AI Systems for Agency Owners | LachlanCB
Field tests

What I tested

Experiments I actually ran, with the numbers that came back — including the ones that returned nothing at all. Anyone can publish the wins. The failures are where the transferable information is.

06
Tests
01
Outright failures
  1. № 01Worked2026

    The tradie site engine

    Can one config file produce a complete, client-ready local-service website, without a developer touching it each time?

    Method
    Built a config-driven Next.js engine after tearing down 26 competitor sites. One config file renders a full site. Ran it end to end against fresh fixtures across five trades, then split it into two engines for the starter and growth tiers.
    Result
    It works. Plumber, electrician, painter, landscaper and handyman sites all generated from config and deployed. The handyman build ran clean with no fixes needed, which is the run that actually validated it. A later feature rebuilds a prospect's existing site from its URL for use on a cold call.
    So what
    Site production stopped being the constraint. The bottleneck moved to sales, which is the correct place for it to sit and the reason the emphasis shifted there.
    5trades generated end to end
  2. № 02Failed2026

    Cold outreach at volume to local trades

    Can volume alone open a market that doesn't know it has the problem?

    Method
    ~5,000 cold emails plus ~100 SMS to local service businesses. Warmed domains, multi-step sequencing, segment-tuned copy — the same setup that produced 60–80% open rates on other campaigns.
    Result
    Zero replies. Not a low reply rate — zero. Deliverability was fine and opens happened, so the failure was downstream of the inbox.
    So what
    The problem wasn't the message, it was the premise: this audience had no felt pain about the thing I was selling. Switched trades to phone-first and reserved email for higher-ticket buyers who already know they have the problem.
    0replies
  3. № 03WorkedJuly 2026

    One room vs. an entire outbound campaign

    Does in-person beat scaled digital outreach for early-stage pipeline?

    Method
    Attended a single afternoon networking meetup and compared the outcome against the full cold-email experiment running in parallel.
    Result
    One room of working people produced more real conversations, replies and an actual employment lead than the entire cold-email experiment combined.
    So what
    Reweighted toward in-person and warm channels for anything early-stage. Cold volume is a scaling tactic, not a starting one — it multiplies a message that already works.
    1 afternoonbeat ~5,000 emails
  4. № 04MixedJuly 2026 — still running

    The 20% Vault — selling with zero audience

    Will a digital product sell organically before any audience exists?

    Method
    Book-synthesis products listed on a third-party marketplace with no promotion, no list and no paid traffic — left to be found. Running alongside two parallel demand tests on other marketplaces and a pinning channel.
    Result
    One organic sale from a buyer who found it themselves, after months at zero. It came from marketplace discovery rather than anything I did — which is the actual finding.
    So what
    Enough signal to keep the channel, nowhere near enough to call it validated. The honest reading is 'first paid sale, n=1' — not product-market fit. Marketplace placement is beating promotion at this stage, so that's where the next effort goes. This one is still in the mix, not decided.
    n=1organic sale, zero marketing
  5. № 05MixedJuly 2026

    Rank-and-rent local site

    Can a local service site generate demand before a business exists behind it?

    Method
    Built and ranked a local trade site as a pure demand test, deliberately bypassing my own rule about validating first.
    Result
    The site ranks and the engine is reusable, but the lead path was left in demo mode — so early enquiries had nowhere to go. A build finished; a system didn't.
    So what
    Produced a reusable site engine for future locations. Also a reminder that 'shipped' and 'working end to end' are different states, and only one of them earns anything.
    livedemand test running
  6. № 06WorkedJuly 2026

    Migrating a 46-post blog off a hosted CMS

    Can you leave a CMS without losing search visibility?

    Method
    Exported 46 posts, 8 pages and 53 tags from Ghost to MDX in a Next.js app. Wrote automated checks for URL parity and content fidelity, then cut DNS over.
    Result
    Every live URL still resolves — 107 direct, 3 intentional redirects, zero broken. Content fidelity verified across all 46 posts. Two real traps: the CMS's own routing didn't match what I assumed, and the host served an expired certificate after cutover.
    So what
    Verify routing against live HTTP status codes rather than trusting documentation or memory. The plan was wrong in four places; only the checks caught it.
    110 / 110URLs preserved
The things I killed