# Training — the full journey

**A guide to building agents and scaling your marketing agency with conversational agents for all your clients.**

The videos are a contextual guide to everything we've been building in skills, so you can not only use ours but build your own and run all your agents by talking to Claude (or ChatGPT). If you get stuck on something, the whole guide is right here.

I recommend watching them anyway, even though you have the wiki to read: the videos give you the knowledge and the context. They're short, and it's best to do them in order, 1 through 6.

---

## Contents

1. **Welcome & connect**
2. **Create, test & deploy your agent**
3. **Continuous improvement: the 3 buckets**
4. **Data & decisions**
5. **Expansion**
6. **Managing multiple agents**

<div class="vcourse" data-carousel>
<div class="vc-viewport">
<div class="vc-track">

<section class="vc-slide">
<div class="vstart soon" data-thumb="assets/thumb-en-01.png">
  <span class="vp-eyebrow">Video 1 · Welcome &amp; connect</span>
  <span class="vs-soon">Coming soon</span>
</div>
</section>

<section class="vc-slide">
<div class="vstart soon" data-thumb="assets/thumb-en-02.png">
  <span class="vp-eyebrow">Video 2 · Create, test &amp; deploy your agent</span>
  <span class="vs-soon">Coming soon</span>
</div>
<aside class="callout"><strong>Covers:</strong> <strong>create</strong> = an interview with Claude (offer, who, ticket, book-a-call vs. close-in-DM, archetype, clone your voice from real chats) · <strong>test</strong> before exposing it (sample conversations, edge cases, quick QA) · <strong>deploy</strong> to your channels (Instagram, WhatsApp, web/SDK) · the importance of <strong>integrations and tracking</strong> (connect the sources well, UTM/attribution, and custom data to measure from day 1) · and the <strong>first-days</strong> habit: read the real conversations every day and make small adjustments.</aside>

<div class="trace">
  <span class="t-tag">The most important thing at launch</span>
  <h4>Connect your conversions for full traceability</h4>
  <p><strong>Every Ninjō agent sends links with an invisible UTM term</strong>, and that UTM is a <strong>unique contact ID</strong> for each person. If you connect <strong>Calendly, GoHighLevel</strong>, or your sales calendar / CRM under <strong>Integrations</strong> in Ninjō, those systems pick up the UTM automatically — so the agent knows <strong>who completed the booking (or the sale) and who didn't</strong>.</p>
  <div class="t-flow"><span>Agent sends link <b>+ UTM</b></span> → <span>Lead books in Calendly/GHL</span> → <span>The UTM <b>returns to Ninjō</b></span> → <span>Funnel traced end to end</span></div>
  <p>That info comes back to Ninjō and lets you see <strong>each person's full path</strong>: how many messages they exchanged, what content they came from, all the way to becoming a booking or a sale. On top of that, Ninjō saves those conversions in its memory as <strong>best practices</strong> → the flow improves on its own.</p>
  <p>Calendly and GoHighLevel are already integrated (connect and the contact is created automatically). Got a <strong>custom CRM</strong>? Make sure the <strong>UTM term</strong> tag can travel to your CRM and back to Ninjō. You configure all of this with Claude.</p>
</div>

<div class="ckl">
<span class="ckl-lbl">Launch checklist</span>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Agent tested and approved</b><small>It passed the sample conversations and edge cases before real people see it.</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Conversions connected (Calendly / GHL / CRM)</b><small>Under Integrations, so the UTM term travels and you get full funnel traceability.</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Follow-ups enabled</b><small>Instagram: a 24 h window for automatic follow-ups, configurable from 30 min up to 23 h 59 min. WhatsApp: the automatic window with the same agent is also 24 h (beyond that, approved templates). Ask Claude for it.</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Reply delay</b><small>Set a minimum so it doesn't feel so bot-like: 5–20 seconds recommended (you can go up to 5 minutes).</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Message splitting</b><small>Ask it to split the reply into 2–3 bubbles so it feels more human (between 1 and 5 text blocks).</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>You know how to pause the agent</b><small>In a specific conversation: ask Claude or go to Messages in Ninjō and turn it off there. When a human steps in: ask for it and it'll ask whether the pause is temporary or how long you want it to last.</small></span></div>
</div>
</section>

<section class="vc-slide">
<div class="vstart soon" data-thumb="assets/thumb-en-03.png">
  <span class="vp-eyebrow">Video 3 · Continuous improvement: the 3 buckets</span>
  <span class="vs-soon">Coming soon</span>
</div>

<aside class="callout"><strong>Goal:</strong> sort QA into three buckets instead of putting out random fires.<br><br><strong>Covers:</strong> why it pays to <strong>classify</strong> each finding before acting, and the three buckets:
<ul>
<li><strong>Fixes</strong> — broken behaviors (repeats itself, sends the link too early, interrogates, makes things up, doesn't sound like you). Targeted fix, one hypothesis at a time, with rollback.</li>
<li><strong>Performance</strong> — already works; now optimize conversion (a better objection reply, link timing, the keyword that converts best).</li>
<li><strong>Scaling</strong> — the agent is ready for more volume or channels; prepare it to grow.</li>
</ul>
Plus the <strong>rhythm</strong> (periodic review → tag each finding into its bucket → attack by priority) and the golden rule: one change at a time, measure, and roll back if it got worse.</aside>

<p>The goal with any agent is to get two things right: <strong>no technical errors</strong> (Fixes) and <strong>good performance</strong> (a good booking and closing rate). What for? So that, once those two are solid, we can <strong>scale the business</strong> — scaling the agent's volume of conversations (with ads or more organic traffic) or scaling use cases (replicate the agent for another client, or bring it to WhatsApp).</p>

<p>Leaning on Claude, the job is: ensure the <strong>fewest technical errors</strong> possible in the agent's configuration. We already did the back-end work to keep them minimal; still, when one shows up, ask Claude for help or to audit it. And on <strong>performance</strong>, keep testing which flow works best.</p>

<p>Once those two are solid, you <strong>start scaling</strong>. I always recommend scaling the <strong>agent's traffic first</strong> to see how far it goes (ads, organic, or finding other ways to bring it more traffic). <strong>Then</strong> you scale <strong>cross-channel</strong> or <strong>cross-client</strong>.</p>

<p><a class="dl-btn" href="https://ninjo-3-buckets-en.pages.dev" target="_blank" rel="noopener"><svg viewBox="0 0 24 24"><path d="M4 4h7v2H6v12h12v-5h2v7H4V4zm9 0h7v7h-2V7.4l-8.3 8.3-1.4-1.4L15.6 6H13V4z"/></svg> Open the 3-buckets presentation</a></p>

<div class="cmd">
<span class="cmd-lbl">Example of what you'd ask (I explain it in the video)</span>
<p>"Analyze my agent's conversations from the last 7 days. Comparing them with the prompt, tell me if it made mistakes at any funnel stage, how the agent's performance looks, how many leads booked and how qualified they were, and what else we could implement."</p>
</div>
</section>

<section class="vc-slide">
<div class="vstart soon" data-thumb="assets/thumb-en-04.png">
  <span class="vp-eyebrow">Video 4 · Data &amp; decisions</span>
  <span class="vs-soon">Coming soon</span>
</div>
<aside class="callout"><strong>Covers:</strong> what to watch (the <strong>funnel</strong>: conversations → qualified → bookings → sales · conversion <strong>by keyword</strong> · % booking · your best and worst conversations) · using <strong>custom data and insights</strong> to find patterns (objections, country, intent) · the definition of a <strong>good lead</strong> (booked + the client qualified it, not the internal lead score) · that the <strong>winning keyword</strong> is the one with the best conversion, not the most volume · and how to go from <strong>data to decision</strong> with statistical honesty.</aside>

<p><a class="dl-btn" href="https://ninjo-skills-reportes-en.pages.dev" target="_blank" rel="noopener"><svg viewBox="0 0 24 24"><path d="M4 4h7v2H6v12h12v-5h2v7H4V4zm9 0h7v7h-2V7.4l-8.3 8.3-1.4-1.4L15.6 6H13V4z"/></svg> See all the reports & data-analysis skills</a></p>
</section>

<section class="vc-slide">
<div class="vstart soon" data-thumb="assets/thumb-en-05.png">
  <span class="vp-eyebrow">Video 5 · Expansion: new agents &amp; features</span>
  <span class="vs-soon">Coming soon</span>
</div>
<aside class="callout"><strong>Covers:</strong> <strong>when</strong> to expand (only once the first one performs and is measured) · <strong>new agents</strong> (another offer, another channel, another funnel stage — for example WhatsApp re-engagement) · <strong>new features</strong> (advanced triggers, follow-ups, alerts, image generation, voice notes) · and how to <strong>reuse</strong> what you learned (clone voice, templates, and archetypes from the first) so every new agent is born already measured and connected.</aside>

<p><a class="dl-btn" href="https://ninjo-expansion-en.pages.dev" target="_blank" rel="noopener"><svg viewBox="0 0 24 24"><path d="M4 4h7v2H6v12h12v-5h2v7H4V4zm9 0h7v7h-2V7.4l-8.3 8.3-1.4-1.4L15.6 6H13V4z"/></svg> See the 4 channels and how we expanded</a>
<a class="dl-btn" href="https://ninjo-expansion-ideas-en.pages.dev" target="_blank" rel="noopener"><svg viewBox="0 0 24 24"><path d="M4 4h7v2H6v12h12v-5h2v7H4V4zm9 0h7v7h-2V7.4l-8.3 8.3-1.4-1.4L15.6 6H13V4z"/></svg> Expansion ideas + WhatsApp guide</a></p>
</section>

<section class="vc-slide">
<div class="vstart soon" data-thumb="assets/thumb-en-06.png">
  <span class="vp-eyebrow">Video 6 · Managing multiple agents</span>
  <span class="vs-soon">Coming soon</span>
</div>
<aside class="callout"><strong>Covers:</strong> <strong>audits</strong> (periodic review of each agent, batch QA) · <strong>automated reports</strong> (weekly per agent, scheduled, nothing built by hand) · <strong>custom dashboards</strong> (per-agent performance, funnels, live) · <strong>win alerts</strong> (find out about sales and bookings without looking) · and how to ask for it all ("audit all my agents", "build and schedule the reports", "make me a dashboard of X"). This is exactly how we operate at Ninjō.</aside>

<div class="ckl">
<span class="ckl-lbl">Best practices for managing multiple clients</span>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Daily audit</b><small>Once a day, sweep the whole portfolio: have Claude review every agent, flag what broke, and prioritize it. You catch a problem the same day, not at month-end.<br><em>Ask:</em> "Audit all my agents and send me today's flags, sorted by urgency."</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Automated reports</b><small>One report per client, scheduled, nothing built by hand. Leave it running and it arrives on its own.<br><em>Ask:</em> "Build the weekly report for each client and schedule it for Monday mornings."</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Build your own skills</b><small>A <b>skill</b> is a reusable capability: an instruction set Claude follows the same way every time. If you repeat a task (a report type, an onboarding), turn it into a skill and stop re-explaining it.<br><em>Ask:</em> "Turn the flow we just did into a reusable skill I can invoke whenever I want."</small></span></div>
<div class="ck"><span class="ck-mark">✓</span><span class="ck-b"><b>Build your own commands</b><small>A <b>command</b> (slash command, e.g. <code>/cortex-report</code>) is the shortcut that fires that task with one word. Great for what you run every day.<br><em>Ask:</em> "Create a /daily command that audits all my agents and posts the flags to my Slack."</small></span></div>
</div>

<div class="trace">
  <span class="t-tag">What the Ninjō team runs for you</span>
  <h4>Customer support operators</h4>
  <p>Beyond your agents, at Ninjō we run <strong>operators</strong>: specialized agents the team operates for you on background tasks — for example <strong>customer support</strong> (they answer questions, onboarding, and post-sale for your clients, and escalate to a human when needed).</p>
  <p>Want to implement your own? <strong>Contact the Ninjō team</strong> and we'll set it up for your operation.</p>
</div>
</section>

</div>
</div>
</div>
