What It Costs to Run a Marketing Function as a System: Our All-In Numbers.

Ask a mid-market company what its marketing costs and you get a salary, an agency retainer, or a media budget. Ask what a blog post costs, or a campaign, or a qualified lead, and the room goes quiet. Marketing is one of the few functions a company will fund for years without being able to price a single unit of its output.

We decided we would not run ours that way. Since the summer we have run CodeStringers' marketing as one governed system on Zoho, with AI agents doing the production and a person approving every dollar. One of the design rules was that every piece and every campaign has to carry its all-in cost, including the cost of the AI that made it. This post is what that looks like in September 2026, with the numbers we actually pay.

The enemy is a cost nobody can trace to a result.

Marketing bought as headcount or as a retainer has a structural problem: it is judged on activity because activity is the only thing it can report. Posts published, emails sent, impressions bought. The cost side is a lump sum and the result side is a dashboard of counts, and nothing connects them.

That is not a people failure. It is what happens when the function runs across an agency, a CRM, an ad account, a blog and an inbox that do not agree with each other. The agency owns the creative, the admin owns the CRM, finance owns the budget, and nobody owns the number that matters: what did it cost to produce a qualified lead, and which piece produced it.

If you are weighing a marketing automation consultant, an agency or a hire right now, the question underneath the decision is the same. Whichever you choose, will you be able to price the output afterwards?

The missing piece is a cost table, not a cheaper vendor.

A system can carry its own cost because every action in it is logged. Ours writes a row for every run an agent makes: which agent, how long it ran, what it produced. Fixed costs are pooled by month and spread across those runs by the minutes they used. Metered costs, like a video render, land directly on the piece that caused them. Media spend is already attributed by campaign. Add the three together per piece and you have an all-in cost you can defend.

Here is what sits in that table for September 2026.

• AI seats: four Claude seats used by the marketing system, two Team Premium and two Team Standard, at 300 dollars for the month. This is the pool that pays for theme scoring, program planning, drafting, image generation, publishing, ads operations and measurement.

• Video tooling: a HeyGen plan at 100 dollars for the month, plus render credits that land on each video directly. The most recent batch of four videos carried 26 dollars and 51 cents of render cost between them.

• Paid media: 1,145 dollars and 56 cents spent on Google Ads from 1 January to 10 September this year, about four and a half dollars a day. September is running at about 31 dollars a day against roughly 122 dollars a day of enabled budget, because most of what the campaigns are allowed to spend has not been spent yet.

Before media, the tooling that runs the whole function costs 400 dollars a month. That is the number to compare with a salary or a retainer, and it is the number most companies have never seen for their own marketing, because the tools are scattered across departments and the labour is invisible inside headcount.

What the table leaves out, on purpose.

Two things are missing, and we would rather say so than let the number look smaller than it is.

The platforms the business already pays for are not in the marketing pool. Zoho One, the website platform, the search-data subscription and the office suite were carried by the company before the system existed and would be carried without it, so we do not attribute them to marketing. A company adopting this from scratch should add its own licence line.

The founder's time at the gates is not in the pool either. Every budget, bid and campaign status change waits for a person to approve it, and that person's hours are real. We have not measured them yet, so they are not in the table. When we have, they will be.

What the numbers do for us.

The point of a cost table is not the table. It is what you can decide with it.

• We can price a piece. When the month closes, the 400 dollar pool spreads across what the system produced. In the first week of September that was ten long-form articles in a single batch and a run of avatar videos; on that kind of output the tooling cost of one article is measured in single or low double digits of dollars before media and before human review.

• We can kill a program on cost. Every program has a hypothesis, a window and a kill rule, and the cost side of the kill rule is a real number rather than an allocation.

• We can hand the ad platform the truth. Qualified leads in Zoho CRM go back to Google Ads as offline conversions, so the campaign's cost per qualified lead is computed on leads a person qualified, not on form fills.

• We can see the delivery gap. The fact that September's campaigns are spending a quarter of what they are allowed to is a diagnosis, not a saving, and the system surfaces it daily rather than at quarter end.

None of that is a feature of any single application. It is what you get when the whole function runs on one set of facts, and it is the reason we built AI Marketing OS around the cost table rather than around the content.

When not to do this.

If your marketing is one person who knows the customers by name and produces a newsletter a month, a system will not pay for itself. Keep the person. Give them a CRM that agrees with the books.

If your problem is that you do not know what to say to the market, no operating model fixes that first. The system scores themes against evidence, but it needs a business that can articulate what it is for.

If you want an agency because you want their taste, buy the taste. Just ask them to report cost per qualified lead, and notice how the conversation changes.

How we know this.

These are our own numbers, from the tables the system writes as it runs, and they are the same numbers we would show a client. The commercial structure is the one we apply to every engagement: discovery is no-risk, so you pay for it only if you proceed, and the estimate is guaranteed, so if we underestimate the work of setting this up on your Zoho, the difference is ours. That structure is only possible because we can price our own work, which is the argument of this post applied to ourselves.

We also build the pieces separately. If the question is narrower than the whole function, our Zoho Marketing Automation practice sets up the sends, the scoring and the journeys, and managed technical operations keeps them running with a named party accountable.

Where this goes.

Agentic AI is going to make marketing cheaper to produce for everyone. That is not the interesting part. The interesting part is that it makes the function priceable for the first time: every action logged, every dollar gated, every piece carrying the cost of the seats, the tools and the media that made it, joined to the qualified pipeline it produced.

Once you can see that number, you stop asking whether to hire, retain or automate and start asking which pieces earn their cost. We will publish our numbers again when the month closes, including the ones that go the wrong way. If you want to see what your own marketing would look like priced this way, a no-risk discovery is where we start: we read your CRM, your ad accounts and your content and show you what a governed system would do differently.