On this page
A marketing playbook is a named, versioned rule that watches one thing about an organization's advertising and raises a finding when a number crosses a line declared in advance. The catalogue below ships with the product; an organization can edit it and add its own rules. Playbooks are selected by their numeric conditions, never by a person browsing a list and never by a language model.
The grammar every playbook is written in#
| Part | Holds | Why |
|---|---|---|
| when | A condition tree: leaves compare a signal with a threshold, joined by AND or OR | A threshold can be a literal, a quantile of the entity's own history, or a change against a baseline (trailing mean, same period last year, year-to-date mean), so a rule travels between clients without retuning |
| eligibility | The evidence floor: grain (brand, campaign class, campaign), objectives, minimum spend, minimum days | Below the floor the answer is "not enough data", never "fine" |
| action | The kind of action and its parameters | Integer parameters only, so a rule cannot carry an invented multiplier |
| measure | The conversion metric, the horizon in days and what counts as success | Declared before any outcome exists, so success cannot be redefined afterwards |
| guardrails | Blocking checks with literal thresholds | A guardrail that depended on history would abstain exactly when it needed to block |
| display | The wording, each bullet bound to a condition leaf | A build audit fails on a bullet that claims something no leaf checks |
Money is integer VND and ratios are basis points. Evaluation is three-valued: a missing signal or an unresolvable baseline is unknown, and unknown never matches. The detection operators are shared with the morning brief (Signals and the morning brief).
The catalogue#
Every rule below measures itself in leads and raises an audit finding; none changes a budget or touches an ad platform. The groups follow the funnel. Status, last verified 24 September 2026: playbook library.
Delivery and creative
| Playbook | Fires when | Evidence floor | Horizon |
|---|---|---|---|
| Delivery cost spike | Cost per thousand impressions more than 40% above the account's year-to-date mean | 500,000 VND spent | 14 days |
| Delivery cost in its worst quartile | Cost per thousand impressions above the entity's own 75th percentile over 90 days | 500,000 VND spent | 14 days |
| Click-through collapse | Click-through rate more than 30% below its 28-day mean | 500,000 VND spent | 14 days |
| Ad fatigue | Frequency at or above 3.0 and click-through more than 20% below its 28-day mean | 500,000 VND spent | 14 days |
| Impression saturation ceiling | Frequency at or above 4.5, on lead, traffic or content campaigns | 1,000,000 VND spent | 14 days |
The two delivery-cost rules are alternatives: year-to-date suits a stable account, the quartile suits a volatile one. Run one per organization.
Landing page
| Playbook | Fires when | Evidence floor | Horizon |
|---|---|---|---|
| Landing page click loss | More than 30% of link clicks never became a page view | 1,000,000 VND spent | 14 days |
| Critical landing page loss | 60% or more of link clicks never arrived | 1,000,000 VND spent | 7 days |
Lead cost and results
| Playbook | Fires when | Evidence floor | Horizon |
|---|---|---|---|
| Cost per lead outside the target band | Lead cost outside the declared band, in either direction | 90,000 VND spent, lead-objective campaigns | 14 days |
| Solution-tier cost per lead outside band | Lead cost above the higher declared band | 750,000 VND spent, lead-objective campaigns | 14 days |
| Lead cost drifting off baseline | Lead cost more than 30% above its 28-day mean | 1,000,000 VND spent | 28 days |
| Result rate collapse | Result rate more than 30% below its 28-day mean | 1,000,000 VND spent | 14 days |
| Spend with zero results | Material spend and results at or below zero | 1,000,000 VND spent, at least 2 days | 7 days |
| Efficient scale opportunity | Lead cost below the efficient ceiling, result rate at or above its 90-day median, material spend | 3,000,000 VND spent | 28 days |
Efficient scale is the one rule that reports good news: where more budget is likely to pay, not only where to stop.
Budget pacing
| Playbook | Fires when | Evidence floor | Horizon |
|---|---|---|---|
| Spending ahead of plan | More than 20% over the pro-rated plan | 7 days in the window, brand grain | 14 days |
| Spending behind plan | More than 20% under the pro-rated plan | 7 days in the window, brand grain | 14 days |
| End-of-flight shortfall | More than 30% behind plan | 14 days in the window, brand grain | 14 days |
Horizons follow what is watched: 7 days for incident-shaped findings, where money bought nothing; 14 days for operational rhythm; 28 days for price and quality drift, which needs a month before the number means anything. The morning brief baselines on 7 periods for the same-day catch; the catalogue uses 28 days for the weekly judgement, where day-of-week effects would otherwise dominate.
Not in the catalogue
Designed but held out of the catalogue, because the signals they need are not yet written to the signal store: site engagement, search visibility, email bounce rate, email engagement, email spam complaints (which also needs a connector), and brand mention and sentiment rules, which belong to the social pack. A rule that names an unpersisted signal evaluates to unknown and never fires, so these are held back rather than listed as apparent coverage. Statistical anomaly detection and explaining a move by its top-contributing dimension are out of scope for this catalogue.
Counter-indications#
Suppression pairs are part of the rule set:
- Zero results suppresses result-rate collapse on the same entity: a fall to exactly zero is an outage, not a decay.
- The two delivery-cost rules exclude each other.
- The cost-band rules outrank baseline drift: if the price is unacceptable, that is the finding.
- Either fatigue rule suppresses efficient scale: more budget on a saturated audience re-buys the same people.
- Delivery-cost rules suppress click-through collapse on the same entity: broader reach raises cost and lowers click-through together, and the cause is the finding.
- Efficient scale is blocked by its pacing guardrail when the organization has no measurable plan: the guardrail is unknown, so it is not cleared.
The frequency ceiling and the zero-result rule check the campaign objective, because high frequency and zero leads are normal for awareness campaigns; a campaign with no recorded objective is skipped rather than judged. A result rate is null when the platform's results column was never mapped and zero when the platform reported nothing; only zero is a finding, which is why the zero-result rule's first remedy is to check the measurement.
How a playbook runs and is measured#
One evaluation path serves four callers: the audit card on the marketing performance page, the data assistant's audit tool, the 07:00 morning brief and the rule editor's dry run.
The scheduled scan runs at 06:30 every Monday, Vietnam time, over the closed Monday-to-Sunday week, so a finding does not appear and retract itself within a week. It writes proposals with no approver; approving, rejecting and activating remain an owner or org admin step with an audit-log entry (Decision engine screens). A flagged finding becomes a proposal only when:
- the organization has declared a conversion metric; without one no proposals are made, and the reason is stated;
- the finding is action-bearing, flagged, and its guardrails passed; anything blocked is counted with a typed reason;
- the conversion feed returned data for the window; otherwise the decision is recorded as learning, with the reason.
Leads resolve through one CRM qualified-lead definition. With no CRM feed connected the resolver returns nothing rather than zero, so a missing feed never scores a decision as a failure (Measurement and honest numbers).
Every rule's action is an audit finding with no lever attached, so nothing in the rule shape can move money or send anything outside the platform; editing a threshold cannot raise that ceiling. Where each autonomy level stands is on Project status.
Adopting and tuning#
- Read the catalogue and its track record in Marketing Systems OS > Industry playbook library ("Hệ vận hành Marketing > Thư viện playbook ngành").
- Adopt the defaults: seeding the default library follows the organization's planes, so a marketing organization receives the marketing catalogue, not a commerce one.
- Edit thresholds, evidence floors, baselines, quantile ranks, horizons and success rules in the rule editor, with a dry run over the last 30 complete days of the organization's data before saving. Every threshold needs a written reason. The editor is in the organization's platform console (Admin > organization > Playbook library), which a super admin opens; the server accepts the same edits from an owner or org admin.
- Versioning: an edit writes a new version; evaluation reads the highest one. Re-seeding never overwrites an organization's edits, and earlier versions stay readable for decisions made under them.
- On and off: turning a playbook off removes it from evaluation, not just from the list.
Automations#
AI Assistant > Automations ("Trợ lý AI > Tự động hoá"), /playbooks, is a separate tool for chaining operational actions, not a catalogue of marketing rules. On a canvas an admin picks a trigger (manual, schedule, on alert, after upload, after sync), drags in actions and connects them; saving compiles the chain into steps. Running an automation adds its steps to the Action Queue as proposals, where approval still applies. It is part of the Marketing Systems OS package and needs the AI management and automation playbook capabilities (owner and admin by default).