Te Whare Matihiko · Learn & Build

AI & automation — learn & build

18 June 2026 · weekly upskilling · research-verified

You don't need to be technical to work powerfully with AI. This week: three concepts that will change how you see your tools — and one hands-on build you can finish before Friday.

Concept 1 · Prompt engineering is a writing skill

What it is A prompt is the instruction you type to an AI. Prompt engineering is simply the practice of writing those instructions well — clearly, specifically, and with enough context so the AI gives you a genuinely useful answer. That's it. No code, no technical background needed.

Jargon decoded: "Prompt" = your message to the AI. "Engineering" sounds technical but just means crafting it thoughtfully — the same way you'd brief a junior copywriter, not program a computer. Skai.io (a marketing intelligence firm) put it well in 2025: "Prompt engineering isn't some new technical specialty."

Why it matters

Research from 2025 found that 78% of AI project failures trace back to poor human-AI communication — not the AI being incapable, but the instructions being vague (Skai.io, 2025). Meanwhile, 62% of businesses have no structured training on how to write good prompts. That's an opening: marketers who learn to write clearly and specifically get dramatically better results from the exact same tools their competitors are using.

A tiny concrete example

Vague prompt (underwhelming result):
"Write an email for my client."

Well-engineered prompt (usable first draft):
"Write a 3-sentence re-engagement email for a Shopify wellness brand. The customer hasn't purchased in 90 days. Tone: warm and genuine, not corporate or pushy. End with a soft call-to-action to browse the new arrivals. Do not offer a discount."

The second prompt gives the AI a format (3 sentences), a context (wellness brand, 90-day gap), a tone brief, and a constraint (no discount). You'll get something usable on the first try instead of the fifth.

Think of it as briefing a smart colleague who knows nothing about your client: the more specific you are, the better the work. Your domain knowledge is the input; the AI is the fast drafter.

Concept 2 · Zapier + MCP = your AI switchboard

What it is You already know Zapier connects apps (when X happens in one app, do Y in another). Now there's a new layer: MCP — a standard that lets AI tools like Claude plug directly into Zapier and reach your entire app stack through plain-language conversation.

Jargon decoded — MCP (Model Context Protocol): Think of it as a universal socket strip that lets AI tools plug into your apps — Shopify, Klaviyo, Gmail, ClickUp, Google Docs — and take actions inside them, not just read them. Anthropic (Claude's maker) published this open standard in November 2024; it's now been adopted across the industry by OpenAI, Google, and others as well (Digiday, 2025). Learning it now is like learning email in 1999 — it's going to be everywhere.

Why it matters

Zapier launched its own MCP server in March 2025, giving Claude access to 30,000+ actions across 7,000+ apps — all without writing code (Digital Applied, 2026). You are already using Zapier. This means Claude can now, in a single workflow, draft a Klaviyo email and schedule it, create a ClickUp task, and update a Google Sheet — because Zapier acts as the bridge between them.

Zapier vs n8n — which one for you?

Zapier — best for non-technical teams. 7,000+ pre-built integrations, a visual point-and-click builder, and MCP support for Claude. Multiple independent 2025-2026 comparisons confirm this is the tool for marketers who don't code.

n8n — more powerful AI capabilities (70+ AI nodes, local model support, LangChain integration) but genuinely requires developer-adjacent skills. Think of it as building your own dishwasher vs. using one that comes ready. Great to know exists; not where to start.

A tiny concrete example

A new Shopify order comes in → Zapier detects it → passes the order details to Claude → Claude writes a personalised thank-you in the brand's voice → Zapier sends it via Klaviyo automatically. No manual writing. No code. Set it up once in Zapier's visual builder.

Concept 3 · Klaviyo's AI sees what's coming

What it is Klaviyo has a feature called predictive analytics — it uses AI to make educated guesses about each customer's future behaviour, based on their purchase history. It predicts when someone is likely to buy next, how much they'll probably spend, and whether they're at risk of never coming back.

Jargon decoded — "predictive analytics" and "churn": Predictive analytics = using past patterns to guess what happens next. Instead of only knowing "this customer bought 3 months ago," Klaviyo also shows you "this customer is likely to buy again in 2 weeks" or "this customer is at high risk of leaving." Churn = when a customer quietly stops buying and probably won't return. Predictive analytics lets you act before that happens.

Why it matters

Most email marketing reacts to what has already happened. Predictive analytics lets you act on what's about to happen. Klaviyo's own Academy course confirms you can build a segment of "customers likely to lapse in the next 30 days" and send them a gentle re-engagement flow before they've actually lapsed — which converts significantly better than a "we miss you" email sent after the fact (Klaviyo Academy, 2025).

The five predictions Klaviyo generates per customer profile: spending potential, expected next order date, average order interval, predicted lifetime value, and churn risk. All usable as segment conditions directly in Klaviyo's builder — no extra tools required.

Important caveat: Klaviyo's predictions require at least 1,000 customers and 12 months of order history to be reliable. For smaller or newer clients, the predictions exist but carry lower confidence — weight them accordingly when advising on strategy.

A tiny concrete example

In Klaviyo: Segments → Create Segment. Add the condition "Predicted next order date is more than 60 days from now" AND "Last order was within the past 6 months." That's your at-risk segment — customers who used to buy but whose next purchase is predicted to be far away. Build an automated flow just for them. Send it before the gap widens into goodbye.

Practice build this week

Build your first AI writing assistant in Zapier — a Zap that drafts a personalised customer thank-you using Claude, ready for you to review.

What you'll build: When a new Shopify order is placed → Claude writes a short, personalised thank-you message in your client's brand voice → it lands in a Google Doc (your review inbox) before anything goes to a customer.

Why start here? It's low-risk (nothing reaches customers without your eyes on it), uses tools you already have, and teaches all three concepts at once: you'll write your first real prompt, see Zapier bridge your apps, and watch AI draft copy you can improve. Once this is working, every other automation follows the same pattern.

Log into Zapier at zapier.com. Click Create → New Zap. You'll see a two-step canvas: a Trigger on the left, an Action on the right.
Set your Trigger: Search for "Shopify" and select it. Choose the trigger event "New Order." Connect the Shopify store and click Test Trigger — Zapier will pull in a real recent order so you can see what data is available: customer name, product name, order total, and more.
Add an AI step: Click the + to add an Action. Search for "Claude AI by Anthropic" (or "AI by Zapier" if Claude isn't visible). Choose "Generate Text" or "Send Message."
Write your prompt — this is Concept 1 in action. In the prompt field, write something like:

"Write a warm, 3-sentence thank-you message for a customer named [first name] who just bought [product name]. The brand is a New Zealand wellness company. Tone: genuine and human, not corporate. Do not mention a discount or offer."
Use Zapier's purple field-picker icon to insert real Shopify data — the customer's first name, the product they ordered — as dynamic tags. The AI will see the actual values when it runs.
Add a final Action: Click + again. Choose "Google Docs" → "Append Text to Document." Pick a Google Doc you'll use as a review inbox (create one now if needed: "AI draft messages"). Map the Claude output to the text field. Each new order will add a fresh draft to your doc.
Test and publish: Click Test Zap — Zapier will run the full sequence using the real order from Step 2. You'll see Claude's draft appear in your Google Doc within seconds. Review it: does it sound right? If yes, click Publish.
Iterate the prompt: The next time a draft doesn't quite hit the brand voice, go back to Step 4 and refine it. Add a line like: "Brand voice example: 'Thank you for choosing us — your order is on its way and we're so glad it found you.'" Each tweak builds your prompting instinct.

Time to build: 30–45 minutes. Cost: within Zapier's free tier for low order volumes. If the store processes high order volume, check zapier.com/pricing for task limits.

Worth knowing this week

MCP (the AI-to-app standard) was published by Anthropic in November 2024 and has since been adopted by OpenAI, Google, and most major AI platforms — learning Zapier's MCP implementation gives you skills that will transfer across the whole industry.
n8n 2.0 (released December 2025) changed how workflows are deployed: Save no longer means Publish. If any client runs n8n in production, flag this — teams used to the old behaviour may think changes are live when they aren't.
Klaviyo's Segments AI feature (natural-language segment builder) is still maturing — it's a useful starting point, but always verify the logic it generates before running a campaign, as it can be overly literal.
The Klaviyo Academy course Use Predictive Analytics to Retain More Customers is free and genuinely practical — worth 90 minutes if you manage any Shopify/Klaviyo client accounts.

What to do with this

Do the Practice Build this week — even a rough first Zap teaches more than reading about it. The pattern you learn applies to every automation after this.
Start a prompt library — a simple Google Doc where you save prompts that worked well. Over time this becomes one of your most valuable professional assets.
Before your next client review, check whether their Klaviyo account has 1,000+ customers and 12 months of data. If yes, the predictive churn-risk segment is a concrete thing you can propose and build.
Explore Zapier's MCP integration (search "Zapier MCP" in their help docs) once the Practice Build is working — this is the bridge to having Claude help you manage client workflows by talking to all your apps at once.