Grokbot Marketing Ops: ICP Research with Cerebro and Clay

If your day is a stack of email addresses, calendars, Slack, and chat accounts, the bottleneck is not another dashboard. It is context switching. Matthew, who builds marketing-ops automations at SpaceX AI, runs Grokbot as a chief of staff that can see all of that traffic, separate tasks from FYIs, and hand a first pass to specialist bots.

The same Grokbot marketing ops pattern can grow a tiny consumer product. On a live build of Thursday Arena, a new game studio with a handful of people, he used a go-to-market bot named Cerebro to decide who should sit in ad slots — not whoever writes a check — then sketched how Clay and Amplemarket would turn those personas into personalized partnership outreach.

You will be able to stand up the agent team, run an ICP workflow against a real product, and know what to do after the bot returns a company list. The method is: one coordinator bot with full inbox access, specialist bots for the work, Cerebro for who you are actually trying to reach, then a short list and a sequenced email that cites live numbers.

Key takeaways

Give a chief of staff Grokbot every inbox you actually use — email, calendar, chat, Slack, personal and professional — or it cannot run a self-completing task list.
The coordinator should not do all the work. It should decide which specialist bot takes the first stab, then let those bots talk to each other when they need data or a draft.
For a large sales org, that routing is how you handle territory disputes against rules of engagement instead of a room full of people.
For a three-person consumer product, start from who you want to entertain, then monetize with cohesive ads and partnerships. Skip pay-to-win if you would not play that game yourself.
Use the Grokbot marketplace to seed ideas (memories, routines, integrations already loaded). Build your own when you can describe the job and the connectors.
Cerebro’s job is ICP, personas, segments, scoring, and signal recipes — a sharper picture of who to put in ad slots, not a generic “advertisers” bucket.
Player growth (PLG) and partnership / ad-slot sales (SLG) feed each other: live user and match stats belong in the cold email, not in a separate data-team ticket.
Clay finds companies and people who match the persona. Amplemarket sequences the mail. A Grokbot on the inbox watches who takes a bite. Human-edit the copy; treat first-pass one-liners as something to react to.

Who this is for — and who should skip it

This is for a founder, operator, marketer, or ops lead who will actually connect tools and live with an agent team. Matthew’s day job is marketing ops (combined with revenue ops) supporting marketing and sales at a large company. The same sequence was applied, live, to a brand-new game studio with a single title at thursdayarena.com, a sponsor form, and a decision not to charge players for power.

You need Grokbot, a willingness to grant inbox and calendar access, and comfort talking to bots the way you would message a colleague. Clay and Amplemarket are the later-stage connectors for lists and sequences; they were not wired up in the live run. Skip this if you want pay-to-win mechanics, if you are happy to sell any unrelated ad that pays, or if you need a click-by-click Clay implementation with live credentials. The speakers treated that last mile as work to do off-stream.

Give a Grokbot chief of staff every inbox

What you are trying to accomplish: stay sane. Reduce context switching. Distill a firehose into tasks versus FYIs, then get unstuck without you being the router.

Prepare the real set of sources, not a polite subset. Matthew’s version is tapped into all email addresses, calendars, different chat accounts, and Slack, across personal and professional use. If you leave a channel out, the self-completing list is fiction.

Stand up a chief of staff agent and point it at those inboxes on a regular cadence. Its job is to receive what comes in, decide which subset is a task, and identify which of your other specialist bots is best positioned to take a first stab. Matthew’s inbox template in the marketplace is named Inbot. Treat the chief of staff as the person you talk to — with the unlock that it can finish work, not just chat about work.

I call it a self-completing task list.

Good looks like: incoming noise becomes a short task list, FYIs stay FYIs, and a specialist already has a first pass before you context-switch. Failure looks like a chief of staff you still have to brief on every inbox because it cannot see them.

Route the first stab to specialist bots

On a combined RevOps / MarOps team, a recurring task is territory planning and territory disputes. A salesperson meets someone at an event, exchanges contact info, then realizes a colleague owns the account. Tickets pile up: should I work this or not? In the past that took human judgment, a walkthrough of the rules, and a reminder that someone clicked the wrong button.

The automated version is: understand what is actually going on, align it to rules of engagement, and state the policy moving forward. That is the pattern to copy even if you do not have a massive sales team. Encode the rule, let a specialist bot take the first pass, keep a human on exceptions.

Other specialists on the live team were Dr. Eggbot (helps you make new bots), a data scientist (charts and pulses), and Cerebro (go-to-market / ICP). Label bots as you add them — GTM for Cerebro — and pin the ones in active use. Message them the way you would message colleagues on iMessage or Slack. When you add a new bot, have the chief of staff or Dr. Eggbot send a brief of what you are actually building so the new hire is not guessing.

Monetize a small product without pay-to-win

A three-person studio with one game is not a large sales org. Matthew’s filter is blunt: every business exists to connect with people and do something valuable, entertaining, or memorable. From the consumer side, get the game to the people who will be most entertained by it. From the “pay the bills” side, people pay or you run ads.

Thursday Arena’s call was no pay-to-win. Charging for power is a mechanic the builders would not enjoy, so they would not ship it. Ads and sponsorship were the path: screen estate, a sponsor form on the site, and later partnerships that might also drive traffic.

Do not take the easy ad sale. If anyone who pays can appear on screen, players get random unrelated ads. You get money and a worse product. The Grokbot marketing ops job is to fill those slots with brands that feel native to the session — other games, snacks, energy drinks, merch, news, the five-minute break — so the ad is part of the experience.

Growth then splits into two buckets that feed each other:

PLG / consumer growth: get more people to play.
SLG / sales-led growth: a smaller set of conversations to sell ad slots and set up partnerships. Those emails should cite PLG numbers — users in Y days, matches in the last hour — not a generic “we have a game.”

Partnerships take longer than a self-serve ad unit. Paid social can fill the top of the funnel, but only if there is something on the other end. Do not spend to acquire users you cannot hold or monetize cohesively.

Start in the Grokbot marketplace, then build your own

The marketplace lives at x.ai/bot/marketplace. Published bots come with memory (preloaded job context), routines (Cooper, a news agent from Tommy Hanson, sends a tight daily Slack briefing at 8:00 a.m. and weekday midday competitor alerts only when something clears the bar), and integrations (Slack is the example called out for knowing what is going on in the business).

Thursday Arena’s in-game bot roster started as marketplace bots. You can share any bot as a template: hit share, it packs itself up. That is how Cerebro moved from Matthew’s team onto the live build.

The marketplace seeds amazing ideas and like you should go to that first, start there.

Unlike tools you copy because the product is too hard to learn, Grokbot is simple enough to build from scratch once you can say what you are trying to do and which connectors you need. Use the marketplace for a first version. Then think, and replace it.

Run Cerebro to find your ICP for ad slots

Cerebro is the go-to-market bot. The metaphor is Professor X at the Cerebro helmet: you are not only listing an industry and a revenue band. You are asking who the human beings are who would be excited if they knew this thing existed, then finding them.

Prepare three things before you invoke it:

A one-pass brief of the product (have the chief of staff or Dr. Eggbot send it so Cerebro knows what the ad slots even are).
A point of view on monetization (ad-supported, not pay-to-win; cohesive brands, not whoever pays).
Any live player insight you already have, even if it is a theory.

Install the template, pin it, assign a GTM label. Skip Clay and Amplemarket plugins until you have an ICP worth searching. If a packaged skill such as “find my ICP” does not show up in the slash menu, try invoking it in natural language anyway — in the live run it was still in memory and fired.

Voice-dictate into the bot. Talk in first principles: who would you actually reach out to and say you have a convened, highly engaged audience in front of a product they would care about. Seed it with adjacent ideas (other indie or casual games, snacks, energy drinks, merch) so it is not guessing in a vacuum.

What good ICP output looks like

Expect a bundle, not a slogan:

ICP: who you are trying to reach when you put the helmet on.
Personas: avatars for the main categories — in the live pass, brands / sponsorship, partnerships and platforms, streaming apps, game platforms that could send traffic.
Segments: another grouping of those personas, with a priority order for who to go after first.
Scoring: how you would rank a target once you have more than a silhouette.
Signal recipes: the data that makes the picture higher resolution so the email is not “game to advertise on, click here.”

A signal recipe is how you stop targeting “any snack brand.” Example from the conversation: a young brand putting out products that are unhealthy-but-made-healthier, trying to reach people who love junk food and want something better. Niche on purpose. The incentives line up; they answer because it is not any email.

Feed Cerebro who is actually on the board. Thursday Arena’s early traffic included the livestream and Grokbot / SpaceX AI fans, plus a working theory: downtime games, a few minutes in the middle of the day, puzzle players, Words With Friends energy, a five-minute break between meetings. After that context, segments came back as work-break casuals and scroll substitutes — forget the Instagram reel, play one round. Those are stories you can tell, not just buckets.

Partner ideas that landed as useful to react to: news apps and newsletter briefings (Morning Brew as a named example), retail / stock-check apps (check the market, then one more round), body-reset / walking. A partnerships one-liner also came back (“We sit in the same five minutes as a news skim, a snack run and one more casual round”). Treat that as draft. Clean the slop. Human touch is the point.

It’s always about inspiring ideas. It’s giving us something to react to and then we can put the human touch on it.

Lock a segment when the personas match how people actually engage: when they open the app, what they want from five minutes, what else they already do in that window. Then move to lists. Do not spray advertisers before that lock.

Turn personas into Clay lists and Amplemarket sequences

What you are trying to accomplish: take conceptual ICP work and get names you can email — companies first, people second, a sequence third — with copy that cites your numbers and their specifics.

Clay’s job in this stack is to search a database of companies and people for who fits the description you just locked. In the live session the Clay MCP was not connected; the bot still had the web and could research. Treat a disconnected plugin as a simulation or a research pass, not a finished list. Ask for a short list on purpose (ten companies to start) so you can judge fit before you industrialize.

What happens next, in order:

Take the company list. Ask for the people at those companies you would actually reach.
Load an Amplemarket (email sequencer) campaign. Many tools like this exist; this is the one named in the plugin list.
Do not send “we have a game, you’re an advertiser, let’s meet.” Reference live product stats the bot team already has, plus specific factors about that company that make them the right advertiser or partner.
Amplemarket sends. Put a Grokbot on the inbox to see which advertisers took a bite.

Good looks like a short, ranked list you recognize as native to the session, emails a human would send, and replies triaged without you living in the inbox. Failure looks like a generic blast, plugins you never connected, or stats you cannot defend because nobody asked the data bot.

When the context is on the agent team, they can just hang out and chat with each other.

That is the operational difference versus the old handoff: partnerships should not file a ticket with a data team to pull last hour’s matches. Cerebro asks the data scientist bot, gets the opener, and you edit.

Use case	Bot / tool	Input needed	Output	Owner
Inbox triage and self-completing tasks	Chief of staff, Inbot	All email, calendars, chat, Slack	Tasks vs FYIs, routed first passes	Founder or marketing ops
Territory disputes / rules of engagement	Specialist bot + policy	Tickets, account ownership, ROE	Who should work the account, policy going forward	RevOps / MarOps
ICP for ads and partners	Cerebro	Product brief, monetization rules, player theory	ICP, personas, segments, scoring, signal recipes	GTM / founder
New specialist bots	Dr. Eggbot	The job and connectors	A new bot, shareable as a template	Whoever owns the workflow
Live product stats	Data scientist bot	Signups, matches, leaderboard, time window	Charts, pulses, numbers for outbound	Founder / growth
Company and people search	Clay MCP	Locked personas and signal recipes	Target companies, then contacts	GTM
Personalized sequences	Amplemarket + inbox Grokbot	People list, stats, company-specific factors	Sent campaign, replies flagged	Partnerships / founder

Prompts, bot instructions, and workflows

These are cleaned from the live dictation and instructions. They are not improved into a different method. Where the speakers only described the next action, the block is labeled as reconstructed.

Brief the new GTM bot from the rest of the team (described, not a full prompt):

Reconstructed from the speaker's description.

To chief of staff or Dr. Eggbot:
Send Cerebro a brief of what we are actually building — the game, the ad slots, and the fact that we are not pay-to-win — so it has context before it runs find-my-ICP.

Invoke ICP for ad slots:

Reconstructed from the speaker's description.

Cerebro, we want to find our ICP in terms of who we should put in these ad slots. Run find my ICP.

Brainstorm who we want filling these slots. If you are sitting and playing this game, you might think about other fun games to discover, or snacks while you pass the time. Come up with creative ideas for who we would reach out to and say: we have an ad slot and a highly interested, highly engaged audience in front of a product they will care about.

Refine segments from how the product is actually played:

Reconstructed from the speaker's description.

Hey Cerebro, we are thinking a little more about our segments based on what you shared.

Thursday Arena is quick and easy to play. Consider users who are more casual gamers — hop in during a five-minute break, or do something with their brain between meetings at work.

That is who uses the game. What other products and services are those people interested in? Maybe they also read the news and get smart between meetings. Use that to think about partnerships and advertisers.

Hand locked personas to Clay:

Reconstructed from the speaker's description.

Take what we have landed on for player personas and the brands we want as advertisers or partners. Plug that into the Clay MCP.

Search its database of companies and people for who fits this description. Start with a list of 10 companies we might reach out to about advertising or partnering.

After the company list (workflow, not a pasted prompt): ask for the people at those companies; draft Amplemarket sequences that cite specific product stats and company factors; connect a Grokbot to the inbox to watch who replies. Do not lead with a generic “you are an advertiser” opener. In the live pass, revised openers used facts such as shipped this morning, about a thousand players, and hundreds of matches in the last hour — and explicitly avoided leading with weaker vanity framing.

Measurement and business impact

No time-saved percentages or cost figures were given. The checkpoint is whether the bots can pull the number you need without a cross-team ticket, and whether those numbers are good enough to lead an email.

On the live game, the data bot was already watching: more than 1,000 users logged in; nine names on a platinum leaderboard; an early spike around 600 signups; later, more than 3,000 matches within a few hours; a 15-minute pulse of stats; a request to add total users over time. One outbound-ready snapshot cited about 1,000 players and 440 matches in the last hour. Those figures are a moment in a livestream, not a benchmark you should promise.

Qualitative test: Cerebro can ask the data scientist for email stats and get a revised opener in the same thread. If you still wait on a separate data team to “pull the latest,” the agent team is not actually sharing context.

Pitfalls and guardrails

Random ads. Selling every slot to whoever pays gets you money and a disconnected player experience. Choose brands that belong in the same five minutes as the session.
Pay-to-win. If you would not play it, do not ship it as the business model.
Generic cold email. “Game to advertise on, click here” is the low-resolution picture. Use signal recipes and live stats, or do not send.
AI first drafts. Partnerships one-liners will read like slop. React, then rewrite. The bot is there to inspire, not to publish.
Skills may not travel with the template. Slash commands can be missing even when the skill still runs from memory. Invoke it in language; do not assume packaging is complete.
Plugins that are not connected. Clay and Amplemarket were named and described, then skipped live. Do not pretend you have a sequenced campaign until those accounts exist. A web-research fallback is not the same as Clay’s database.
Paid acquisition with no far side. Traffic without a cohesive offer or a way to pay the bills is just spend.
Wrong-size process. Territory-dispute automation is for a large sales team with real ROE. A three-person studio still needs ICP and outreach; it does not need a war room about account ownership.
Human still in the loop. Rules of engagement, brand on the screen, final email, and who you actually partner with stay with people. Bots take the first stab.

7-day implementation plan

Day 1. Install Grokbot. Create the chief of staff. Connect every inbox, calendar, chat account, and Slack you actually use. Browse x.ai/bot/marketplace. Note Inbot, Cooper-style routine bots, and any GTM template you can share in. Do not build Clay sequences yet. Document one workflow you already hate (inbox triage or “who is this product for”).

Days 2–3. Share or install Cerebro. Label it GTM. Have the chief of staff or Dr. Eggbot brief it on the product, the no-pay-to-win rule, and whether you are selling ads, sponsorship, or both. Dictate the first ICP pass. Read back ICP, personas, segments, scoring, and signal recipes. If the slash skill is missing, ask in prose anyway.

Days 4–5. Feed who is actually using the product — even a theory (five-minute break, casual, scroll substitute). Lock one or two segments. Write signal recipes specific enough that “any snack brand” would fail them. Human-edit partner one-liners and the story of why the product exists. Stand up or promote a data bot so the next email can cite real volume, not vibes.

Days 6–7. Connect Clay if you have it; otherwise run a short research pass and label it as such. Ask for ten companies, then the people. Draft Amplemarket sequences that cite a handful of live stats and one true thing about each company. Put a Grokbot on the inbox for replies. Do not scale the send until a human has read the first batch. If plugins are still off, stop at a ranked list and copy deck rather than pretending the sequencer is live.

Close the loop on who you are for

The business outcome is not more bots on a sidebar. It is a coordinator that keeps you out of the inboxes, an ICP sharp enough that ad slots feel native, and a short path from persona to a personalized email that can cite what happened this morning. Start with one Grokbot — the chief of staff or Cerebro — and document that first workflow before you add Clay.

FAQ

Do I need a large sales team to use this Grokbot marketing ops setup?

No. Territory disputes and rules of engagement are the large-org example. The same coordinator-plus-specialists pattern was applied to a three-person game studio with one title. What changes is the work you route: ICP, ad-slot fit, and a small number of partnership conversations instead of account-ownership tickets.

Grokbot marketplace vs building a bot from scratch — which should I do?

Start in the marketplace. Bots arrive with memory, routines, and integrations, which is enough to steal a good idea. Matthew’s point is that this is not a tool you copy because it is too complicated to learn. Once you can describe the job and the connectors, build your own and share it as a template.

What is Cerebro, and do I need that specific bot?

Cerebro is a go-to-market template whose job is to find an ICP: personas, segments, scoring, and signal recipes, with a bias toward who would actually be excited the product exists. You need that function. You do not need that name if you build an equivalent GTM bot and brief it the same way.

Ads vs pay-to-win — what did they actually choose?

No pay-to-win. The studio would not play a game that sells power, so they would not ship one. Monetization moved to ads, sponsorship, merch, and partnerships, with a hard preference for brands that belong in the same break as the game rather than unrelated buyers.

PLG vs SLG — which one is this?

Both, on purpose. PLG is getting more people to play. SLG is the smaller set of people who buy ad slots or form partnerships. The flywheel is that player numbers make the outbound stronger, and the right partners send more players. Cite the first bucket in the second bucket’s email.

Do I need Clay and Amplemarket, or is Grokbot enough?

Grokbot is enough for chief-of-staff routing, ICP thinking, and drafting. Clay is the named way to turn a persona into companies and people. Amplemarket is the named sequencer. There are other tools like it. In the live run those plugins were off, so the last mile was described, not executed. Do not wait on them to start Cerebro; do not claim a campaign is running until they are actually connected.

What is a signal recipe?

It is the data that turns a fuzzy silhouette (“snack brands”) into something you can say in an email (a new brand, a specific product bet, an audience that already wants that bet). Without it you are sending a generic ad-slot pitch. With it, the incentives are more likely to match and the message is not interchangeable.

Where does a human still have to stay in the loop?

Connecting inboxes, locking the segment, deciding what may appear on screen, rewriting outbound, and handling exceptions to rules of engagement. Bots take the first stab, pull stats from each other, and watch the inbox. They do not get to publish slop or sell a disconnected ad because it was easy.