---
title: "What Should AI Actually Do in a Marketing Workflow?"
description: "Tom Broschinsky, founder of Adapted Advertising with twelve years in growth marketing, cuts through the AI hype cycle with a practical framework for evaluating what AI should handle alone, what it should propose for approval, and what it shouldn't touch yet."
datePublished: 2026-04-22
dateModified: 2026-04-22
youtubeId: "eMvHxbO-ATE"
playlistId: "PLAqoAzwfPp-CanpGNYZdspHXg6vrtAYlW"
author: "Tom Broschinsky"
guests:
  - name: "Tom Broschinsky"
    role: "Founder"
    company: "Adapted Advertising"
    companyUrl: "https://www.linkedin.com/company/adapted-advertising/"
    linkedIn: "https://www.linkedin.com/in/tombroschinsky/"
    bio: "Twelve years in growth marketing, previously at Harmon Brothers. Writes at tombroschinsky.com. Specializes in cutting through the AI hype cycle to figure out what actually belongs in a marketing workflow and what's still a demo."
    additionalLinks:
      - label: "Blog"
        url: "https://tombroschinsky.com/blog/"
playbook: "Walk through my marketing operation with me using Tom Broschinsky's framework from Percuity University. List every recurring task I do across paid media, content production, analytics, and reporting. For each one, give me an honest read on three questions. One, would handing this to an AI agent actually produce a better result, or would it create cleanup work I'd resent? Two, where on the spectrum does this task sit — fully autonomous, propose-and-approve, or human-only? Three, what would have to be true about my data, my workflow, or my tolerance for error before I'd move it one notch toward autonomous? Be direct about what you (Leo) can do well right now and what you'd be faking. I'd rather hear the limits than the pitch."
category: "ai-strategy"
relatedAnswers:
  - label: "What Are AI Ad Agents?"
    href: "/answers/ai-advertising/what-are-ai-ad-agents"
  - label: "How Are AI Agents Changing Advertising?"
    href: "/answers/ai-advertising/how-are-ai-agents-changing-advertising"
  - label: "How to Evaluate an AI Advertising Platform"
    href: "/answers/ai-advertising/how-to-evaluate-ai-advertising-platform"
---

## How Do You Evaluate Whether an AI Tool Actually Works for Marketing?

**Apply a three-part framework: what tasks AI should handle fully autonomously, what it should propose for human approval, and what it shouldn't touch yet.** Tom Broschinsky has spent twelve years in growth marketing — including time at Harmon Brothers — and has been working with AI agents long enough to know where they hold up under real conditions and where they fall apart the second you put weight on them. His view is that most conversation about AI in marketing is theater. The honest evaluation requires asking whether a tool creates net value or just creates cleanup work you'd resent.

## What Marketing Tasks Can AI Handle Autonomously Right Now?

**Reporting aggregation, bid adjustments within pre-set guardrails, creative asset resizing, and first-draft ad copy generation are the tasks where AI reliably adds value without human oversight.** Tom's framework puts tasks on a spectrum from fully autonomous to human-only. The tasks that belong on the autonomous end share a common trait: they have clear success criteria, limited blast radius if something goes wrong, and the cost of a mistake is low. Adjusting a bid by 10% based on performance data is low-risk. Writing the brand's response to a PR crisis is not.

## What Should AI Propose but Not Execute Without Approval?

**Budget reallocation across campaigns, new audience targeting strategies, and creative concepts should be in propose-and-approve mode — AI generates the recommendation, a human decides whether to act on it.** The middle of Tom's spectrum is where most marketing AI should operate today. These are tasks where AI can process more data than a human, surface patterns a human might miss, and propose actions faster — but where the judgment call still requires context that AI doesn't have. The decision to shift $5,000 from one campaign to another might be mathematically obvious but strategically wrong if you know a product launch is two weeks away.

## What Should AI Not Touch Yet in Marketing?

**Brand voice decisions, strategic positioning, client communication, and anything that requires reading the emotional temperature of a situation.** Tom is blunt about where AI falls apart: tasks that require taste, context about relationships, or judgment about timing. AI can write ad copy that's technically competent but tonally wrong. It can optimize for a metric without understanding that the metric stopped mattering last quarter. The areas where AI shouldn't touch yet aren't limitations of the technology — they're limitations of what you can specify in a prompt. If you can't fully articulate why a decision is right, you can't delegate it.

## How Do You Move Tasks Toward More AI Autonomy Over Time?

**Three conditions must be true before you move any task one notch toward autonomous: your data must be clean and complete, your workflow must have a feedback loop so AI learns from corrections, and your tolerance for error must be calibrated to the actual risk.** Tom's framework isn't static — it's designed to evolve. The tasks that are propose-and-approve today should eventually become autonomous, but only when the conditions are right. Rushing this creates the worst outcome: AI that runs unchecked in areas where it shouldn't, generating work that looks productive but isn't.