Tim asked three questions on the 13 August call. They were aimed at Cody, but they are the right questions to ask about our own software — so here are the answers, in plain English, with the numbers. Including the one that is zero.
Honest answer: there is still a person doing a lot of the work.
A marketing person does six jobs. Our software does the first three well, and struggles with the last three — which happen to be the three that matter to a business owner, because they are the ones that put work in front of customers and prove it paid off.
Find the people who are actually in the market, rather than blasting everybody.
For exampleIt watched 103 company pages on LinkedIn and collected the 2,004 people who liked or commented on their posts. Then it sorted the buyers from the hangers-on and kept 207 — a purchasing manager at KB Home, a director of production at LGI Homes. Not the sales reps from rival suppliers, who follow everybody.
This part genuinely works.
Know the words the industry uses, so what you write does not sound like an outsider wrote it.
For exampleIt reads what companies in a trade actually post about and learns from it. For magnetics it picked up “rare earths” and “critical minerals” on its own. For the windmill company it worked out the trade is water wells and pumps — which it had previously got wrong, and was describing them as if they sold marketing software.
Fixed this week. It now knows 76 different trades.
The ads, the pages, the posts. Not a plan to write them — the words themselves.
For exampleFor the windmill company it wrote a Google ad inside Google’s real 30-character limit — “Water Wells in Texas”, “Local Water Wells” — plus a Facebook post, a 20-second video shot list, and nine complete web pages now live on their site — an FAQ, a service-area page, and a page for each of the six things they sell.
This part works, and it is markedly better than it was a week ago.
Publishing. Getting the work out where customers will see it.
For exampleIt can add pages to a website by itself, and it did — nine of them this week. The connector that hands social posts to our publishing hub was also switched on and proven end to end. It still cannot place an ad on Google, send an email, or update a Google Business listing, and it cannot post to a client’s own social accounts until those accounts are connected — a one-time job per client.
Two channels out of about seven. Still the biggest gap.
Knowing that the thing you paid for brought somebody in.
For exampleIf someone taps the phone number or the email link on any of our twelve client sites, we record it and we know whether they arrived from Google, Facebook or a search. We do not know whether they actually rang, whether anyone answered, or whether a job came out of it. For a trade, most enquiries are a phone call at seven in the morning.
Now on all twelve sites that have a contact link. It has recorded nothing yet, because no campaign has run.
Look at the numbers and kill what is not working, spend more on what is.
For exampleThe software knows how to say “that advert is costing $32 for every customer, you said your limit was $5, so I am stopping it”. It also refuses to spend a penny until the free work is done, and it will not go over a monthly limit even if you approve it.
All of that is built and tested — and has never once had a real number to work with.
The software is good at working out what should be done and poor at doing it and proving it worked. If you hired it as a marketing manager, it would hand you an excellent plan every morning and then wait for you to carry it out.
No. Not one.
| People who tapped the phone number on a client site | 0 |
| Enquiry forms filled in | 0 |
| Advertising campaigns ever run | 0 |
| Money ever spent on advertising | $0 |
| Client sites where we can even see a phone tap | 1 of 63 |
What changed this week: until Wednesday we had no way of knowing whether anything worked at all. Now, when someone taps the phone number on the windmill company’s site, we record it and we know where they came from. That is the measuring stick. Nothing has been measured with it yet, because no campaign has been run.
What we still cannot see: whether the phone actually rang, whether anyone picked up, and whether a job came of it. For a trade, that is most of the story — and it is why a tracked phone number is the most important thing on the to-do list.
Almost everything, still — but less than a week ago.
| Pieces of work it produced for this one client | 64 |
| It carried out on its own, start to finish | 11 |
| Still sitting waiting for a human | 45 |
What “waiting for a human” actually means. The software wrote a Google advert — the headlines, the descriptions, all inside Google’s character limits. Somebody still has to log into Google Ads and type it in. It wrote a Facebook post; somebody has to paste it into Facebook. It wrote an article outline; somebody has to write the article.
On Wednesday this number was zero — the software could change a page’s title and its hidden search-engine tags, and nothing else. It is now able to write and publish whole pages, which is where the jump came from.
Worth knowing whether 20% is a ceiling or a starting point. Going through the 45 items one by one:
| 20 of them the software could do, with work we already know how to do | → 55% |
| 12 more unlock the day you buy the AI account | → 77% |
| 13 genuinely need a person | the ceiling |
So the honest answer is about 77%. The last 23% is deciding a price, holding a camera, agreeing what the business stands for, and choosing whether to spend money. A marketing department that claims it needs nobody at all is lying, and it seems better to tell you where the line sits than to pretend it is not there.
Nine complete pages, live on the client’s site, written from things the business had already said about itself. Nothing invented. There is now a page for each of the six services they sell, so someone searching for “line-shaft turbine repair” lands on a page about line-shaft turbines instead of a homepage.
A questions page — royallswindmill.com/faq. Real questions from it:
A service-area page — royallswindmill.com/service-area, listing the towns the company itself already named: Luling, Lockhart, Seguin, Gonzales, San Marcos, Martindale, Maxwell, Prairie Lea and more.
Both pages tell Google exactly what this business does and where, in the format Google reads. That is the sort of thing an agency charges a few hundred dollars for and takes a couple of weeks over.
Google gives you 30 characters for a headline. Go over and it gets chopped off mid-word. These all fit:
Water Wells in TexasTexas Water WellsLocal Water WellsWater Wells Near You
Averaging makes it look like steady progress across the board. It is not. We are close to finished on working out what to do, and almost nothing on doing it and proving it paid. If you only remember one line from this page, that is the one.
Marked in red where it needs money spent, in amber where it is a hard problem rather than a long one, and plain where it is just work.
Three jobs need it: sorting the borderline prospects, writing campaign wording, and reading replies to work out who is interested. Everything else in the system runs without AI on purpose, which is why it is cheap to run.
Right now the system can find the person but not their email address. This is the bit we said we already had on the 13 August call. We do not — and that correction has gone out in writing.
It is on the free tier and the free tier ran out halfway through the home-builder search. That is why we got 19 companies instead of 60.
A separate number that forwards to the client’s real phone, so we can count actual calls instead of guessing from taps. Without this, any “cost per customer” figure we quote is partly guesswork — and we would rather not quote one at all than quote one we cannot stand behind.
The earlier version of this list said 63 sites could have it. That was wrong: only 12 have a phone number or email link at all. The other 50 are news and content sites with no way to contact anyone, so there is nothing on them to measure.
Their Google listing, some posts, some directory entries. All free. The system will not propose spending money until this is done.
This is the thing that turns your second question from “nothing” into a number.
We already run a separate server that does this for our own sites. It needs connecting up.
For a local trade this is the single most valuable free thing there is. Several well-drilling companies we found have dozens of reviews and no website at all — the listing is doing all the work.
Cold email has to go out from separate throwaway domains, never the client’s real one. Get that wrong and their normal email stops arriving.
Somebody writes back — the software reads it, works out whether they are interested, and answers toward booking a call.
Putting the ad up is straightforward. Knowing when to kill a bad one is judgement, and that part will be mediocre until it has run on real accounts for a while.
A page per service, a pricing explainer. Same approach as the two it just published.
31 of the 56 well-drilling companies we found have no working website. We built one for the windmill company by hand in ten minutes. Doing that automatically roughly doubles who we can sell to.
There are 47 items waiting for a human on one client. Going through them one at a time is why Dawn hit a wall.
Which of them the software should learn to do, and which will always need a person.
Right now it learns the words an industry uses. It does not yet know the arguments — what tradesmen disagree about, what customers get wrong. That is the difference between an article that is accurate and one worth reading.
$250 of advertising at $32 a customer is eight customers. You cannot tell which of 25 adverts is best from eight. Testing 25 versions properly needs a much bigger budget or a much longer run — no amount of clever software changes that, and anyone who tells you otherwise is selling something.
Buy the three things in group A. Get a tracked phone number. Finish the free work on the windmill company. Run one real campaign and see what a customer actually costs.
That is about three weeks, and at the end of it question two stops being “no” and starts being a number. Everything after that is making it a product you can sell to more than one person at a time.