AI Agent for Marketing at a Small Team: What It Reports and Changes from Slack, What You Approve, and When You Still Need a Marketer

If your team spends on Google Ads without a marketer, Monday starts with a chore. Someone has to open Google Ads, then Analytics, then Search Console, pick last week's dates in each, and paste the numbers into Slack. When that person is the founder, it slips. A campaign overspends for a week, a page quietly loses its clicks, and nobody notices until the month closes. An AI agent for marketing can take the reading part of that job off a person, from the Slack thread where the team already talks, and it can make a small set of changes when someone asks and approves them. This guide shows which weekly marketing chores fit an agent, which step waits for your approval, and which work still needs a marketer you hire.

The short answer

For a 5 to 100 person team that runs its own ads, the split looks like this:

1. Let the agent read. Weekly Google Ads spend by campaign, the search terms that spent the most, Analytics active users and Search Console queries and clicks are all reads. They are the bulk of the Monday chore, and they are safe to schedule.
2. Approve every change. Pausing a campaign, enabling one or changing a daily budget is a change in a live ad account. Ask for it in the thread, check the card, and approve it yourself.
3. Keep judgement human. Deciding what the numbers mean, what to try next, and writing the ads, the posts and the pages stays with a person.

If you only do one thing after reading this, set up the weekly report as a read-only scheduled request and leave every change as a live request.

Why the weekly check keeps slipping

The checks themselves are not hard. A practitioner checklist for what to do in a Google Ads account each week is short: read the search terms report and add negatives, note converting queries, check spend pacing against the monthly budget, confirm conversion tracking is firing, and leave campaigns in "Learning" alone. The same author puts a single-campaign small account at "thirty to forty minutes a week and two hours a month".

The time is small. The work is spread across three logins and needs the same date range picked three times, so it gets done on good weeks and skipped on busy ones. That is when a founder starts looking at a marketing coordinator or a virtual assistant, because the job recurs every week and nobody owns it.

The founders who do look often look too often. The same checklist warns: "Changing something because you looked is what costs money." A founder who opens the account on a bad Tuesday and cuts a budget can undo a week of learning. A routine that keeps reading and changing apart prevents that.

What the current AI marketing agents assume

Search for an AI agent for marketing and most results describe a different buyer. HubSpot's overview says AI agents for marketing show up in four key areas: content creation, customer support, campaign optimization and data analysis. It adds, fairly, "Mind you: AI agents don't replace strategy or creativity." Salesforce defines them as powerful software systems that can automate a variety of marketing tasks, and its examples are briefs, emails, landing pages, audiences and journeys built inside its marketing cloud, with marketers refining the output before use.

Those are reasonable descriptions for a team with a marketing platform and people inside it all day. They say little to a ten-person firm whose marketing is one Google Ads account, a website and a CRM, and whose conversation happens in Slack. That team needs the Monday numbers to show up, and someone to notice when they look wrong.

There is a third option between a platform agent and a hire: a reporting tool. One vendor guide to automating a weekly marketing report from GA4, Google Ads and Search Console connects the three sources, schedules a Monday report to email or Slack, and can send only when a condition is met. If all you want is a fixed weekly dashboard and nobody ever asks a follow-up question, a reporting tool fits. The case for an agent is the follow-up: "why did spend jump on the brand campaign?" asked in the same thread, answered from the same data.

Which marketing chores an agent can run from Slack

The table below covers the recurring chores a small team usually hands to a coordinator. The "Kipwise agent" column is limited to what our agent does today with each tool connected; the last column says when something else is the better choice.

The Google Ads rows touch money. With Google Ads connected, the agent pulls campaign performance, keywords and search terms, and when you explicitly ask it can pause or enable a named campaign or change its daily budget, provided that budget is not shared with other campaigns. Enabling a campaign starts spending that budget, so treat "turn it back on" as a spending decision. Google itself allows frequent edits: you can change your average daily budget at any time and as many times as you like. Because edits are that easy, the approval step should sit with a person.

Analytics and Search Console access is read-only. The agent reports top queries, clicks, positions or active users over a period, states the date range next to every figure, and only reports what it actually fetched. It can also list the sitemaps Google already has. It cannot change anything in either property.

With HubSpot connected, the agent looks up people, companies, deals and who owns what, using your own CRM access. When you explicitly ask, it can create or update a record, log a note, add a task or move a deal.

A worked example: the Monday report thread

This example is made up, but every step uses a capability described above.

On Monday at 9:00 the agent posts in #marketing, from a scheduled request a teammate set up the week before: last week's Google Ads spend by campaign, the five search terms that spent the most, Analytics active users for the week and the top Search Console queries, each with its date range.

Dana, the founder, reads it and replies in the same thread: "Brand Test spent 40 percent more than usual. Which search terms drove that?" The agent pulls the search terms for that campaign over the same dates and lists them. Two of them are competitor names nobody meant to bid on.

Dana replies: "Pause Brand Test until Thursday's call." The agent shows an Approve card in the thread. Only Dana, who asked, can press it, and the card expires after 30 minutes. Dana approves, the campaign pauses, and the thread now holds the numbers, the question, the reason and the change in one place.

Adding the competitor names as negative keywords, and deciding whether Brand Test comes back at all, stays with Dana and whoever runs the ads on Thursday. The agent did the reading and made one approved change.

Keep the Monday report read-only

In Kipwise, before a live request changes something in a connected service or posts a message, the agent shows an Approve or Deny card, and only the person who asked can use it. Scheduled requests work differently: creating a scheduled task needs approval once, and its later runs do not ask again.

So keep the Monday report read-only and ask for budget changes live, because a scheduled task is approved once and its later runs do not ask again. If you write a schedule like "every Monday, report spend and pause any campaign over budget", you approve the pausing once and then it happens every week without a card. That might be what you want for a mature account with clear rules. For a small team still learning what normal looks like, it removes the one moment where a person looks before money moves.

The practical version: the schedule says "post", never "pause" or "change". Every change starts as a reply in the thread, where a person reads the numbers first and approves the card. This matches the general advice in Anthropic's guide to building effective agents, which recommends human checkpoints and finding the simplest solution that works before adding autonomy.

Ask for the weekly figure, not seven daily ones

A weekly report often gets built by adding up daily numbers, because that is how a dashboard export arrives. For users, that is wrong: one person who visits on Monday and Thursday counts twice. The reporting-tool guide above tested this on its own data and found that adding up daily users overstated weekly users by 24.3%, and its rule is "Never add daily users, or weekly users, together."

Our recommendation: ask for the weekly figure as a weekly figure, because weekly users are not the sum of seven daily counts. Write the scheduled request with the exact period, for example "active users for the week Monday to Sunday", and check that the agent states that date range next to the number. Spend and clicks add up safely; users and sessions per user do not.

Search Console has its own timing catch. Its help page notes that the newest data can be preliminary, meaning it is still being collected and might change in the next few hours. A report that runs at 9:00 on Monday about a week ending Sunday night may show a short last day. If Sunday always looks low, end the reporting week a day earlier or treat the last day as provisional.

Separate looking from changing

The worked example has two steps. First the numbers arrive and someone reads them. Then, if something needs to change, a person asks for that change on purpose and approves it.

We recommend you separate looking from changing: the weekly report is where you read the numbers, and a budget change is a second, deliberate request with its own approval. This is the discipline the practitioner checklist asks of human ad managers, written into the workflow instead of left to willpower. It also gives you an audit trail you did not have before. Each change sits in Slack next to the numbers that prompted it and the name of the person who approved it.

If a change was a mistake, the fix is another request in the same thread: "enable Brand Test again", approved the same way. Remember that enabling starts spending, so check the daily budget in the reply before you approve.

When you still need a marketer

An AI agent for marketing in this setup is an operations helper. Hire a person, or bring in an agency, when:

• The work is mostly creating. Ad copy, landing pages, social posts and design need a person or a writing tool. The agent described here does not write or post them.
• You need a client-facing report. Agencies that send branded monthly reports to clients need a reporting tool built for that, not a Slack thread.
• Nobody can judge the numbers. A report that arrives every Monday is useless if nobody on the team knows whether a 40 percent jump is good or bad. An agent can surface it; a person has to decide.
• The account is large and fast-moving. Accounts with many campaigns and daily optimisation need someone who lives in Google Ads, plus Google's own bidding tools.

Many small teams sit in between. They do not need a full-time marketer yet, but they need the reading and checking done every week. That is the part to hand off first, before you post a job ad.

What it costs and how to check it worked

The Kipwise agent is priced per user, with included credit that pays for AI usage at cost; current plan and trial details are on the pricing page.

Three weeks in, check three things:

1. Did the report arrive every Monday, with date ranges on every figure? If a run failed, ask the agent what happened on previous runs; it can list your scheduled jobs and their failures.
2. Did anyone reply to it? A report nobody questions is a dashboard. If the thread is silent, the report is too long or covers the wrong numbers.
3. Did every change go through a card? Scroll the thread. Each pause, enable or budget change should appear as a request from a named person followed by their approval.

What to do next

Connect Google Ads, then ask the agent in your marketing channel: "Every Monday at 9, post last week's Google Ads spend by campaign and the five search terms that spent the most, with the date range." Approve that one schedule, read the first report next Monday, and make your first change as a reply in the thread. For the approval mechanics behind that card, see how human approval works for AI agent workflows in Slack, and for the other chores worth handing off first across the business, read AI agents for a small business: what to hand off first.