Skip to content
    Pointer Strategy
    Back to Blog
    Growth10 min read7 May 2026

    What's an AI Growth System and What Do I Need to Build One? (May 2026 Edition)

    Luke Marshall on AI growth systems in May 2026: the stack, the signals, the feedback loops, and why your first campaign will stink but iteration ten won't.

    What's an AI Growth System and What Do I Need to Build One? (May 2026 Edition)

    This is a recording of a live session from the GTM ANZ Community, featuring Luke Marshall (Head of Marketing Recruitment, Pointer Strategy) and Ricky Pearl (Founder, Pointer Strategy).

    Why "AI Growth System" Needs a Definition

    The first four months of 2026 have produced a release cadence that even the loudest thought leaders struggle to keep up with. If you feel overwhelmed by it, you are not alone.

    What you see on LinkedIn is the polished demo: one button, outrageous numbers, finalised output. Almost none of it is honest about the weeks or months of iteration sitting underneath. Luke's working definition cuts through that:

    "An AI growth system is a sales or marketing campaign mechanism you can run repeatedly, with AI executing some or all of it, in a continuous improvement loop. The last part is the important bit. It is not set and forget."

    The real value shows up on iteration four, five, even ten. The first run will stink. If you cannot live with that, this is not the play for you.

    These principles apply across paid, funnels, lifecycle, and measurement. This post focuses on outbound, because that is where the iteration loop produces the cleanest signal and where the rewards are biggest for B2B teams selling to hard-to-reach decision makers.

    The Reality Check (Before You Spend a Dollar)

    A few things every founder, growth lead, or RevOps person should know before they touch a tool:

  1. Software cost: roughly $500 to $1,000 per month for the first month of a serious stack. You can cut corners with a bigger TAM or a more generic audience, but efficacy drops and spam risk goes up.
  2. Setup time: real, and not optional. There are no instant returns. Anyone telling you otherwise is selling something.
  3. Audience size: this works best when your target is hard to reach and your deal value justifies effort. If you are selling $50 widgets, the math does not work. If you are selling $50,000 to $100,000 per year, small needle movement matters.
  4. Trust: the first pass will produce underwhelming results. Without internal trust, the system gets killed before iteration three.
  5. The Four Components of an AI Growth System

    Every system Luke has built or audited has the same four pieces.

    1. Audience

    This starts with a list, then layers on enrichment and signals.

    List sources, ordered roughly by trade-offs:

    SourceTrade-off
    LinkedIn manual / Sales NavigatorAbove board, slow, hard to scale
    Microsoft Dynamics + Sales Nav integration100% above board, expensive
    Apify and other scrapersPowerful, faster since Claude Code, sits in a grey zone
    ApolloOne-size-fits-all, decent if your market is broad
    FirmableBest-in-class data for ANZ, especially for niche or licensed categories
    LLM as list builderDoes not work. Don't bother.
    In-browser agents and Chrome extensionsPossible, not recommended over the above

    Enrichment and signals is where most teams stop too early. Getting an email or a job title is easy. Getting cut-through is not. Signals are the markers in someone's online world that suggest interest or fit.

    Examples that have worked:

  6. Following a competitor or category page (e.g. someone who follows the Microsoft Dynamics page probably understands enterprise CRM)
  7. Recent technical job postings at the company (Clay does this well)
  8. Last 50 to 100 LinkedIn posts from the target (e.g. has the prospect written about AI in their last 100 posts?)
  9. Competitor case studies the company appears in
  10. Rotating chairs: a Head of Sales role re-posted three times in 12 months tells you it is a leadership problem, not a talent problem
  11. "This is where the game is won. But you have to run these things multiple times to unlock that information."

    >

    Luke Marshall

    2. Campaigns and Messaging

    The job here is to match what you can find from a signal to messaging that lines up against your target.

    The classic "started a new role in the last three months" angle is over-played. It can still work, but it needs an extra signal layered on top. A more interesting one Luke has tested: a *lifer* in a role. Tech moves fast in most categories. A CRM lead who has been in seat for ten years is a legitimate conversation starter for the leader above them.

    Or the rotating chair example from above, where the message becomes:

    "You are burning through Heads of Sales. Want some help?"

    Right message, right time, right signal. That is the loop you are trying to build.

    3. Tools

    Not exhaustive, but a working May 2026 stack:

    JobTools
    Cold email (sending, warming, account rental)Smartlead, Instantly
    LinkedIn outboundHeyReach (best price-to-value right now)
    List building (broad)Apollo
    List building (precision and ANZ)Apify, Firmable, Clay, Freckle
    OrchestrationClaude Code (Luke's pick), Codex (Nathan's pick), purpose-built agent builders, n8n for workflow-style automation

    If you want a sense of what AI-fluent execution looks like at the role level, Luke covered the same theme from the hiring side in Is AI Coming for Your Marketing Jobs?.

    4. Feedback Loops

    This is the one most teams skip and most teams regret skipping.

    A simple version:

  12. If a contact has not been reached in three or six months, run a recontact with a fresh angle
  13. Run evergreen offers (newsletter, community invite) to reactivate non-responders
  14. Look at performance over a six to twelve month window, not the 48 hours after launch
  15. Track everything in one place. Pointer pipes campaign data to Supabase so we can see what is moving over time, not just per send.
  16. Once you have done the heavy lifting on signals and tools, *introducing new campaigns becomes cheap*. A niche conference is coming up? Enrich previous attendees and people who follow the conference page. Hit them up. Same loop, new angle.

    The Pointer Stack (For Reference)

    Just so this is not theoretical, here is the stack we run today:

  17. Audience: LinkedIn, Apollo, Firmable, Apify
  18. Enrichment and scoring: Clay, Apify, Claude Code (lead scoring)
  19. Sending: HeyReach (Smartlead may join soon)
  20. Orchestration: Claude Code as primary orchestrator, with separate evaluator agents grading each campaign
  21. Feedback loop: Supabase as the long-term data store
  22. Nothing exotic. The leverage comes from running the loop, not from any single tool.

    Watch-Outs (the Things That Break)

    A few that catch teams out:

  23. CDOs hate this. Most of these tools sit outside the official org stack. If you ask permission, the answer will be no. Find a way to work around it without putting customer data, SOC, or ISO posture at risk.
  24. Never run cold email from your main domain. If you get blacklisted, your invoicing and client comms go with it. Rent accounts via Smartlead or Instantly. This is non-negotiable.
  25. Do not generate-and-send. Pure AI-drafted copy reads like AI-drafted copy. Train on your past messaging, templatise where structure matters, and run small audiences first to make sure nothing sounds wonky.
  26. Smaller TAMs need more human, not less. More on that below.
  27. This tech is three years old at most. Anyone calling themselves an AI growth expert with five years of experience is making it up.
  28. Q&A: What People Actually Asked on the Day

    What guardrails matter most for agent orchestration?

    Three things:

    1
    High-level system prompting that codifies the must-dos and must-don'ts for your industry.
    2
    Separate the campaign agent from the evaluator agent. As soon as the campaign agent gets more context, it starts marking its own homework. The evaluator agent should come in cold, with less context, and ask "does this pass the barbecue test?"
    3
    Optional third agent for error-checking and spot-checks against a different parameter set.

    The setup is easier than it used to be. Ask Claude Code (or your equivalent) to design each agent. Have the evaluator return a score and a few lines of feedback. Loop the campaign agent against the evaluator five, ten, fifteen times until the score hits your bar. Run all of this *before* a single real prospect sees a message.

    Ricky's add: you can also run loops *across* LLMs. Draft in Claude, sanity check in Codex, then route through an LLM council where each model wears a different hat.

    How do you make this work in a small ANZ TAM?

    This is the question every ANZ team asks. The North American playbooks assume an effectively infinite TAM. Australia is different. Burn the list and you are done.

    The fix is to *dial up the human* rather than dial down the system. With a TAM of 500, you might:

  29. Run mini campaigns of 10 to 50 contacts at a time
  30. Personalise more aggressively (mutual connections, recent posts, specific signals)
  31. Add a phone call or text based on what comes back from the email or LinkedIn touch
  32. Smaller volumes, but better-targeted, with a tighter human-in-the-loop. Provided the thinking is good, it will outperform a 5,000-contact send to the same audience.

    What if LinkedIn is useless for our category?

    Common in mortgages, finance, real estate, construction, trades, anywhere that operates more like a walled garden.

    Other places to find signal:

  33. Google Maps reviews (scrape and read)
  34. Job postings with industry-specific jargon (rising or falling job counts is a directional indicator on its own)
  35. Industry-specific portals like realestate.com.au or EstimateOne for tenders
  36. Licensing registries (mortgage brokers, master builders, financial planners; most regulated industries publish a check-the-licence database that is scrapeable)
  37. Technographics via BuiltWith or paid alternatives
  38. Manual phone outreach as part of the discovery (e.g. "when is your contract up for renewal?")
  39. Ricky's add: knowing *what* data to target is the creative exercise. *How* to get it is the practical one. The combination of domain knowledge and technical capability is the moat. Marshy will spot signals nobody else sees in a category because he has been in that category for years.

    Key Takeaways

  40. A growth system is a loop, not a launch. First run will stink. Iteration four to ten is where the value shows up. If your team cannot tolerate that, do not start.
  41. Signals beat lists. Anyone can buy an email address. The teams winning right now are the ones combining domain knowledge with technical chops to find the markers nobody else is reading.
  42. Match message to signal. Generic personalisation gets ignored. Specific signals (lifer in role, rotating chair, recent technical hire, follows-this-page) earn the reply.
  43. Separate your agents. Campaign agent and evaluator agent should never share context. Score, loop, score, loop. Run all of it before a single real prospect is contacted.
  44. ANZ TAMs need more human, not less. When the audience is 500 instead of 50,000, dial up personalisation, mutual connections, and direct outreach. Smaller campaigns, bigger care.
  45. Trust is the load-bearing component. Without it, the system gets killed at iteration two.
  46. Three years old at most. Anyone calling themselves an "AI growth expert" with a five-year track record is fabricating.
  47. Your Speakers

    Luke "Marshy" Marshall is Head of Marketing Recruitment at Pointer Strategy. Ex-Google, ex-Facebook. Helped launch Instagram advertising in Australia. 15+ years across BMW, L'Oreal, P&G, and Unilever. He sees what every growth team in ANZ is hiring for and where the real bets are being placed.

    Ricky Pearl is the Founder of Pointer Strategy, where he has worked with 200+ GTM teams across APAC on strategy, hiring, and implementation.

    *This session is part of the weekly GTM ANZ Community series. Join free for conversations on marketing, revenue enablement, partnerships, and AI, every Thursday.*

    Ready to Stop Gambling on Recruitment?

    No upfront fees. Practitioner-certified talent. 12 months of training. Your next hire should come with proof, not promises.

    Book a Discovery Call

    No commitment. No pitch deck. Just a conversation.