Updated August 2026. Originally published July 2024.
I asked one of our SDRs to write down every step of his daily workflow.
The first version was clean. About ten steps. It was the version he thought I wanted to see, the workflow as it would appear in a job description: tidy and defensible.
I pushed him to write it again and include everything. Every step, including the ones he considered too small to mention.
The second version had more than a dozen steps, and the gap between the two lists was where all the value was.
The steps nobody writes down
Here is what showed up in version two that was missing from version one:
- Opening a browser and pulling up LinkedIn before every call
- Copying context between tabs
- Formatting internal notification emails from scratch, every time
- Retyping notes into the CRM
- Drafting follow-up and confirmation emails by hand
None of those are the job. Not one of them is what you hire an SDR to do. They are the connective tissue between the parts that matter, and because they are small, nobody counts them.
We counted them. They added up to 35 minutes per meeting.
Our best month at that point was 57 meetings. That is roughly 33 hours a month, on one person, spent on work that generates nothing.
Why the first list is always wrong
This is the part executives get wrong, and I got it wrong too until I watched it happen.
When you ask someone to document their workflow, they document what they think matters. They edit as they write. They leave out the small things because the small things feel beneath mentioning, and because listing them feels like admitting you spend your day on trivia.
I now use an analogy that gets the point across faster than any instruction I have written.
I ask: what is the first thing you do in the morning?
They say something like “log into the CRM to review warm opportunities.”
And I say: no. The first thing you do in the morning is get up from the bed.
That is the whole lesson. I do not want the professional summary. I want every step, in order, including the ones that seem too obvious or too small to write down. The automation opportunities are never in the steps people are proud of. They are in the ones people skip past.
The sequence matters as much as the content.
Document everything. Then automate. Then optimize, with them rather than at them.
Most teams try to optimize first, because optimizing feels strategic and documenting feels like admin. It does not work. You cannot improve a process you have only seen the summary of.
What happened next
We automated five of those tasks.
He stopped spending 35 minutes per meeting on work that produced nothing, and started spending it on work that did. He became a team lead, and we applied the same process to his team.
The part I did not expect: the other SDRs asked for it. Nobody was told to adopt anything. They watched what happened to his numbers and came asking. That is the only adoption strategy I have ever seen work, and you cannot manufacture it with a training session.
Then the honest version of the results, because I have seen too many of these stories told loosely.
The following month we closed 123 appointments, up from 57. That was not automation alone. In the same period I added two headcount to the team, automated onboarding and training so those two ramped in a fraction of the usual time, and put proper tooling licenses in everyone’s hands.
Automation was the thing that made the other three worth doing. It was not the whole story, and if I told you it was, any SDR reading this would know I was inflating.
The number I would actually defend is the 35 minutes. It is small, it is checkable, and anyone who books meetings for a living will recognise their own day in it.
Two years later: the interface moved
Everything above is what I wrote in 2024. The method still holds. What changed is the ceiling.
In 2024, the answer to “how does a marketing team use AI” was to paste your text into a chat window and prompt it well. That is what the original version of this article taught, in detail, with prompts you could copy. I took the same approach to content at the time, which I wrote up in using AI for website content.
That advice is now obsolete, and not because the prompts got worse.
Today nobody on the team opens a chat window to start their day. The work arrives already done.
- A morning intent digest posts to the SDR group. Accounts that replied to outreach or visited our site recently, ranked by intent, with contact details and visit history attached. Nobody requests it.
- A hot reply alert fires when a cold email gets a positive response. It includes a summary of what the prospect actually said, a drafted reply ready to send, the right sales asset attached, and the reasoning for why it was classified as hot. The job is to approve or adjust, not to compose.
- A pre-call briefing goes out automatically to the sales manager for that region the moment an appointment is booked: contact, company, headcount, leadership, background. No more drafting announcement emails by hand. That was one of the five tasks we automated.
- Call transcripts flow from our phone system into the GTM engine, which runs sentiment and call analysis and feeds it back as real-time coaching.
If you are earlier in this than we were, the first AI project most companies skip is the one I would start with, and it is closer to where this article now ends than where it begins.
Look at that list against the five in-between tasks from the original workflow. They are the same tasks. The difference is that in 2024 we automated them one at a time with prompts, and now they are a system that runs whether anyone is thinking about it or not.
Where the 2024 version stopped short
I thought the ceiling was faster tasks. It was not.
In my opinion, what we were using in 2024 was not really AI. It was assistive. You prompted it, and whether it knew the answer or not, it would convince you that it did. It gathered information from known places and returned it in the shape of a conversation. No real thinking. No access to our historical data. No skills. No ability to act on its own.
What we run now is a different category of thing. There is a harness folder holding our metrics and sales enablement documents going back to 2021, along with skills built specifically for our business: an MSP solutions manager and an MSP sales enablement skill. It is self-hosted. It knows what we have tried, what worked, and what did not.
That last part is the whole difference. This is also why I stopped treating AI adoption as a tooling decision and started treating it as an operating one, the same way I argue for budget in Engineer Your GTM Like a CFO.
A general model can tell you what an RFP response usually looks like. A system with five years of your own outcomes in it can tell you what has actually closed business for you.
Running as a solution manager, it has produced sales enablement documents, a site seller, an RFP generator, and more. Not drafts I then rewrote. Working output.
None of that was possible in July 2024, and no amount of prompt engineering would have got us there.
If you are starting this today
The method has not changed, and I would still start exactly where I started two years ago.
Sit down with one person. Ask them to write every step of their day, and when they hand you the tidy version, send them back to do it properly. Count the minutes in the gaps, and price them the way you would price anything else you are paying for. Automate the five worst offenders. Let the results recruit the rest of the team for you.
What I would tell you that I did not know in 2024: do not stop at faster tasks. The tasks were never the point. Getting your own history, your own metrics, and your own way of working into a system that can act on them is where the difference is, and the documentation exercise you start with is what makes it possible later.
Document everything. Then automate. Then optimize.
You still have to get up from the bed first.
This article was substantially rewritten in August 2026. The original July 2024 version documented a five-step process for moving a marketing team onto AI using chat-based prompting. That method still works as a starting point and is preserved above. The sections covering what the system became, and what I got wrong about its ceiling, are new.
