You have sat through four AI pitches since June. One was a chatbot, one was a content tool, one was a whole platform with a per-seat price, and one was your existing vendor renaming a feature they already had. Every one of them was confident. None of them told you which job in your business it was replacing, or what happens the day it gets something wrong. You are not trying to understand how a model works. You are trying to work out whether the next subscription earns its place or just joins the other eleven on the credit card statement.
Start with the job, not the tool.
The first question is not what the tool does. It is what job is currently being done badly, late, or not at all. Write the job down as a sentence with a verb in it. “Somebody replies to the form fills from the website.” “Somebody writes the monthly email.” “Somebody checks the list before it goes out.” If you cannot write the sentence, there is no job, and a tool applied to no job does not produce a saving. It produces a new expense and one more thing to check on Friday.
This sounds obvious. It is the single most common reason AI spending disappears without a trace. A vendor demonstrates something impressive, you buy it, and it sits alongside the work rather than inside it, because nobody was doing that task in the first place and nobody is now accountable for its output. The saving was never real. The subscription is.
So before the demo, make a short list of the jobs in your marketing operation and mark the ones that are genuinely done badly. Late replies. Drafts that never get finished. A list nobody has cleaned since 2021. That list is your assessment criteria. Anything a vendor shows you that is not on it is theatre, however good it looks.
The four places it earns its keep.
In a small marketing operation — one or two people, an owner who does marketing between other things — there are four jobs where we have seen this technology pay for itself reliably. Not four categories. Four jobs, each with a person still attached to it.
Notice what they have in common. In every one, the machine does the first pass and a person does the last. None of them end with something going out unread.
- Answering an inbound lead in minutes rather than days. A draft reply, written against the enquiry, sitting in a queue for someone to read and send. The value is not the writing. It is the speed.
- Getting a first draft onto the page. The work then starts at editing instead of at a blank document, which is the point at which most monthly emails die.
- Reading back what was produced and checking it against the source. Dates, prices, hours, spellings of names, links that resolve. A second reader that never gets bored.
- Sorting a list so a send goes to the right people. Who bought what, who has not opened anything in a year, who moved, who is a supplier and not a customer.
The arithmetic on a faster reply.
Do this with your own numbers, not ours. Take last month’s inbound enquiries — form fills, calls you missed, emails to the general address. Say there were 40. Say you closed 8. That is a 20% close rate. Now ask how long the average first reply took. If the honest answer is “a day or two, and some on Monday,” you have a measurable problem, because the well-known Harvard Business Review work on lead response time found that the odds of qualifying a lead fall steeply once you are past the first hour. We are not going to attach a percentage to your business that we did not measure in your business.
Here is the shape of the arithmetic. 40 enquiries, 8 closed, average job worth $6,000. That is $48,000 of work from that month’s enquiries. If replying within fifteen minutes moves the close rate from 8 in 40 to 10 in 40, the month is worth $60,000 instead of $48,000. Twelve thousand dollars, from replying sooner. Against a tool at $150 a month, the question answers itself — and it stays answered even if the improvement is half what you hoped.
Now run it the other way, because that is the discipline. If you get 6 enquiries a month and you already answer them within the hour because they come to your phone, the same tool returns nothing. Same technology, same price, no job to do. This is why the job comes first and the tool second.
The arithmetic on a first draft.
A drafting tool is easier to assess than people expect, because the input is hours and you already know what an hour costs you. Time yourself writing one monthly email from a blank page, start to send. Most owners we have watched land somewhere between 90 minutes and three hours, with the largest part of it spent staring at nothing. Then time the same job starting from a draft you edit. If it comes down from two hours to forty minutes, you saved eighty minutes.
Put a price on it. A $200-a-month tool that saves eighty minutes on one email is costing you $150 an hour for the time it gave back. That is a bad trade for most people. The same $200 tool used across four emails, twelve social posts and a set of lead replies might save six hours, which is $33 an hour. That is a good trade if your own time is worth more than $33, and yours is.
So the test is not whether the output is good. The test is volume. Drafting tools pay off when there is a lot to draft and lose money when there is not. If you publish once a month, the honest answer is that you do not need a subscription. You need forty-five minutes in the calendar that nobody is allowed to take.
Where it quietly makes things worse.
There are two places we will not go, and they are both variations of the same failure. The first is publishing without a person reading it. The second is any job where being confidently wrong costs more than being slow. A model will produce a number, a date or a price that looks exactly right and is not. It does not hedge. It does not flag the guess. That is the whole risk, and no amount of setup removes it — it only moves who is responsible for catching it.
We learned this on our own work. A while back we had a drafted email queued for a client’s list, and two hours before the send one of us caught a discount in the second paragraph that the client had never agreed to. The model had put it there because a previous campaign had one, and it read perfectly. We caught it because a person reads every send line by line against the source. If we had been running that queue on a schedule with nobody reading, it would have gone out to the whole list with our name on the footer.
So the rule at this desk is simple and it is not negotiable. AI drafts, sorts, checks and suggests. A person approves every send. Anything a vendor sells you that ends in publication with no human in the loop is not a saving of your time — it is a transfer of your risk onto a system that cannot be held accountable for it.
How to assess a tool in an afternoon.
You do not need a trial period, a committee or a technical review. You need five questions and a straight answer to each. Ask them in the demo, out loud, and write the answers down. A vendor who cannot answer them in plain English is telling you something.
Score it honestly. Two soft answers out of five and you are buying a subscription, not a capability.
- What does it replace? Name the job and who does it today. If the answer is “it adds capability,” there is no job.
- What does it cost per month against the hours it saves? Divide one by the other. You now have a dollar-per-hour figure to compare against your own.
- Who reads its output before anyone outside the company sees it? A name, not a role.
- What happens the day it is wrong? Who notices, how fast, and what is the cost of the wrong thing having gone out.
- Can you leave? Where does your list live, can you export it today, and what is the notice period. Ask for the export while you are still on the call.
What to measure afterwards.
Once it is in, measure the job, not the technology. “AI adoption” is not a number that pays anybody. Three things are worth writing on a whiteboard and checking monthly: median time from enquiry to first reply, pieces published on schedule versus planned, and hours the owner personally spends on marketing. All three you can count without a dashboard.
Take a baseline before you buy anything. Go back through last month’s enquiries and record the reply times — it takes about twenty minutes and it is the single most useful twenty minutes in this whole exercise. Count how many of the things you planned to publish actually went out. Ask yourself, or your marketing manager, roughly how many hours went into it. Now you have a before, which means in ninety days you will have an after.
If the reply time did not fall, the tool did not work, no matter how good the drafts read. If publishing is still slipping, the constraint was never drafting — it was approval, or photography, or the fact that no one owns the calendar. That is a useful finding, and it is cheaper to discover in ninety days than in two years.
Where we come into this.
We are two partners in Brentwood. We have been doing this since 2014, we use these tools every day, and we say so openly — and a person reads every send before it goes. We are not selling you a platform, because we do not have one to sell.
This is what a Growth Assessment does. Two weeks, fixed price, credited toward any build. We look at how you actually get customers today, where the time goes, and which of those four jobs — the reply, the draft, the check, the list — is costing you the most. You get back three prioritized moves with the arithmetic under each one. Fairly often one of those moves is that AI is not the answer and a fifteen-minute change to who answers the phone is. We would rather say that in week two than bill you for it in year two.
The details are here: etchmarketingservices.com/growth-assessment.html. If the job that is done worst is a list nobody has mailed in two years, that is a specific piece of work with specific arithmetic, and it lives here: etchmarketingservices.com/customer-reactivation.html. If you would rather just ask a question, a real person answers at etchmarketingservices.com/contact.html.
The smallest next action.
Pick the one marketing job that is done worst in your business today. Not the most important one. The worst one. Then time it. Actually time it, with a phone on the desk, from the moment you start to the moment it is finished and out the door.
That number is the beginning of every assessment worth doing. It tells you what the job costs now, which tells you what any tool has to beat, which tells you whether the next pitch is worth an hour of your attention. It also has a habit of embarrassing everybody into fixing the obvious thing first, which is usually free.
Do it this week. Then go back to the vendor with the five questions and see how the conversation changes when you are the one holding a number.