Skip to content

Notes on: ZapConnect 2024

Notes on ZapConnect 2024. This post has two functions: notes for future me, and knowledge sharing.

From the website:

ZapConnect brings together industry leaders, automation experts, and users like you for a powerful half-day of growth and connection. Dive into tailored sessions that uplevel your work and unlock new potential with unmissable insights on scalable automation.

Keynote by Zapier leadership

  • Prediction by Wade Foster: There will be a billion dollar business created by a team of less than 10 using Zapier and AI.
  • Remote (the remote hiring service) uses 580+ zaps! Scaling people as well as business. The IT team (three people) scaled themselves with Zapier. Automation may be saving headcount by a lot (might need to be twice the size without it), and thus saving the company.
  • Arden Insurance: automated sending invoice notices, then moved on to more. 34,000 hours a year saved.
  • My note: Zapier has changed a lot since I last looked. Interfaces and tables . . . and canvas to map multiple workflows. That's very nice.
  • AI usage is high.
  • My note: Zapier Central looks super-powerful: an AI assistant that can scrape web pages (among other things). Pair with chatbot.
  • And the big announcement: Zapier Functions. Write automation logic in Python (!!!).
    • Looks like you can 'call' integrations (and zaps?) from code? And the function can be called from triggers.

My note: thinking as an ex-n8n-er, can n8n compete? Yes, I believe so: self-hosting is the key differentiator, n8n need to lean into that. And n8n pricing is very competitive, especially for more complex workflows. And who are Zapier competing with at this point? Maybe Microsoft Power Platform?

Automation architecture: Master your business logic for automation success by Alistair Wilson

A speedy run-through of a substantial process. There was definitely value here, but it felt rushed - trying to pack a 30min talk into a 15min timeslot.

Talk outline:

  • CEO of Flowmondo: UK automation agency.
  • 'Automation is the fastest way to scale a mess'
  • Key definitions:
    • Business logic: the rules by which your business produces its output.
    • Data model
    • Codification: describing these well enough that a stranger could reliably execute those processes.
  • The human factor: the way work happens is complex to describe. Reality often deviates from documentation. Humans can infer, guess, improvise, make judgements . . .
  • Use a DPD: 'done perfectly demonstration'
    • Observe, atomize (break down the demo), formalise (make repeatable, train people)
  • Avoid automating immature/unstable processes: if you must, bake in safety and/or add human in the loop.

Igniting innovation: How to use AI to create, innovate, and scale by Wade Foster and Dharmesh Shah

Lots of nice stuff in this talk in terms of high-level industry opinions. However the conversation felt a little disjointed. This could have been a series of Bluesky threads.

Talk outline:

  • Dharmesh's process on deciding what to build:
    • Solosoft: solo software, just for him. No users, no worrying about testing etc.
    • Picks projects that help him learn what he wants to learn.
    • Then eventually some things become commercial. Most things relate to HubSpot to some extent. Looks for early signals: gets a rough product out the door, then looks for how much curiosity it garners and whether people try it. If it gets curiosity, then build it out to something polished. Reach out to early users to encourage them back once things improve.
  • Automation value: how often does the thing get done, how much does it cost, and how much does it take to automate it?
  • Nike's data-driven marketing woes: how do you balance data-driven and intuition? HubSpot loves data, but humans make decisions. You can't just say "the data said this so that's what we're going to do". Instinct (from experience) is the deciding factor in major success. The right answers aren't hiding in the data - but the right questions are.
  • AI use cases for businesses today:
    • First draft writing
  • What matters in content is the value to the customer, not who/what wrote it.
  • Before automating, ask: is what I'm doing valuable? If not, make it valuable before automating. Otherwise you're just producing more efficient junk.
  • Dharmesh hot take: companies that are 'just' thin wrappers around AI models are not automatically low-value. If you're solving a big enough problem, you'll make money.
  • How to get rewards of AI while mitigating risks: you can't wait until it's safe to start learning about AI. Find ways to experiment (sandbox, low risk use cases etc.). You can wait to deploy, but don't wait to learn.
  • Where's a good place to start with AI: engineering is where there's the clearest and most meaningful impact (measurable increase in productivity). And also in customer service: from triaging tickets to trying to actually auto-respond.
  • How has HubSpot survived/been so resilient long term: the sense of a moral obligation to help customers, which means HubSpot has to be dynamic, and keep moving with tech.
  • As a founder: keep your hands on the things you enjoy. Don't let yourself be completely pulled away from 'your' thing.

Documentation decoded: Squash automation issues before they begin by Philip Lakin

The best talk I saw today. Well-structured, at a sensible pace.

Talk outline:

  • How no-code gets complex:
    • Easy to start
    • Grows complex quickly, especially with multiple builders
    • Bus factor
    • Hidden errors
    • Cascading issues
  • Life before docs: why did we build this, what does it do, what does it impact . . .
  • Builder vs process participant: process participants use the workflow, but didn't build it. They need to know how to use the system. Builders create the workflows. Their docs need to explain how the workflow fits together, to help with handoff and maintenance. You need docs for both, but this talk focuses on the builder side (he recommends tango.us for participant side docs).
  • Four docs formats:
    • Wikis: explain why decisions were made, documenting reasoning and how Zaps support business objectives. Tip: use templates.
    • Dependency tables: visualize triggers, actions, and components of workflows. Track triggers and dependencies across workflows.
    • Flowcharts: visually map Zaps for better planning and communication. One possible tool: Zapier Canvas.
    • Automatic changelogs: automatically track changes to Zaps and keep docs up to date.
  • Case study - AI changelog: Zapier have used Coda and Slack. Changes tracked in Coda, and notifications sent in Slack.
    • Trigger: a Slack reactji from a team member
    • System generates summary of change using OpenAI
    • Log to Coda
    • A Slack notification to alert about the changelog update.
  • Life after docs:
    • Debugging is easier: changes are visible through the changelog, the dependency table helps you track sideeffects, wiki helps understand background and who to talk to.
  • Takeaways:
    • Docs provide a complete view for faster troubleshooting.
    • Next steps: automated changelog, work towards a linked docs system, and avoid writing cryptic shorthand notes!