You Are Budgeting for Tools Before You Have Priced the Problem

Every publishing business has a category of work that nobody planned, nobody budgeted for, and nobody can see on a P&L. It is the checking.

Somebody logs into the ad server each morning to see whether campaigns are pacing. Somebody pulls numbers into a spreadsheet at month end because three systems disagree. Somebody notices on a Thursday that a campaign is going to miss its guaranteed impressions, and the make-good conversation starts. None of that work appears in a job description, all of it is paid for.

I spent twenty years running operations, most recently as group COO of a services business across six countries, and the checking is the most reliable thing I find. It’s a different sector, but the problems are the same. Work that exists only because a system cannot tell a person when something has gone wrong.

It is also, conveniently, the easiest waste in a business to put a number against. Which matters, because you cannot confidently decide what to automate until you know what the current situation costs.

The first mistake is to ask how the process is supposed to work, ask the staff what they did last week instead.

In practice this means sitting with the person doing the work and watching them. Ask them to talk you through the last month-end, what they do first thing on a Monday or what they check twice because they do not trust the number. That last question is usually the most productive one you will ask all week.

You are listening for three things:

1. Polling - Somebody looking at a system to find out whether anything has changed. Checking delivery pacing across live campaigns is polling. So is refreshing a dashboard.

2. Reconciliation - Somebody making two sources agree. Ad server vs the advertiser’s own numbers. Booked revenue vs delivered revenue vs invoiced revenue.

3. Re-keying - The same data typed into a second place. Campaign details from the insertion order into the ad server. Delivery numbers from a report into a client deck.

Each of these is a symptom of the same underlying problem: two systems that do not communicate, bridged by a person.

Now put a number on it. There are two components, and most people only count the first.

The labour. Hours per week multiplied by a fully loaded hourly rate, which is salary plus employer costs plus overhead.

Take an illustrative case. Two people spending five hours a week each on pacing checks, discrepancy chasing and report building. At a loaded rate of £40 an hour, that is £400 a week, or around £20,000 a year. Substitute your own figures, but be honest about the time. People consistently underestimate this, because the work is usually scattered in short blocks throughout the day.

The cost of finding out late. This is often the omitted piece, and it is usually the larger number.

Checking is not free even when it works, because checking is periodic and problems are not. If someone reviews pacing every morning, the average problem lives for half a day before anyone sees it. If the review happens weekly, the average problem lives three and a half days.

Now attach a consequence. A campaign that has fallen behind can be corrected cheaply in week one by widening targeting or adding placements. In the final week the only options left are expensive: a make-good, a credit, or over-delivering by taking inventory that another campaign needed. If that situation arises, say, six times a year, and the average cost of a late catch is £1,500 in make-goods and lost inventory, that is £9,000 a year attributable purely to detection lag, not to the underlying problem.

Again, use your own numbers. The point is the structure: Labour plus delay. Many businesses count the first and are baffled that automating it did not deliver what they expected.

By now you will have found more than one of these. Do the same sum for every piece of checking on your list, then rank them by what each costs you a year.

The ranking is the deliverable. It is worth more than any individual fix, because it turns an unbounded question, “where do we even start?” into a bounded one: here are our seven most expensive pieces of manual work, in order, and here is what each is worth.

Very little of the above needs AI, and it is worth being clear about that.

Most checking is replaced by exception alerting, which is ordinary automation. The data already exists in your ad server, and it has an API. A scheduled job can compare actual delivery against required pace overnight and tell someone only when a campaign drifts outside tolerance. The half day detection lag becomes minutes, and the labour disappears along with it.

AI earns its place higher up the stack:

1. Predicting which campaigns are likely to under-deliver before they visibly do.

2. Reading unstructured inputs, like a booking confirmation in an email, and turning them into structured data.

3. Drafting the narrative for a month end report from reconciled numbers.

That distinction matters commercially. If you go to market asking for an AI solution, you will be sold one, and you will pay AI prices for something a scheduled script could have done. Knowing which of your problems require which solution is worth more than any particular tool.

The checking is usually being done by your most experienced people. They do it because they have learned, correctly, that the system cannot be trusted to tell them when something is wrong. It is a rational response to an unreliable process.

Which means the conversation you need to have is not about whether they are doing valuable work. It is about the fact that the business is paying senior people to compensate for a gap that could easily be closed. Frame it that way and the people doing the checking usually become your best allies, because they have wanted to get rid of it for years.

Start by counting. You cannot make a sensible decision about automation, AI, or anything else until you know what today already costs you.

Jeremy Thorburn is the founder of LatenZ, an independent operations and AI advisory. He represents no vendor or platform.