How I Would Automate Your Sales Process, and the One Part I Would Never Touch
Only automate steps that already work by hand, and expect quality to dip at first. Use AI for speed and volume. Never automate the sales call. Write down RACE and list what you do at each stage. Then build each piece as an automation, a Claude routine or a Claude skill.
- Automate only a step that already works when you do it by hand
- Use AI when volume matters, like 90,000 old leads to reactivate
- Never automate the sales call or anything that touches a person directly
- For most agencies, start with speed to lead and the follow-ups
Everyone wants to point AI at their sales process right now. Good instinct. Most people will do it wrong.
This post is the written version of my video. Watch it here: https://youtu.be/X8eVJ6by-TM
I have helped more than 55 businesses use AI in their sales. So I know a thing or two about what works and what does not. And the part nobody makes videos about is what you should not automate. That is where people lose money.
So here is how I would automate your sales process. The rules first, then what to build, then how.
Rule one: it has to work by hand first
Automation is taking a manual step that already works and making it faster. That is all it is.
If you do not have a manual step, you cannot expect the automation to work. I have seen this so many times. Someone automates their marketing. I ask how much money it made. Zero. They did not make money from manual marketing either, so why would the automated version be different?
Before you automate something, you need to know it already works by hand. Then the only thing you changed is the AI. If it breaks, you know where to look.
Rule two: expect the dip
This one is not obvious. When you automate a process that works, expect the quality to drop at the start.
Why? Because manual things have better quality than automated things. That is just logic. Your best person doing it by hand beats the AI on day one.
It gets better. You train it, you test it, you change it, you test again. But on day one it is worse, and you need to plan for that.
So there are things I would rather keep manual, because I know the quality drop would hurt. Anything at the front of your sales process. Anything close to the sale.
Let's say I am doing Loom videos for Upwork applications. I do those by hand. I can automate everything around the video. The video itself stays me. A five percent drop in quality there can mean I lose clients, and my business stops growing.
Speed and volume
So why automate at all? I use AI for two things. Speed and volume.
I have a client with 90,000 leads sitting in their database. They cannot reactivate all of those by hand. Maybe they could do 30 a day, very custom, and the quality would be high. But 30 a day against 90,000 people is nothing.
If AI does it, maybe the quality is a bit lower. But they get through the list, because AI can do 100 times more than a human can.
That is the trade. Automate when volume beats polish. If you are small enough to do it by hand, do it by hand.
Rule three: never the sales call
And the hard line. Never the sales call. Never anything that touches a human directly.
I would rather do the call myself. A five percent mistake on a call is the lead. Run that a hundred times and it is real clients gone.
Sort it with RACE
Now you know why, and why not. What do you actually automate?
There is a framework, maybe you already know it. RACE. Reach, Acquire, Convert, Expand.
| Stage | What it means | What I would automate |
|---|---|---|
| Reach | Getting seen | Making the creatives, or a system that helps you make them. Watching the ad KPIs daily so you get a notification when something is off. Killing the losing ads on their own. |
| Acquire | Attention to booked calls | Speed to lead, the big one. A better form. A better funnel. Follow-ups that get them to book. |
| Convert | Booked call to closed deal | Reminders so they show up. Nurture before the call, a testimonial, a VSL. The proposal itself. Follow-ups after the call. |
| Expand | The clients you already have | More attention, more revenue from the people already paying you. |
Here is how I would use it. Write RACE down. Under each letter, list everything you are doing right now for that stage, and the things you are not doing yet. Bullet points. That list is your automation plan.
My Upwork example: the split
One of the ways I get clients is Upwork. An application has a lot of steps. Read the job. Prepare. Build a proposal. Fill out the form. Shoot a Loom video. Send.
I automate everything around it. And I keep three things for myself. Judging the job. Recording the Loom. Hitting send.
That split does not hurt the quality, because the parts I gave to the system are the parts I do the same way every time. The parts that touch the person stayed me.
That is the whole test. Does it touch the person? Would they know you did it with AI? If yes, be careful.
Three shapes to build it
You did the RACE sort. You know what to automate. Now, how do you build it? There are three shapes.
| Shape | How it runs | Use it for |
|---|---|---|
| An automation | Completely on its own, no you involved | Deterministic stuff. Speed to lead. Follow-ups. If your CRM has workflows, like GoHighLevel or Close, use those. |
| A Claude routine | On a schedule, a bit flexible | Something that checks your KPIs and gives you a daily briefing on whether the ads are on target. |
| A Claude skill | You in the loop, back and forth | Content. Ad creatives. Anything where you need the judgment in the middle. |
The skill is the one people miss. If you are doing something by hand today and it needs your judgment, a skill is the first step. You keep the control. You lose the typing.
FAQ
Should I automate my sales calls with AI? No. Never anything that touches a human directly. A five percent mistake on a call is the lead. Automate what is around the call: the reminders, the research, the follow-ups, the proposal.
My automation is worse than doing it by hand. Did I do it wrong? Probably not. Expect the dip. On day one the automated version is worse than your best person. Train it, test it, change it, test again. It gets better. If it is still worse after that, ask if the manual version ever worked in the first place.
What should I automate first? Write down RACE and list what you do at each stage. Start with what already works by hand and where volume matters most. For most agencies that is speed to lead and the follow-ups.
Automation, routine or skill? If it should run while you sleep and the steps are always the same, an automation. If it needs a bit of flexibility on a schedule, a Claude routine. If it needs you in the middle, a Claude skill.
What about my old lead list? That is the volume case. If you have thousands of old leads and you can only follow up with 30 a day by hand, AI is how you get through the list. The quality per message is a bit lower. The money is in the list anyway.
Hamza Oulad runs Qemoza, an AI ops partner for paid-media agencies. He and Finn build the sales and fulfillment systems in this post for agency owners who are stuck doing the work instead of selling.
If there is one automation you want and you are stuck on how, tell me which one. I will show you how I would build it. That is also what I do for agencies at qemoza.ai.
