A demo can produce an impressive account summary. A sales development rep preparing for a real call needs more than that. The reason to reach out must be current, the product claim defensible and the proof relevant to that account. And the result has to land in the CRM in a form the rep can use.

Start with an ordinary employee’s access.

Run the task with the access the team has today. Ask your company’s AI lead and the head of sales development to watch the same run. At each stop, write down what was missing. A model that cannot do the reasoning, a source the rep cannot open and a CRM that is not connected are three different pieces of work.

A briefing or campaign task · Q12

Where approved AI hands the job back.

01

Writes the answer

People check it and move it into work systems.

32%
02

Finds approved information

People still perform the action in another system.

15%
03

Prepares work for approval

People approve work the assistant has prepared inside the work systems.

21%
04

Completes permitted steps

People handle exceptions and higher-risk decisions.

14%
We do not use approved AI for these tasks: 14%Don’t know: 5%
581 people were asked; blank answers are left out. Each answer is one stopping point, so the four add up to a total rather than building on each other. Inspect the full question ↗

It is also a practical way to settle marketing’s role in the company’s AI plan. The CMO names the priority task, the functional head says what a usable result looks like, and your company’s AI lead commits to delivering it. Put those three names, the date and the test for done on one page and bring that page to the planning meeting rather than a slide. A platform someone else pays for becomes useful to marketing when those three sign the same page.

Capgemini, CMO playbook 2025

Match the permission to the action.

Take the three categories to legal in a particular order. Start with internal preparation, where nothing leaves the building and the approval is easy to give. Use that agreement as the template for external publication, where product marketing and a named approver sign off. Leave spending decisions for last, with the budget owner’s limit written in. Legal says yes to narrow permissions faster than to broad ones, and the order matters more than the wording.

Q13

The most recent delay in AI-supported marketing work

Waiting for an approval that did not distinguish the task’s risk16%
94 responses
Missing or conflicting company information41%
234 responses
No permission to use the required model or connected tool9%
49 responses
Correcting an inaccurate or unsuitable result12%
67 responses
Coordinating people across tools and handoffs9%
52 responses
There was no material delay5%
26 responses
We have not attempted such a task7%
38 responses
Don’t know2%
14 responses
574 people answered. All listed answers, including don’t know, stay in the total. Full readout and cross-tabs ↗

A suggested approval design

Internal account brief

AI preparesRetrieve approved sources and prepare a cited brief.

People decideAccount owner checks missing or conflicting facts. No buyer contact.

Public competitive claim

AI preparesDraft a comparison using current product evidence.

People decideProduct marketing and a named approver sign off before it goes out.

Campaign spend change

AI preparesAnalyze results and recommend a budget adjustment.

People decideThe budget owner approves the amount before anything changes.

Software teams call the setup around an assistant a harness. In marketing it is the approved sources, the written instructions, the connected tools, a record of what the assistant did, the quality checks and the person who handles exceptions. It is also the part you keep when the model or the vendor changes, which is why chapter 2 asks you to keep it outside any one tool. The engineering write-up below explains why the pieces matter for AI assistants that write software; the judgment about accuracy, brand and permission is marketing’s own.

Anthropic, engineering notes on long-running agents

Agree what “done” means before anyone builds.

For the account brief, require a usable record in the CRM, a source for every claim that matters and a clear flag where information is missing. Let the rep and the account owner judge whether it helps the next call. Count how many times a person had to step in or correct something, then compare with how the brief gets made today.

Set the date now. Ask for the finished brief within thirty days, made with a normal employee login, and put that brief in front of the sales team instead of a demo. What the rep does with it on the next call is the only acceptance test that counts.

xAI’s Grok Bot, an assistant with its own computer and toolsMuse, an assistant that keeps context and asks before it acts

Bring this discussion to your GTM team.

The team benchmark and workshop connects the operating questions with your own workflows, priorities and people.

Request the team benchmark and workshop ↗