“There is nothing so useless as doing efficiently that which should not be done at all.” Peter Drucker
Last week I handed you two lists and forty five minutes. Steal the roster feeding your competitors, wake the dead partners in your own dashboard, twenty five real notes out the door before Friday.
The reply I got most was some version of the same question.
Can I just have AI do all of that?
Mostly, yes. And I need to say this clearly before anybody gets the wrong idea about where this issue is going, because I am not writing the one you think I am writing. I am not nervous about AI in this channel. I think it makes recruiting and management materially better, not incrementally better, and the tedious seventy percent of this job is finally going away. I will not be attending its funeral.
But there is a version of yes that costs you a year, and I bought it.
Years ago at SmarterChaos I turned on a set of automated rules for a client, because I was tired of running the same optimization by hand every Monday like a man churning butter. The rules worked. Volume up, cost per acquisition down, dashboard greener every week. I was extremely pleased with myself for about five months, right until a quarterly review where somebody asked a mild question about partner mix and I pulled it up live, in front of everyone, and discovered that the program had quietly reweighted almost entirely to the bottom of the funnel. Every content partner who ever introduced us to a buyer had been slowly starved out by my clever little rules.
The rules did exactly what I told them to do. That is the trouble with rules. They listen.
Nobody approved that. Not the client, not me, not anybody in that room. It just happened, on schedule, every week, while the numbers looked great.
I wrote an issue a while back arguing that a tool multiplies whatever judgment you feed it. Still true. This is the other half of it, and it is the half nobody audits. A tool also multiplies whatever you let it decide.
What You’ll Get in This Issue
The one question that sorts what to automate from what to keep, and why “right answer or trade off” beats every framework you got sold this year. The two lists to write down before you buy another tool. Three ways I am watching this go sideways in real programs right now, including the auto approval flood that looks like growth on the report and the generated outreach quietly getting you blacklisted by the exact publishers you wanted most. Six things a well run program actually looks like from the inside, all of them visible this week without a single new tool. And the off switch test, which is the entire audit compressed into one uncomfortable question you can answer this afternoon.
Part One: Right Answer or Trade Off
Here is the part everybody skips. When you switch on an automation you are not buying speed. You are handing over a decision. Speed is just the part you notice.
We all pick tools by asking what the tool can do. Almost nobody writes down what the tool is allowed to decide, which means that gets settled by whoever configured it, on a Thursday, probably without telling anyone, and then it runs unsupervised for two quarters.
So here is the line I hold, and it has survived every wave of tooling since 1999.
Automate anything with a right answer. Keep a human on anything with a trade off.
A broken link has a right answer. It is broken or it is not. A machine will find it faster and more reliably than your team will at six on a Friday, and it will not sigh about it. Whether a partner deserves a rate bump is a trade off. That is money you do not get back, a relationship you cannot repurchase, and a bet on what this partner is worth a year from now, which is not a fact sitting in your account waiting to be queried.
The danger was never that machines are bad at trade offs. It is that they answer both kinds of question in the same confident voice, and a spreadsheet does not blush. Nothing pings you to say a judgment call just got made by a rule. You get a number, and the number looks fine.
Part Two: The Playbook
Three moves. The first one is thirty minutes with a pen and it is the one that saves you the year.
Move 1: Draw the Line.
Two lists. Actually write them, because the entire failure mode here is that everybody thinks this is understood and nobody has ever said it out loud.
The machine owns anything with a right answer. Compliance monitoring for trademark bidding, coupon injection, adware, and toolbar behavior, running all the time instead of once a quarter when somebody remembers to look. Link audits across every partner in the program, because dead links are the quietest money leak in this business. Payment validation. Report assembly. Alerts when EPC falls off a cliff, when average order value lurches, when a tracking ID nobody recognizes starts doing real volume. First pass scoring on applications so nobody on your team ever opens a blank record and starts Googling. Every one of those has a correct answer sitting right there in the data, and putting a human on it is a waste of a human.
You own anything with a trade off. Who gets real investment. What the rate actually is. Placement, exclusivity, co marketing. Your top twenty relationships. Whether a partner is genuinely incremental or just standing near the finish line looking helpful at the moment of conversion. When to end a relationship, and how you tell them. Every one of those costs money you do not get back or trust you cannot rebuy, and a rule will make the call badly while producing a report that says otherwise.
Then there is the middle, which is where the good tools actually live now. The machine forms an opinion, a human confirms it. A scored application with a recommendation attached. A flagged partner with the case already assembled. A suggested rate with the math shown. That is the sweet spot, and what makes it work is not the model. It is the confirm step. Which is, reliably, the first thing anybody turns off, usually around week three, usually by the person with the most confidence and the least context.
→ Every automation hands over a decision, so write down which ones you handed over. Machines take anything with a right answer, you keep anything with a trade off, and the middle only works while somebody still has to click confirm.
Move 2: Watch the Three Places This Goes Sideways.
Same three failures, program after program, right now. None of them exotic. All of them look like progress on the way down.
The auto approval flood. Somebody needs to hit a recruitment number, so applications get auto approved against a filter with the structural integrity of a screen door. Two quarters later you have hundreds of new partners, most of them deal sites, browser extensions, and at least one enterprising soul bidding on your own brand name and invoicing you for the privilege. Revenue is up on the report. Almost none of it is incremental. You are paying commission on carts that were already closing, and last click will defend it to the death, because last click always does.
Do not turn auto approval off. Bound it. Auto approve inside a set of partner types and traffic sources you have already decided you want, and route everything outside that set to a human with a same day deadline. Speed is a recruiting weapon and I am not asking you to hand it back.
Generated outreach at volume. The pitch is two thousand personalized notes a week. The reality is that a decent publisher now gets forty of those a week and clocks the pattern in the first line, and once they have clocked it you are filtered forever. You did not save time. You spent your welcome.
Point the machine at building the list, finding the exact page, and pulling the evidence. Then have a person write four sentences. Twenty five researched notes beat two thousand generated ones, which was true before any of this and is more true now that everybody else went the other way.
Rules nobody owns. This is my SmarterChaos story and it is the expensive one, because it takes months to surface and it never announces itself. Auto optimization walks your budget toward whatever converts closest to the click. Left alone it starves the top of the funnel, irons your rate card down to one number, and rebuilds your entire partner mix without a single person choosing that. Nobody chose it. The rules chose it, and the rules do not attend the QBR.
The fix is unglamorous and it works. Every rule gets a name and a review date. If nobody owns it, that is not automation. That is drift with a login.
→ The failures do not look like failures. Auto approval buys volume you already had, generated outreach spends welcome you cannot get back, and unowned rules quietly redesign your partner mix, so bound the first, keep a human on the second, and put a name and a date on the third.
Move 3: Know What Good Looks Like.
Fine, but how do you know if you got there. Six things. All of them visible this week, none of them requiring a purchase.
Every partner has a tier and a next action. No dormant middle sitting there collecting a cookie.
Application to decision runs in hours, because qualification is automatic and the judgment call is a scheduled fifteen minutes instead of a someday.
Your top twenty partners get a real conversation every quarter. All twenty. This is the one that separates a program from a dashboard, and it is always first to slip when things get busy, which is exactly when it matters.
You know your incremental rate and not just your last click rate, because you ran a holdout and you can say out loud what this program is worth if it vanished tomorrow. Most people cannot. It is a great question to ask on a first date with a new agency.
Every rule has an owner who can tell you in one sentence what it is allowed to decide.
And the calendar moved. Less time assembling numbers, more time negotiating. If your team automated a quarter of their week and spent the free hours making more reports, you did not buy leverage. You bought a nicer version of the same week.
→ Good is visible. Tiers and next actions on everybody, decisions in hours, twenty real conversations a quarter, a known incremental rate, an owner on every rule, and a calendar that moved from assembling numbers to negotiating.
Part Three: The Off Switch Test
Whole audit, one question, one afternoon.
Turn off every automation in your program tomorrow. Do your decisions get worse, or do they just get slower?
If they only get slower, you built it right. The machines are buying your team time and your team is spending it on judgment. When this is working it does not feel like magic. It feels like your best people finally having room to be your best people, which is less exciting and considerably more profitable.
If your decisions actually get worse, the tools are not saving you time. They are making calls nobody signed off on, and it looks fine right up until the quarter it does not, and by then your partner mix has been quietly rebuilt underneath you and the repair is a year long. I have done that repair. It is not fun and everybody in the room already knows whose rules did it.
You do not have to switch anything off to run this. Do it on paper. List every automation, write next to each one whether it surfaces something or decides something, then look at the decides column and ask who approved that.
Most of the time the answer is nobody. It was a default. Set by somebody who has since left the company. In a tool nobody has opened since the demo.
Your Weekly Chaos Challenge
Twenty minutes and one sheet of paper.
Column one: every rule, filter, alert, and integration running in your program right now. Include the ones your network runs on your behalf, because those count and everybody forgets them.
Column two: does it surface something, or decide something?
Column three: who owns it, and when did they last look at it?
Anything in the decides column with an empty box in column three is your week. Put a human in front of it or put a name and a review date on it. Either is fine. Walking past it is not.
→ List every automation, mark it surfaces or decides, put a name on every decides, and put a confirm step in front of anything that spends money or ends a relationship.
And in the comments, tell me the dumbest thing an automated rule ever did to one of your programs. I went first. Mine is up top and I have not fully recovered.
Final Thought
The industry is about to spend two years arguing over whether AI replaces partner managers. It is the wrong argument, it will take four conference seasons, and there will be a robot on the slide.
AI replaces the work that never needed a person. That work is enormous. It is most of what burned out every affiliate manager you have ever hired, and losing it is a gift, not a threat. What it cannot do is decide what your program is for, which partners are worth real money, or who gets a phone call in a bad quarter instead of an automated pause. That last one is not a workflow. That is the job.
What it will do, quickly and without asking permission, is show everyone whether there was a strategy under the busywork. Programs run on judgment are about to get much better. Programs run on activity are about to find out.
If you would rather not draw that line yourself, this is what our agency side, Chief of Chaos, does for the programs we run. We come in, find every rule nobody owns, and decide out loud what the machine is allowed to do.
And if you already have a sharp team, good, keep them. This is also why we built Alfie the way we did. A hundred years of affiliate judgment baked into the machine, so it works the tedium at full speed and hands you the calls that need a person instead of quietly making them on your behalf. The goal was never a tool that decides for you. It was a tool that brings you the decision with the homework already done.
Automate the work. Elevate the strategy. Just be able to say which is which, because most programs cannot right now, and the ones that can are about to walk away from the rest.
Until next week.
Run toward the chaos.
Matt Frary, Chief of Chaos
President & COO, XPFlow
#AffiliateMarketing #PerformanceMarketing #PartnerMarketing #AI #ChaosToGrow
