads every two weeks instead of weekly. There isn't enough data in a week to act on.
05 Channel processes
Each channel has a goal, its tasks by cadence, and its three kinds of measures.
05 Channel processes · Paid search
Paid search
Goal: show up when prospects search for the things the client most wants to be found for. Spend the most on the client's own terms and the bullseye terms, and only extend outward while the economics hold.
Workmanship
- Search terms reviewed on schedule
- Proposals waiting, and for how long
- Share of spend on bullseye terms (proposed; needs a per-client definition)
- Quality Score in the healthy ranges
Signs of life
- Impression share and top-of-page rate on bullseye terms
- CTR on core terms
- Platform vs warehouse conversions ratio
Outcomes
- Qualified leads, opportunities, revenue (CRM)
- Cost per qualified lead
Paid search, full text
| Cadence | Tasks |
|---|---|
| Daily | Pacing against the approved budget (exists). Disapprovals and anomalies (exists). Audit of agent changes (exists on one account). |
| Weekly | Search terms review and negatives under the Addison method (exists). Impression share on brand and bullseye terms. Quality Score components. Status of ad and landing page tests. Changes made by people since the last check. |
| Monthly | Spend against plan. Impression share lost to budget and to rank on core terms. Cost per qualified lead, from the CRM. |
| Level | Measures |
|---|---|
| Workmanship | Search terms reviewed on schedule. Proposals waiting, and how long they've waited. Share of spend on terms that match the bullseye (proposed; needs a definition per client). Quality Score in the Addison healthy ranges (brand 9-10, non-brand 6-8). |
| Signs of life | Search impression share and top-of-page rate on bullseye terms. Click-through rate on core terms. Ratio of platform-reported conversions to warehouse conversions. |
| Outcomes | Qualified leads, opportunities, and revenue matched in the CRM. Cost per qualified lead. |
05 Channel processes · SEO
SEO
Goal: rank for the client's core queries, on pages that carry real experience and expertise.
Workmanship
- Pages published against plan
- Core queries with a page built for them
- Technical issues open past the client's standard
- New pages linked from related pages within a week
Signs of life
- Non-brand impressions
- New ranking queries
- Position movement on core queries
Outcomes
- Leads from non-brand organic search
- Revenue matched in the CRM
SEO, full text
| Cadence | Tasks |
|---|---|
| Daily | Nothing by hand. Automated alerts for indexing drops or the site going down. |
| Weekly | Search Console movement on core queries. New queries appearing. Index coverage errors. Pages published and internal links added to them. |
| Monthly | Positions on core queries (AgencyAnalytics; DataForSEO weekly where it runs). Non-brand impressions and clicks. Number of pages earning impressions. Core Web Vitals. |
| Level | Measures |
|---|---|
| Workmanship | Pages published against plan. Core queries that have a page built for them. Technical issues open longer than the client's standard. New pages linked from related pages within a week of publishing. |
| Signs of life | Non-brand impressions. New ranking queries. Position movement on core queries. |
| Outcomes | Leads from non-brand organic search. Revenue matched in the CRM. |
05 Channel processes · Content
Content (the Workshop)
Goal: turn what the client knows into published pages, and get faster and more accurate with each cycle. This is where "the system learns" can be measured.
Workmanship
- Days from a client call to the Foundation update
- Days from brief to publish
- Client corrections per draft
- Editorial log entries, and how many turned into rules
- Pages published against plan
Signs of life
- Impressions and AI mentions on pages published in the last 90 days
Outcomes
- Leads whose path includes new content
05 Channel processes · Website
Website
Goal: a fast site that people and machines can read, and a lead path that always works.
Outcome
- Conversion rate of engaged visits
05 Channel processes · Social
Social
Goal: raise awareness among the right people and keep current customers engaged with what the client is doing.
Workmanship
- Posts published against plan, written approval first exists, GEN-11
- Time to respond to comments
- Employee amplification used ("Notify Employees" on LinkedIn)
Signs of life
- Impressions among target job titles and industries (LinkedIn audience data)
- Follower growth inside the target audience
- Engagement rate
- Branded search impressions in Search Console our inference, not proof
Outcomes
- Leads referred by social
- Leads who say they heard about the client on social
There is no social playbook yet. The only cadence material on the drive is one unpresented client program and the approval rule in the Ops SOP.
Social, full text
LinkedIn for most B2B clients. Meta where the client sells to consumers or runs Meta ads.
| Level | Measures |
|---|---|
| Workmanship | Posts published against plan, with written approval first (exists, GEN-11). Time to respond to comments. Employee amplification used ("Notify Employees" on LinkedIn). |
| Signs of life | Impressions among target job titles and industries (LinkedIn audience data). Follower growth inside the target audience. Engagement rate. Branded search impressions in Search Console, as a rough awareness signal (our inference, not proof). |
| Outcomes | Leads referred by social. Leads who say they heard about the client on social. |
Paid Meta has its own measures where it runs: pacing, creative hold rate, and warehouse-based CPA and ROAS.
05 Channel processes · Data and tracking
Data and tracking
Goal: every other measure depends on this. If the tracking is broken, every inspection is wrong.
Workmanship
- Each data source loaded on schedule
- Share of leads with a captured source
- Share of leads with a quality status within the standard window
Outcome
Not applicable. This is the plumbing.
06 AEO: reading intent from what we can see
We can't see what AI assistants tell our clients' buyers.
No assistant reports which prompts named a company. So we collect every indirect signal, infer what the buyer was asking, and check whether a page answers it.
06 AEO · Fan-outs
One buyer question becomes many searches.
"query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously."
Google, describing AI Mode
06 AEO · Query map and evidence
Each client gets a query map, fed by the evidence we can see.
Evidence table, full text
| Source | What it shows | Limits |
|---|---|---|
| AEO tracker | How often assistants name the client for our prompts, with and without web search (exists). | These are our prompts, not real buyers' prompts. It runs free-tier models. |
| Server logs | Assistant fetches by page. OpenAI says ChatGPT-User visits a page when a user asks ChatGPT a question. A fetch means a real person's question led the assistant to that page (exists on some accounts). | Spoofed user agents must be filtered. Not every assistant identifies itself. |
| Search Console Generative AI report | Impressions by page in AI Overviews and AI Mode. Available for all sites since 2026-08-31. | No queries and no clicks. Not yet in any warehouse (planned). |
| Search Console Performance report | Queries and pages. AI Mode clicks and impressions are included in web search totals. Long, question-shaped queries hint at fan-outs. | AI Mode isn't separated from regular search. |
| GA4 AI assistant channel | Sessions referred by chatgpt.com, perplexity.ai, gemini.google.com, copilot, claude.ai, and similar, with landing pages (exists on one account). | Many AI visits arrive with no referrer. |
| Direct visits to deep pages | Direct sessions that land on interior pages. Our experience is that most of these come from AI. | This is an inference. Label it as one. |
| On-site paths | What the visitor read next, in PostHog or Clarity. The path tells us what they were trying to figure out. | Needs session tooling on the site. |
| Sales calls and the CRM | The questions buyers ask on calls. "How did you hear about us" answers that name an assistant. | Depends on the client's sales team recording it. |
06 AEO · The weekly intent review
Five steps every week, one every month.
Intent tiers (from AEO discovery work already done on one account)
The six steps, full text
- Pull last week's AI evidence by page: assistant fetches, AI referrals, generative AI impressions, and direct entries on deep pages.
- For each page with evidence, write a one-line intent note: the question the buyer was most likely asking. Tag it with an intent tier: buyable, pre-purchase, reference, branded, or noise. These tiers come from AEO discovery work already done on one account.
- Match each intent note to the query map. If it doesn't fit an existing fan-out, add it as a candidate.
- Check coverage. Does a page answer that question directly, near the top? If not, add it to the content backlog in the Blueprint.
- Explore. Ask an assistant three to five core questions by hand, with web search on. Record the sub-questions it seems to answer and the sources it cites. New sub-questions go on the candidate list.
- Monthly: move the strongest candidates into the AEO tracker and retire prompts that have stopped telling us anything.
06 AEO · Measures
AEO gets the same three kinds of measures.
Workmanship
- Intent review done each week
- Fan-outs identified
- Share of fan-outs with a page that answers them directly
- Days from finding a fan-out to publishing the answer
- Tracker prompts refreshed each month
Signs of life
- Tracker presence rate, with and without web search
- Generative AI impressions by page
- Assistant fetches by page
- AI-referred sessions and how engaged they were
Outcomes
- Leads with an AI referral, a direct entry on a deep page, or a self-reported AI source
HBB-PRACTICE asks for AI visibility claims to be re-baselined each quarter. A stale number is a liability in front of a client.
07 Tailoring per client
The cycle and the measures stay the same. The spec sheet changes.
What changes per client is written on a one-page spec sheet in the client's Blueprint.
Most client KPI files have an empty "Confirmed KPIs" table. Getting each client to confirm its outcomes is the first line of its spec sheet.
Spec sheet and factors, full text
- Channels in use, and which are handled by agents
- The spec for each task, and any thresholds (pacing band, Quality Score floor, response time)
- Cadence changes, such as reviewing ads every two weeks when volume is low
- Core queries, bullseye terms, and where the query map lives
- Which signs of life matter most for this client
- The outcomes the client has confirmed it cares about, and how each is measured
- What the client receives, and how often
| Factor | What it changes |
|---|---|
| Lead volume | Low volume: review less often, and lean on signs of life and warehouse signals over platform conversions. |
| Sales cycle | Long cycles: outcomes get judged quarterly or longer, and pipeline counts more than closed revenue in year one. |
| Data access | Without a CRM integration, outcome measures are counts only. Fixing that becomes a process item. |
| Business type | B2B: LinkedIn and search lead. Consumer-facing: Meta and reviews carry more weight. |
| Tooling in place | Whether the account has server logs, session tooling, and the AEO tracker decides how much of section 6 we can run. |
08 What clients see
Showing clients the process changes is part of the point.
It shows them the work gets better each week, even in a month when the outcomes haven't moved.
Daily
Weekly
Monthly
09 What to build first
Roughly in order. The first two cost almost nothing.
- Start the weekly walkthrough and the build logOn every active account. Use the existing editorial and agent-ops learnings files.
- Write a spec sheetFor each client, starting with confirmed outcomes.
- Start the AEO intent review by handOn the accounts that already have server logs, then add the others.
- Add a GA4 AI assistant channelTo the analytics onboarding checklist so every client has one.
- Load the Search Console Generative AI reportInto the warehouse. It's already planned as a tracker feature.
- Build an inspection viewPer client, from data we already have: agent run records, ClickUp task dates, publish dates, editorial log entries.
- Write channel playbooksFor SEO, AEO, and social. Paid search already has the Addison manual.
- Fill the empty monthly report templateIn
templates/reporting/monthly-report/, using the section 8 structure.