5 Steps to Use AI in Sales Without Losing the Human Touch

A simple framework for deciding what AI should handle and what should stay human-led.

I have been getting a lot of sales emails and DMs lately that technically mention the right things.

They reference my newsletter, even the number of my subscribers. They mention AI. Sometimes they even name a specific post.

But most of them still do not feel like a human actually read anything.

And maybe that bothers me more because I use AI so much. I know the shape of an AI-assisted message. I can feel when someone asked a tool to “personalize” something without actually doing the thinking underneath it.

That is the part that feels annoying. The message is trying to sound researched, but it still feels performative. Like someone wanted the credit for doing their homework without actually caring about the person on the other side.

So when David sent me this piece, the frame clicked for me.

Most people are asking, “How do I use AI to do more sales work?”

David is asking a better question: “How do I use AI to remove the admin drag so I can show up more human in the moments that matter?”

That feels much closer to the AI Maker way of thinking about this stuff. AI should help with the notes, the prep, the first draft, and the messy follow-up work. But the judgment still has to come from you.

David Roy writes Eng Sales, a newsletter for technical founders who have become their own first sales rep.

If you want to explore more of his thinking, here are three places to start:

1. TFT:You Hit $500K ARR So Why Did Growth Suddenly Stall?

2. TFT: I was Rejected by My Buyer’s AI Agent

3. The Revenue Flywheel System

In this post, he breaks down a simple rule for using AI in sales without burning trust:

AI drafts. Humans decide. Trust is the asset.

Here’s David.

---

Hello 👋🏻

Most founders ruin AI in sales the same way.

They slap an AI bot on top of a broken system then act surprised when it doesn’t work.

Of course it doesn’t work.

AI cannot fix a broken sales system (more on that here)

Sales is about problem solving with another human. You need trust to do that. AI used correctly helps you build trust.

I’ve seen AI destroy trust with a customer, because I was trying to implement AI tools too fast.

So here’s the mantra I now run my sales process by:

AI drafts. Humans decide. Trust is the asset.

I like this because my focus with founders is to help them reframe sales as problem solving with another human. This mantra keeps the human in the loop working with another human, which is key in a world of increasing AI.

Today I want to share what I’ve learned as I’ve implemented more AI into my own sales workflow. These examples are from working with technical founders with a great product who are building momentum but don’t love sales and small service businesses like HVAC (Heating, Ventilation, and Air Conditioning) companies implementing their first CRM and sales process where customers still want a human.

My AI Sales Stack

Here’s the simple version of my AI stack today:

• Call notes + summaries: transcript → summary → action items (busywork)

• Research + prep: industry/ICP context, account notes, stakeholder mapping (busywork)

• Writing support: tighten your drafts for follow-ups and proposals (busywork)

• Process hygiene: turn insights into next steps and reminders (busywork)

Notice what’s missing: nothing here replaces human-to-human engagement. Because sales is just problem solving with another human. AI is great at removing the admin drag.

Step 1: Don’t Automate Until You Can Name Your Process

Before you build a single AI workflow, ask yourself one question:

Do you have a sales process that actually works today?

Not “we follow up when we remember.” A real, repeatable sequence.

If you can’t write down the steps you take for every deal, AI won’t help you.

Here’s what I want founders to do first: write out the 8–10 steps of your process, then identify which steps are:

• Busywork: admin and pattern work (safe to automate)

• Trust moments: anything that changes the relationship (keep human-led)

This is the first step to your AI implementation. The items marked “Trust” will stay human-led. The “Busywork” can be automated with AI.

Look at your last 5 closed deals to get a clear picture of what those steps actually are. If you want more on building out the sales process I talked in detail about that here.

Step 2: Start With What AI Is Actually Good At

For me, AI became immediately useful in two places.

Getting Clarity on My ICP in Unfamiliar Industries

I came from energy and industrial manufacturing. But I’ve sold into industries I’m less familiar with: HR companies, virtual assistant companies, HVAC, website design, and business consulting.

That’s where AI became a cheat code. It didn’t magically know the customer. But it helped me rapidly understand what’s happening in their industry right now, the common pain points and trends, and the language buyers actually use when they describe their problems.

AI shortened the learning curve. I still had to decide what mattered.

Here’s the exact research sequence I use to get up to speed on a new account:

The ICP Research Prompt Sequence:

“I’m going to engage with [customer name] from [company name] on how the Revenue Flywheel System (my framework) could apply to their business. Based on the resources you can pull on their company, what challenges might they be facing? [included company website]”

Real example — DFW HVAC Company (HVAC/plumbing):

I ran this sequence on a 65-year-old, highly-rated independent contractor. The research surfaced three things I wouldn’t have found in a 30-minute call: their own customers didn’t know they did plumbing (a cross-sell blind spot), they had no maintenance or membership program (a recurring revenue gap), and an underused referral engine all quantified against industry benchmarks before I walked in the door.

AI got me to the conversation faster. I still had to ask the right questions.

Other prompts I’ve used with Deep Research:

Prompt 1: “Research [company]. Give me their founding, ownership, locations, service lines, certifications, and estimated headcount using only public sources. Cite each fact.”

Prompt 2: “Build a competitor table for [market]: rating, review count, services offered, and ownership type. Flag PE-backed players.”

Prompt 3: “Give me [industry] benchmarks for customer lifetime value, retention, and recurring-revenue programs, with sources.”

Step 3: Treat AI-Generated Communication Like a First Draft. Not the Final Word.

Here’s where AI has burned me and probably burned you too.

When I give AI an email and say “respond using my framework,” it often produces generic wording, makes things up, or misses the nuance of where the customer actually is.

When that happens, the outcome is predictable: best case, the message gets ignored. Sometimes, the customer replies confused. Worst case, you lose trust and the opportunity quietly dies.

I once used AI to draft a follow-up email after a call. It sounded polished, but it included one confident detail that wasn’t true. The buyer replied politely and corrected it. I didn’t lose the deal on the spot, but I could feel the trust drop. Everything got harder.

That’s the danger zone: when AI sounds certain, but you haven’t verified the facts.

The Better Workflow I Use Today

Instead of asking AI to “write the email,” here’s what I actually do:

Step 1 — I write my rough draft first:

“Hi Marcus, Great call today. Wanted to follow up on the service-to-sales handoff issue you mentioned. Your service techs are walking past replacement opportunities every day. We can help capture those signals automatically and route them to your sales team so nothing falls through. Worth a quick 30-min call to map it out?”

Step 2 — I add context and keywords only I know, then ask AI to tighten it:

Prompt: “Here’s my rough draft email. The key context: the customer is an HVAC owner, we talked about his service and sales teams not communicating, and he’s concerned about being too “salesy” with existing customers. His name is Marcus. Tighten this email keep it under 100 words, keep my specific details, and make the ask clearer. Do not add anything I haven’t mentioned.”

What AI gives back:

“Hi Marcus, Following up on our conversation. Your service team is walking past real revenue every day, units due for replacement, accounts ready for an upgrade. We can help flag those signals automatically and route them to sales before they go cold. Would a 30-minute call this week help us map out exactly where those handoffs are breaking down?”

The difference: The first version was mine and not bad, but the second version is tighter and cleaner. I just ensured that every fact in it came from me. AI refined the words. I owned the substance.

Because in sales, the details are where trust is built.

Step 4: Automate the Busywork First

If you want the fastest safest wins with AI, start with what steals time from you, but doesn’t require trust.

Here are the first automations I’ve gotten real value from with specific tools and outputs:

AI note-taking on customer calls (my biggest ROI)

Tools: Notion AI, Granola, Firefly, or your CRM’s built-in transcript.

What I do: Record every customer call with Notion AI. After the call, I get the transcript as well as a summary with action items.

If you just have your transcript use this prompt:

“Summarize this call in 5 bullet points. List: (1) what they care about most, (2) objections raised, (3) their decision criteria, (4) confirmed next steps, and (5) anything I should follow up on that didn’t get resolved.”

This gives me a clean brief I can action in 2 minutes instead of trying to remember the call 9 days later. I love that Notion autogenerates the actions and summary. That way I don’t have to go back and run the prompt myself.

Proposal refinement

What I do: Write the proposal myself first. Then paste it into AI and ask:

“You’re a skeptical technical buyer. What’s unclear in this proposal? What would make you hesitate to sign? What’s missing?”

The output shows me the gaps before the buyer finds them. Which is helpful with new buyers. I also like to combine this with my transcript after the call. While it’s not perfect. The AI helps me pick up on questions the customer asked that can help refine the proposal. For that I like to use a prompt similar to the one above, it looks like:

“Here’s my transcript from the customer call. What was unclear to the customer? What was missing from the proposal? What are 3 key follow-ups I should focus on?”

Yes, you were listening during the call so you should already know this. But what I’ve seen is the AI will pick up on at least one item I did not in the call. Which means when I go to draft my follow-up e-mail, I’m hitting on all of their concerns, not just the 2 or 3 that I picked up on.

How this looked with the HVAC company I’m working with was I knew they needed a sales process and to connect their service and new installs teams, but I missed a small opportunity with their follow-ups. When I ran my call through AI with a similar prompt to the one above I got this back as the first action:

Turn their existing customer data into a proactive CapEx pipeline. They described a gold mine: they already track equipment lists, repair spend per unit, and customer budget cycles (January, April, even July starts). They’re already having consultative conversations about repair-vs-replace and planned expenditures. But it sounds informal, it happens when they remember or when the customer’s unit fails. The Upsell Driver’s expansion fit diagnosis maps directly here.

I might have gotten to this action step eventually, but it for sure would not have been as fast.

Draft recaps after meetings

I do this a little different now since Notion actually gives me the full summary. But if you are working with a tool that just gives you the transcript use a prompt like this:

Paste meeting notes and ask:

“Write a 5-sentence recap email I can send to the customer. Include: what we agreed on, what the next step is, who owns it, and by when. Don’t add anything I haven’t mentioned.”

The consistent principle: AI handles the admin drag so human energy goes where it actually matters.

Step 5: Keep Humans in the Trust Moments

There are parts of sales you should not automate, because they are the high trust moment. The trust moment filter I use: if it changes the relationship, a human goes first.

That means discovery calls and asking the real questions. Objection handling, especially on pricing, timing, or “do we need this?” Negotiation and terms language. Pushback emails when the buyer is tense, confused, or skeptical.

Really anything emotional or high-stakes needs a human in the loop. I know that may sound obvious but you might be surprised what people will try to automate.

I’ve seen AI Agents conducting interviews. And even when they call the prospect on the wrong day, they continue with the interview. Interviews in my mind are still a human to human interaction.

Where AI can help without replacing you: prep (likely objections, agenda, questions to ask), after-call recap drafts, risks and next-step options, and tightening your own wording after you’ve written the truth. I’ve used the prompt:

“I’ve got a little more insight here talking to the owner of the company. They are looking to expand more in the commercial space, which leads perfectly into the Flywheel framework because these are relationship driven contracts. I also learned that their primary salesman for this role is leaving the company. I still have a meeting with [company leader] on Monday and I want to prepare some key insights and questions to help me understand their situation and to help them understand how the flywheel system will help them.”

This prompt gave me 5 key topics/questions that I needed to make sure I covered on the call. Which meant I focused our time on their challenges and not my framework, or just basic questions. I was able to tailor the call to the key insights that would help them the most.

AI can support these moments. But the human should handle the human.

Sales is problem-solving with another human. Founders forget this because they think sales means persuasion, which it doesn’t. It means understanding a problem clearly enough that the customer feels seen.

The Real AI Sales Stack Summary

Don’t overcomplicate it.

1. Audit your sales motion

2. Automate low-stakes friction

3. Keep humans in trust moments

4. Iterate one step at a time

And remember the rule:

AI drafts. Humans decide. Trust is the asset.

Action Step

If you do one thing this week, do this:

The “AI Notes → Insights → Follow-up” Workflow:

Pick your note-taking tool (Notion AI, Granola, Firefly, or built-in CRM transcript).

Record your next 3 customer calls.

After each one, have AI summarize what they care about, their objections, their decision criteria, and the confirmed next steps.

You do a human review.

Then schedule follow-ups based on those insights.

This is a high-leverage workflow because the customer interaction stays completely human while AI handles the entire admin layer.