Back to blog
AI & GTM Engineering

How 8 Grok Bots run my outbound and booked 27 meetings

Xavier Caffrey
Xavier CaffreyNovember 4, 2026 · 10 min read
How 8 Grok Bots run outbound at OneAway: the bot for each step from lead list to CRM, the human checks, and the four follow-up rule.

To run outbound with Grok Bot, give each step of the funnel its own bot with its own tools: one plans the campaign, one builds the lead list, one enriches the data, one writes the copy, one answers interested leads, one preps your calls and one keeps the CRM up to date. I run my whole outbound motion this way with eight bots, I check their work at three points, and the rule that booked most of my 27 meetings is simple: reply to an interested lead within two minutes, then follow up four times.

I recorded the video above walking through the streets of Bali, with nothing in my hands, so this post is the written version with the details in order. It follows the work the way it actually happens at OneAway: from the ideal customer profile to the CRM.

You can name your bots anything you like. Mine have names because Grok Bot feels like talking to a team, and names make it easier to tell who did what.

Quick answer

Eight bots, one job each: Jacob hands out tasks, Mark plans campaigns, Cody builds lead lists with AI Ark, Pradeep enriches the data with scrapers, Bobby writes the copy, Sally replies to interested leads, Reid writes call briefs and Harper runs the CRM. They coordinate in a group chat. I check the lead list, the scraped data and the copy before anything is sent. The whole setup replaced about $5,500 a month in people for a $200 a month subscription.


The 8 bots at a glance

  • Jacob, chief of AI — Hands out tasks to the other bots from a Notion board.
  • Mark, head of growth — Writes the ICP pages, plans campaigns with me and creates tasks for the outbound bots.
  • Cody, lead lists — Builds the list of people to contact, using AI Ark.
  • Pradeep, data enrichment — Finds the facts the copy needs, using Apify, Context.dev, Firecrawl and Scrape.do.
  • Bobby, copywriting — Writes the cold emails with a skill file trained on Josh Braun’s teachings.
  • Sally, replies — Answers interested leads within two minutes and follows up four times.
  • Reid, call briefs — Writes me a brief before every discovery call.
  • Harper, CRM — Keeps the CRM up to date after each call and sends the follow-up email.

Jacob and Mark plan the work

Every setup like this needs a chief of AI. Mine is Jacob, short for “just a chill operating bot”, because Jarvis felt overused. Jacob’s job is to coordinate the other bots. He creates cards for everybody on a Kanban board in Notion, so he oversees everything, but he plays a small part in the outbound itself. I explained that board step by step in how to connect Grok Bot to Notion.

The most important bot is Mark, my head of growth. Where Jacob is general, Mark only does outbound. He keeps the cold email going, writes ICP pages (ICP means ideal customer profile: the kind of company and person you want to reach), which we keep in a GitHub repo, brainstorms with me which campaigns to run, and creates the tasks for the outbound bots.


Cody and Pradeep build the list and the data

Cody builds the lead list

Once Mark and I approve an ICP, we tell Cody who we want to go after and he starts building the list. But no LLM builds a lead list on its own. You have to give it access to tools that are very good at one thing. For lead lists, ours is AI Ark, a B2B data tool, connected to Cody inside Grok Bot. With that connection, list building is easier and I trust the quality of the list.

I still check it. Cody sends me a preview of the leads and asks if they look right. I say yes, or no because of X, Y and Z.

Pradeep enriches the data

Next, Pradeep enriches the leads with whatever the campaign needs. Say I tell Mark I want to target an ICP and mention which CRM each company uses: “Saw that you’re using HubSpot for your CRM.” Basic, but we first need to know whether they use a CRM, and which one.

Like Cody, Pradeep works through tools. He’s connected to four: Apify (an app store for web scrapers; Apify describes its store as “ready-to-run tools” to scrape, crawl and extract web data), Context.dev, Firecrawl and Scrape.do. There are two reasons for so many. I like testing which scraper is best, and some websites need a stronger, more expensive scraper than others. In my experience Firecrawl handles sites behind Cloudflare’s security well, but these tools change all the time, so I’d rather give the agent several and make sure it’s never blocked. If you want the method behind this step, see B2B data enrichment from scratch.


Bobby writes the copy, and I still edit it

Copywriting is the hardest part to hand to a bot. I’m still the one who edits the copy and has the final say, and if you think you can do this part better, I’d like to compare notes.

What’s working for me now is a skill file built on the teachings of Josh Braun, a well-known voice in outbound. It makes Bobby write better emails than you get from asking a chatbot to “write me a cold email”.

Bobby also helps me choose the variables Pradeep needs to scrape. Nine times out of ten, the variables for a cold email are about the company, not the person. Most people don’t have enough public information online to personalise on. When someone does, use it. But if your goal is to grow fast, company variables are more than enough. Bobby knows that, and so do Cody and Mark.

The bots work in a group chat

All of this happens in one group chat. Grok Bot lets you add several bots to a chat where they talk to each other to finish a task. xAI’s announcement of Grok Bot describes the same feature: bots in a group chat “can coordinate on their own”. Launching an outbound campaign has a lot of moving parts, so it helps that the bots can see what’s blocking someone else’s work. And every bot knows one thing: there’s a step where a human checks the work, and the human is me.


Where I still check the work

I don’t want to be removed from the process. I want to keep a pulse on it. I check three things before a campaign goes out:

  • The copy — Most of the time I’m writing it or editing it myself.
  • The lead list — Did we get a filter wrong? Did the bot misread something in the ICP page and add a filter I don’t want?
  • The scraped data, and how it reads in the email — If we pulled the CRM, the email has to say “Saw that you use HubSpot”, not “We found that this company uses HubSpot”. That would be awful. My job is to catch the moments the bot gets it wrong or makes something up, because sometimes it does.

The golden rule: follow up four times

Launching a campaign is only half of it. What happens after a lead replies has been the biggest weakness of my agency, OneAway, for the last four years. Campaigns took so much time, money and effort to launch that we didn’t have the attention left to help clients with the middle of the funnel. Grok Bot fixes that too.

When an interested lead comes in, Grok Bot hears about it through a webhook from the tool we send with. A webhook is simply one piece of software telling another that something happened; here, the event is “an interested lead replied”. Sally replies within two minutes. That speed is why we booked around 27 meetings when we launched campaigns with Grok Bot.

Then comes the golden rule. If you take one thing from this post, take this: if a lead shows intent, the bot follows up four times, and the count starts again every time they reply.

Here’s what that looks like. I email John. John says he’s interested. I reply, then follow up a second and a third time. John replies to the third one, but still hasn’t booked. Now I follow up four more times after that reply.

What to say in each follow-up

People worry about sounding repetitive. You’re overthinking it. All you need to do is follow up:

  • Offer times — “I’m free at this time or this time, or grab a slot here,” with your booking link.
  • Offer new times — “That time got taken. Does this one work?”
  • Change it up on the fourth — Before her fourth reply, Sally does a quick search on the lead’s company domain and mentions one relevant fact, for example: “Since you’re selling to CMOs at fintech companies, I thought you’d be interested in X.”

Other than that, the job is to stay top of mind. Sally does that for us.


Reid and Harper take it from booked meeting to CRM

Once a meeting is booked, Reid writes me a discovery call brief. My calendar is full most of the time, so I want a quick reminder: who this person is, what their company does, and whether they really fit our ICP. It’s a double check that the lead fits my offer.

After the discovery call, the lead moves to a separate chat for the bottom of the funnel and is added to the CRM. From then on, Harper handles everything about that lead: the call transcript (we record calls with Grain), the next steps and the CRM, which is our source of truth.

Harper also writes the follow-up email after the call: here’s everything we talked about, and here’s our next call. That’s my non-negotiable rule. Every discovery call ends with the next call booked.


What this replaced

Full disclosure, I used to pay people to do all of this. One person cost me $2,500 a month just to build cold email campaigns for my agency. Another, who managed the inbox and took discovery calls, cost me about $3,000 a month. Both were good, but not exceptional. That’s about $5,500 a month.

Today, a Grok subscription that costs me $200 a month covers all of it. One thing to know if you’re pricing this yourself: xAI’s announcement says Grok Bot is available on its SuperGrok plans (and on some Cursor plans) and comes with its own usage, separate from those plans, so check xAI’s current terms before you budget.

If you want to copy the setup, I’ve put every bot, its skills, the setup prompt and the safety rules in a free outbound agent kit.


Frequently asked questions

Can Grok Bot build a lead list by itself?

Not well. An LLM doesn’t have a lead database. Connect it to a tool built for lists, such as AI Ark in my setup, and have it send you a preview before it uses the list.

How many bots do I need to run outbound?

I use eight because each one has a single job, from planning to the CRM. You can start smaller, but keep one bot per job so you can see where something went wrong.

Which parts should a human still check?

The lead list filters, the scraped data and how it reads in the email, and the copy. Those are the places where a bot misreads the ICP or makes something up.

Should I personalise cold emails with facts about the person or the company?

Usually the company. Nine times out of ten, company facts are all you need, because most people don’t have enough public information online. Use personal facts when you can find them.

How fast should I reply to an interested lead?

As fast as you can. My bot replies within two minutes, triggered by a webhook from the cold email tool, and that speed is the main reason we booked around 27 meetings.

How many times should I follow up with an interested lead?

Four times, and start counting again each time they reply. Offer times, offer new times, and on the fourth message mention something specific about their company.

How much does this cost compared with hiring?

For me, it replaced about $5,500 a month in people with a $200 a month Grok subscription. Check xAI’s current plans, because Grok Bot has its own usage on top of the plan.


Key takeaways

  • Give each step of outbound its own bot, and give each bot the specific tools it needs: a lead database for lists, several scrapers for data.
  • Let the bots coordinate in one group chat, but make a human check part of the process.
  • Check the lead list, the scraped data and the copy. That’s where bots get things wrong.
  • Company facts are enough to personalise most cold emails.
  • Reply to interested leads within minutes, follow up four times, and restart the count every time they reply.
  • End every discovery call with the next call booked, and let a bot write the follow-up email and update the CRM.

Want outbound run by AI agents we built?

We run cold email and LinkedIn outbound for B2B companies, from lead lists and data validation to copywriting and CRM sync, and book qualified meetings onto your calendar. Junction grew from around $3M to $20M ARR during the almost three years we ran their outbound. Read the Junction case study.

Book a call