What happens in the first month of an AI marketing build? A week-by-week guide.

01What happens when you hire an AI marketing agency?

In a well-run engagement, the first month turns a conversation into one working system. Week one produces a written scope and a positioning brief. Week two installs the first agent with a person approving its output. Week three runs it and measures it against a baseline. Week four hands it over with documentation and picks the next job.

An AI marketing build is a fixed project in which a provider sets up AI agents to run specific marketing jobs inside your business, starting from a written scope and ending with a documented handover. An AI marketing agent is the software doing one of those jobs, such as keeping your site visible in AI search or drafting a month of content, with a person reviewing what it produces.

This guide describes how a good first month runs in general, whichever provider you choose, with Flow AI's process as one example near the end. If you are still deciding what to buy, the guide to what AI marketing services include covers the menu, and the wider path is in the guide to AI marketing transformation for small business.

02What should you have ready before week one?

Have access ready for the accounts the first agent will use, and name one person who can make decisions each week. Bring the numbers you track today too, even rough ones, so there is a baseline. A provider can work around missing material. Quick access and a decision maker available each week are what stop a four-week build stretching into three months.

The account list depends on the first agent. For search work it is the website or CMS, Google Search Console, Google Business Profile and Bing Webmaster Tools. For content it is the social accounts and the scheduling tool. For outbound it is the email domain and the CRM. Grant access through user roles where the tool allows it, so you can remove the provider later without changing passwords.

If the agent will handle customer or prospect data, think about privacy before it starts. New Zealand's Office of the Privacy Commissioner expects organisations to complete a privacy impact assessment before using AI tools and to update it regularly¹. A good provider will raise this in week one and keep personal data out of any tool that does not need it.

03Week 1: the written scope and positioning brief

The first week ends with two documents you approve. The written scope names the first agent, what it may and may not do, who reviews its work and how success will be measured. The positioning brief sets out who you serve and the proof behind each claim you make, so every agent writes from the same facts.

Expect a working session of about an hour early in the week, where the provider asks about your customers, your best and worst months, and the marketing jobs that never get done. They should have read your website, reviews and competitors beforehand. By mid-week you have drafts to mark up, and by the end of the week, final versions.

The scope is the most important document of the month. It turns "help us with AI marketing" into a specific job, such as "keep our service pages crawlable and quotable, and report monthly on which AI assistants name us". The guide to what an AI marketing scope of work should include lists every section to check before you sign.

You provide: access, an hour of conversation, examples of writing you like and dislike, any claims you can prove. Your team receives: the scope and the positioning brief. A person approves: both documents, before any agent is built.

04Week 2: building the first agent, with a human review step

In week two the provider builds the first agent inside your accounts and sets up the point where a person checks its work. Nothing reaches a customer until that review step has been tested on real output. By Friday the agent has produced its first batch, and your reviewer has approved or sent back each piece.

Which agent comes first depends on the scope. A business that buyers cannot find usually starts with search. One with plenty of traffic and few enquiries starts with conversion. One that relies on referrals may start with outbound. The guide to AI agents for marketing sets out how to choose.

The review step is where most of the design work sits. It should say exactly what the agent may do on its own (draft, research, schedule internal reports) and what waits for a person (anything published, sent or spent). This is human in the loop in practice. The reviewer should be someone who knows the business, usually the owner or marketing lead, with a clear way to send work back.

You provide: a reviewer with 30 to 60 minutes this week. Your team receives: a working agent and its first output. A person approves: every item before it goes live, plus the review rules themselves.

05Week 3: run it and measure

Week three is a live run. The agent does its job on a normal schedule, your reviewer approves its work in the agreed way, and the provider watches for errors and tunes the instructions. At the end of the week you compare results against the baseline taken in week one, using the measure the scope names.

Keep the measure close to money. For search, that might be enquiries from organic search and whether ChatGPT or Perplexity names you for five agreed prompts. For outbound, positive replies and meetings booked. For content, pieces approved and published on schedule, and the enquiries they bring. Raw output counts, such as posts drafted, say little on their own.

Most marketing results take longer than three weeks to judge, and an honest provider will say so. What week three can show is whether the agent runs reliably and the review step catches its mistakes. Early signals are a bonus. Ask for a written note of what was changed during the week and why.

You provide: approvals on time. Your team receives: a first results note against the baseline. A person approves: each piece of customer-facing work, and any change to the agent's rules.

06Week 4: handover documents and the next agent

The last week moves the agent into your team's hands. You receive documentation that explains what the agent does, how to run it, the files and settings it uses, the review rules and what to do when something breaks. Then you agree which job to build next, based on what the first three weeks showed.

A handover session of an hour, recorded if possible, should walk the person who will own the agent through a full cycle. Check that every account, tool subscription and file sits in your business's name. If the provider leaves tomorrow, you should be able to keep the agent running from the documents alone.

The choice of second agent should come from evidence. If search is working and enquiries now arrive faster than you can answer them, conversion or reporting may come next. If content is drafted but nobody publishes it, the gap is process, and the next build should fix that first.

You provide: the owner for handover, and a decision on the next job. Your team receives: the documentation, the account list and a scope for the next agent. A person approves: the handover as complete, and the next scope.

07The first month on one table

The table below puts the four weeks side by side: what you hand over, what comes back and what a person signs off. Use it as a checklist when you talk to providers. If a proposal cannot show you something similar, ask how they handle each row, especially the approvals column.

08Who approves what during the build?

A person who knows the business approves anything a customer will see or that spends money, along with any change to what the agent is allowed to do. The agent can research, draft, sort and report on its own. Writing those boundaries into the scope protects your reputation and keeps the agent useful while it learns your standards.

The evidence for keeping a person in the loop is plain. KPMG and the University of Melbourne's 2025 New Zealand snapshot, surveyed over summer 2024-25, found 46% of New Zealand employees had made mistakes in their work due to AI and 51% had relied on AI output without evaluating its accuracy². Customers notice too: Constant Contact's 2026 research with Australian and New Zealand small businesses found 27.6% said customers had raised concerns about data privacy and a lack of human oversight³.

Some approvals carry legal weight. For email and text messages, New Zealand's consumer protection guidance on the Unsolicited Electronic Messages Act says commercial messages need consent, accurate business contact details and a free unsubscribe that works within five business days⁴. A reviewer checking outbound work should check those as well as tone.

09How can you tell whether the build is on track?

A build is on track when each week ends with the agreed output and your reviewer approves most of the agent's work with light edits. Warning signs include vague weekly updates, changes nobody explains in writing, an agent that went live before its review step was tested, and a handover date that keeps slipping.

Plenty of AI projects stall. Gartner predicted in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls⁵. Smaller organisations find scaling hard: McKinsey's State of AI 2026, surveying 1,719 participants in mid 2026, found the share of smaller organisations reporting scaling AI agents stayed flat at 22%⁶, against 40% of large organisations.

The cure for both is the same: one agent, a clear measure, a tested review step and a real handover. Relationships also fail when the provider never learns the business. Setup's 2025 Marketing Relationship Survey, a small and mostly US sample of about 100 responses, found 44% of clients who ended an agency relationship said the agency did not understand their business⁷. Week one exists to prevent that. The guide to choosing an AI marketing provider lists the questions to ask before you start.

10Where Flow AI fits

Flow AI provides AI marketing services for businesses in New Zealand and Australia. Amber Lan, an AI marketing engineer in Auckland, builds the AI agents that run search, content, outbound and reporting with your team, then hands them over. Its process follows the shape above, compressed: each sprint builds one agent in a week.

The first step is a written scope agreed inside week one. Each sprint runs Monday to Friday with six fixed deliverables and a documented handover, and a marketing engineer reviews every send. Builds are priced per project, and your team keeps the agent. The overview of Flow AI's marketing services has the detail, and companies hiring can reach Amber at the hiring page.

11The short version

A good first month of an AI marketing build has one output per week: a written scope and positioning brief, then a first agent with a tested review step, then a measured live run, then a documented handover and the choice of the next agent.

Your part is access, a decision maker and an hour or two of approvals each week. Keep every account in your name, and make sure a person signs off anything a customer will see.

To plan your first month together, talk to Amber.