Your Data Architect, Not Your Analyst: AMP's Alex Robinson on Putting AI to Work on Partner Metrics

Getting the full picture of an affiliate program on a Monday morning takes time. The numbers are all there, spread across reports and endpoints, and somebody has to pull them together before anyone can act.
AI tools look like the obvious fix: export everything, hand it to an LLM, and ask what's going on. Alex Robinson thinks that's exactly the wrong move. LLMs can hallucinate or make assumptions, and he calls that "hugely dangerous" when the numbers are going to a client or to leadership. Worse, there's no way to check the answer. If the math is wrong, you only find out after the output is already in front of someone.
That warning carries more weight from someone who has already built the alternative. Alex's team runs affiliate program management for clients in financial services, where a wrong number can become a compliance problem.
Alexander Robinson is co-founder of Affiliate Marketing Partners (AMP), a Pennsylvania-based agency that runs affiliate programs on both the advertiser and publisher side, with deep roots in financial services, a highly regulated space. Today he sits on the Performance Marketing Association's AI Council and co-wrote the PMA's AI automation guide for affiliate managers with Everflow's Michael Cole. His approach flips the usual question. Instead of asking AI for answers, AMP uses it to build the systems that make the answers obvious, and every number stays one a person can check.
Below, Alex walks you through that process in six steps, from what the standard setup misses to the guardrails to set before anyone starts building. By the end, you'll know how to build a weekly view of which partners are leading and slipping, narrow compliance checks to the few placements a human needs to see, and spot fraud in data you already have, while winning back hours for the decisions that grow a program. It starts with why the default dashboards aren't enough.
Step 1: See what the standard setup misses
Every program cares about something different: different events, different funnels, different definitions of a good week. Platform dashboards are built for every kind of company, not around one program's priorities.
When nobody does that customization, the metrics a program cares about most stay buried, and the work falls back on people. On AMP's publisher-side work the problem compounds: a publisher working with 20 partners has to pull from 20 different sources. The old fix was downloading everything into a spreadsheet and aggregating it by hand, the kind of work that can eat 10 to 20 hours of a team's week. The newer shortcut, pasting that export into an LLM, trades the hours for answers nobody can audit.
The fix starts earlier, with the job you give AI in the first place.
Step 2: Make AI the architect, not the analyst
This is the core of Alex's method. He never drops raw data into an LLM and asks it to analyze. He uses it to build the pipeline instead.
As an architect, AI helps design how data is ingested and transformed, so the formula behind every metric is known and checkable. The human still reads the result, makes the call, and reviews everything before it goes out. In Alex's words, "nothing goes out unreviewed."
The end result is a single, customized view of the program, by day, brand, partner, and channel, built on formulas the team understands. When leadership asks where a number came from, there's an answer. With that foundation in place, the next question is what to put in front of the team each week.
Step 3: Build the view that answers the question
Asked which prompts reveal revenue leakage or offer fatigue early, Alex's answer was blunt.
What works is a view built for the question. Every AMP client dashboard has a prior-week tab Alex calls the "leaders and bleeders." The data is already ingested and transformed so the strongest and weakest performers are easy to see, then color-coded red and green on the metrics that matter most. Nobody has to ask which offers are running out of gas, so the conversation goes straight to action: which partner needs attention, and which one is crushing it.
Performance is only one place this pays off. The same approach matters even more where a missed issue costs the most: compliance and fraud.
Step 4: Let AI narrow the field on compliance and fraud
AMP works in financial services, where a placement problem isn't a price that's off by a couple of dollars. In Alex's words, it "could be an FTC violation." Every placement needs the right disclosures, and certain terms can't appear at all.
AMP built a tool around that. The team defines the rule set (required disclaimer language, banned terms, required formats) and gives it the list of placements to monitor: 150 URLs. The tool checks each one against the rules and reports back on every record.
A human reviews every flag, because sometimes the tool is wrong. But the work shifts from a reactive manual sweep to catching issues before a client or leadership spots them. Other tools can capture screenshots and snapshots at each check, a timeline Alex calls "gold" for a financial company that has to go through an audit.
Fraud follows the same pattern. Alex's advice is to start by asking an LLM which signals of bot or fraudulent traffic to look for. The answers are familiar: click-to-install times under three seconds, geographic clusters, a thousand orders at 3 a.m. from one IP address. And "all that data lives in Everflow already." The job is a formula, such as time from click to install by partner, that becomes a fraud-risk view on a dashboard.
Both of those tools began as a tedious manual check. That's exactly where Alex tells teams to start their own builds.
Step 5: Automate the task you hate most
When teams ask what to automate first, Alex asks what tedious, repeatable task they dread every week or month. That's the starting point.
His usual starting point is Google Apps Script, which sits inside every Google Sheet. He and Everflow's Michael Cole walk through it in their PMA guide, AI Foundations: Going From Formulas To App Scripts for Affiliate Managers. Describe the task the way a product manager would (the columns, how the data looks, what you do to it each week) and ask an AI tool to write the script. Paste it in, run it, and put it on a schedule. No new software, no extensions. In Alex's view, the barrier to writing code has never been lower.
That low barrier cuts both ways. Once anyone on the team can build a script or hand a tool their data, the same ease that saves hours can put client information into the wrong place or send an unreviewed email to a client. Alex is direct that it "can be very risky," which is why AMP set its rules before anyone started building.
Step 6: Set guardrails before rollout
When AMP rolled out its AI tools, the team set the governance framework first. No PII or identifiable information goes into the tools. Alex's examples of the defaults a team can set: every email reviewed by a human before it goes out, no automated emails to clients or leadership, and no access to production data, secrets, or passwords. Then came a training session on what the tools can't do, partly to "kind of scare them a little bit." Using AI to build systems, rather than feeding it raw data, keeps that risk small by design.
With those six steps in place, the change shows up in how a team spends its week.
The payoff: better decisions
Alex doesn't want his team spending 10 to 20 hours a week in spreadsheets, because those are hours spent not acting, solving problems, doing outreach, or making decisions. But he's careful about how he measures the change.
The better measure is quality: the decisions made because nobody was buried in a spreadsheet, and the relationships built with that focus. For an industry built on relationships, that's where the value shows up.
It also changes how partners get evaluated. Alex pairs performance data with AI visibility data, meaning which publishers LLMs cite when consumers ask for recommendations. A partner can rank tenth on performance and first in AI citations, which turns a volume conversation with leadership into a quality one.
From raw data to clear decisions
Go back to that Monday morning. The risk was handing AI the analyst's job and trusting answers no one could check. Alex's process puts AI one step back from the data instead, designing the ingestion, the formulas, and the views, so the insight is visible and auditable before anyone has to interpret it. The human stays where the judgment is: reviewing flags, deciding on outreach, building the partner relationships the channel runs on.
It's also why he starts with a plan before any action. Teams that jump straight to "here's my data, here's my task, do this task" skip the step that makes everything after it trustworthy.
For performance leaders, that's the prize: a program where the right action is obvious on Monday morning and compliance issues surface early.
