A decision guide for business leaders and marketing teams
Many companies begin an AI marketing agency engagement with high expectations. A month later, they have attended kickoff meetings, received an attractive report and approved a long activity list—yet still cannot say what is complete, what has passed quality control or what evidence should guide the next decision. The problem is not always a lack of effort. It is often the absence of an agreed definition of what the first 90 days must build.
What should an AI marketing agency deliver in the first 90 days?
A deliverable should not mean only a number of posts, campaigns or report pages. It should be something the client can inspect, use and retain. Strong delivery connects three layers: the output, such as a landing page or content asset; evidence that it works as specified, such as tracking tests and QA records; and the decision it enables, such as scaling, repairing a funnel or stopping an unsupported assumption.
Before execution accelerates, the agency should demonstrate that it understands the actual growth constraint rather than assembling a service menu. When priority remains unclear, the appropriate route is the Customer Growth Blueprint. When the constraint is already known, the engagement can move into a focused installation roadmap.
Vault Mark 90-Day Deliverable Acceptance Matrix
This matrix is Vault Mark professional methodology, not an industry benchmark or universal contract standard. Adapt it to the agreed scope, available data, internal capacity, channel mix and sales cycle.
| Delivery layer | Expected deliverable | Acceptance evidence | Decision enabled |
|---|---|---|---|
| 1. Direction | Business problem, 90-day objective, priorities and explicit exclusions | Approved decision brief with assumptions and owner | Are both parties solving the same constraint? |
| 2. Baseline | Starting state for traffic, leads, journey, content, access and data gaps | Dated snapshot with sources and limitations | How will change be distinguished from noise? |
| 3. Measurement | Event/key-event map, UTM rules, dashboard and QA plan | Realtime, DebugView or platform diagnostic evidence | Is the data reliable enough to decide? |
| 4. Production | Live campaigns, pages, content or workflows within scope | URL or asset, version, approval, QA and ownership record | Is the work usable and fit for purpose? |
| 5. Learning | Experiment register, findings, failed assumptions and next tests | Hypothesis, method, observed result and limitations | What learning changes the next action? |
| 6. Governance | RACI, cadence, escalation, AI/tool register and change log | Named owner, reviewer, approver and history | Who is accountable when work or data fails? |
| 7. Decision | 30/60/90-day reviews and continue/change/stop recommendations | Decision memo separating fact, inference and recommendation | Where should budget and capacity go next? |
Days 1–30: establish shared truth before accelerating production
1. An approved decision brief and scope boundary
The first substantive document should state the business objective, the suspected constraint and what the 90-day engagement is intended to prove or repair. It should also state what is outside scope. This prevents “AI marketing” from being interpreted as every channel and tool at once. The broader AI Marketing Solutions architecture can provide context, but the 90-day plan should remain selective.
2. A baseline and access inventory
The agency should record account ownership, access levels, tracking status and starting metrics without overstating incomplete data. Acceptance requires more than a spreadsheet: each metric needs a source, date range, definition and known gap. The client should also retain appropriate ownership and administrative control over core accounts and assets.
3. A measurement plan that has been tested
Google Analytics 4 allows important actions to be marked as key events and checked through Realtime or DebugView, according to Google’s key-event reporting documentation. A credible deliverable therefore includes an event dictionary, trigger, owner and test evidence—not only a dashboard screenshot. Where teams have multiple conflicting reports, the AI-Data & Measurement OS explains the need for a shared signal spine.
Days 31–60: ship usable work and begin testing assumptions
By this stage, the client should see live assets aligned with the selected priority: perhaps a landing page, campaign structure, content cluster, CRM handoff or reporting workflow. Not every possible asset needs to launch. Each delivered item does need an acceptance test—for example, approved messaging, mobile usability, correct event firing, accessible interaction and a named owner.
For paid media, conversion measurement should represent actions that matter to the business rather than clicks alone. Google Ads defines conversion tracking around valuable actions such as leads or sales following an ad interaction. Launching campaigns before conversion definitions and QA are complete can produce traffic while leaving the organisation unable to judge its value.
For search work, useful first-90-day outputs include query ownership, page briefs, internal-link plans and published URLs—not an oversized keyword list. Vault Mark’s AI Search Content Factory describes a cluster-first flow that connects briefs, drafts, FAQs and answer blocks while keeping truth and final judgment under human control.
Days 61–90: turn evidence into a decision, not a larger slide deck
The 90-day review should separate outputs (what was completed), signals (what began to change) and business outcomes (effects on lead quality, pipeline, revenue or cost). Completing the work is not proof of commercial success. Equally, a final outcome may be premature when the buying cycle or search horizon is longer than 90 days.
The decision memo should offer three explicit paths: continue work supported by evidence; change work where the hypothesis remains plausible but execution or data is insufficient; and stop work that is disconnected from the priority or carries excessive opportunity cost. The AI Experimentation OS is relevant where teams need hypotheses, stop rules and learning records rather than unstructured testing.
What additional deliverables are required when the agency uses AI?
- AI and tool register: tool name, workflow stage, permitted data and approver.
- Human review gate: named fact, brand, legal/compliance and bilingual reviewers.
- Source log: source, review date, supported claim and limitation.
- Version history: draft, revisions, approver and reason for material changes.
- Ownership record: account, prompt asset, source file, audience and published-asset rights.
AI can accelerate analysis and production, but it cannot become the accountable owner of accuracy, judgment or risk. When the operating workflow itself is weak, the AI-Ops OS provides a clearer view of how people, tools and handoffs should work together.
Practical scenario: lead volume rises, but sales says the leads are wrong
Consider a B2B company that hires an agency to generate demand through search and content. A dashboard may show more form submissions and make the programme look successful. The acceptance matrix forces a deeper check: Does the form capture qualification information? Does the CRM retain source and campaign data? Does sales update qualified and disqualified statuses? Is that feedback returned to the keyword, message and landing-page decisions?
The day-90 decision may therefore be neither “scale immediately” nor “fire the agency.” It may be to repair the qualified-lead definition, CRM handoff and landing-page message, then run a better test. This is consistent with the AI Lead Generation principle that lead quality and downstream handling matter more than raw form volume.
Mistakes that make the first 90 days busy but inconclusive
- Starting every channel before selecting the primary constraint.
- Counting outputs without acceptance criteria or QA evidence.
- Treating a dashboard as the main deliverable while event definitions remain disputed.
- Using AI to increase volume without sources, reviewers or an approval trail.
- Changing KPIs without documenting the reason and baseline impact.
- Allowing the agency to control core accounts or assets without a clear owner/admin model.
- Waiting until day 90 to surface issues instead of reviewing at days 30 and 60.
Ten questions before approving month four
- Is the original constraint still the most important one?
- Which deliverables are complete but not yet acceptable, and why?
- Where did the baseline or measurement definition change?
- What is live, and who owns the account and source files?
- Which findings are observed facts and which are inferences?
- Where was AI used, and who owns the final judgment?
- What has been learned, and what remains an assumption?
- What should continue, change or stop?
- Which client-side dependency must be removed next?
- What evidence justifies the next allocation of budget and capacity?
Your next decision: request a roadmap or diagnose the priority first?
If the organisation already knows its constraint, request a 90-day roadmap with deliverables, owners, acceptance criteria, measurement and review cadence. If leaders still disagree on whether SEO, ads, content, CRM or the website should come first, avoid buying several implementation packages at once. Start with the Customer Growth Blueprint to clarify the problem, evidence and sequence.