Every business has that one process.

The one that eats 4 hours every Monday. The one that requires copying data between 6 different tools. The one that only Sarah knows how to do, and Sarah’s on vacation.

Maybe it’s qualifying leads. Maybe it’s processing invoices. Maybe it’s turning customer feedback into product insights.

Whatever it is, you’ve probably thought: “There has to be a better way.”

There is. And you can build it yourself. Today.

What Is a Simple Machine?

A Simple Machine is a visual AI workflow that runs on every row of your spreadsheet.

Think of it like this:

Your data + Your logic = Your output

Each row becomes a pipeline. Each cell becomes a step. And you control exactly what happens at every stage.

No code. No developers. No waiting 6 weeks for IT to build you a custom integration.

Just drag, connect, run.

The Business Case for Custom Workflows

You Already Have Workflows

Every business runs on processes:

  • Lead comes in → qualify → route → follow up
  • Support ticket → categorize → prioritize → respond
  • Content published → repurpose → distribute → track
  • Invoice received → extract data → validate → approve

These workflows exist whether you’ve documented them or not. The question is: are they automated, or are they eating your team’s time?

The Real Cost of Manual Processes

That 4-hour Monday process? Let’s do the math:

  • 4 hours × 52 weeks = 208 hours/year
  • At $50/hour loaded cost = $10,400/year
  • For one process
  • For one person

Now multiply by every manual process across your team.

The number gets uncomfortable fast.

Why Existing Tools Don’t Solve This

Zapier/Make: Great for simple triggers, but try building multi-step AI logic. Good luck.

Custom development: 6-week timeline. $50k budget. Breaks when requirements change.

Hiring more people: Doesn’t scale. Adds communication overhead. People get bored with repetitive work.

Simple Machines: Build it Tuesday. Run it Wednesday. Modify it Thursday when requirements change.

The Building Blocks

Simple Machines are built from primitives — small, single-purpose nodes that you connect together.

AI Primitives

NodeWhat It DoesBusiness Use
GenerateCustom AI prompts with any modelDraft responses, generate content, analyze text
ClassifySort into categoriesRoute tickets, tag content, qualify leads
ExtractPull out structured dataGet entities, keywords, dates from documents
SentimentDetect emotional tonePrioritize support, monitor reviews, gauge feedback
SummarizeCondense long textDigest articles, create TLDRs, compress feedback
TranslateConvert languagesLocalize content, support global customers

Multimodal Primitives

NodeWhat It DoesBusiness Use
Describe ImageAnalyze images with AI visionGenerate alt text, catalog products, process screenshots
Generate ImageCreate images from textMarketing visuals, product mockups, social content

Logic Primitives

NodeWhat It DoesBusiness Use
If/ThenBranch based on conditionRoute based on value, handle edge cases
SwitchMulti-way branchingRoute to multiple teams, handle multiple categories
Map/LoopProcess lists item by itemHandle multiple items per row
FilterKeep matching itemsRemove noise, focus on relevant data
FallbackBackup if primary failsHandle errors gracefully, ensure output

Data Primitives

NodeWhat It DoesBusiness Use
TemplateCombine values into textBuild prompts, format output, create messages
TransformFormat and convertClean data, parse JSON, format dates
SplitBreak text into listHandle multi-value fields, parse CSVs
MergeCombine multiple inputsAggregate data, build objects
RegexPattern matchingExtract specific formats, validate data

External Primitives

NodeWhat It DoesBusiness Use
HTTP RequestCall any APIIntegrate with your stack, enrich data
Web ScrapeExtract web contentMonitor competitors, gather research

Building Your First Custom Machine

Let’s build something real: a Customer Feedback Processor that takes raw feedback, categorizes it, extracts the feature request (if any), and drafts a response.

Step 1: Define the Problem

Input: Raw customer feedback text

Output: Category, extracted feature request, draft response

Current process: Someone reads each feedback, decides category, looks for feature requests, writes a response. Takes 3-5 minutes per item.

With Simple Machine: 10 seconds per item. No human needed for 80% of cases.

Step 2: Create a New Machine

  1. Open Bwocks
  2. Go to Simple Machines in the sidebar
  3. Click “New Machine”
  4. Name it “Feedback Processor”

You’ll see a blank canvas. This is where the magic happens.

Step 3: Add Your Inputs

Drag an Input node onto the canvas. This is where feedback enters the machine.

Click the node to configure:

  • Input Name: feedback
  • Type: text
  • Required: true

This creates a parameter you’ll map to a column when using the machine.

Step 4: Build the Processing Logic

Add a Classify node:

Drag Classify onto the canvas. Connect feedback → Classify.

Configure:

  • Categories: Bug Report, Feature Request, Question, Praise, Complaint
  • Allow Multiple: false

Add an Extract node:

Drag Extract onto the canvas. Connect feedback → Extract.

Configure:

  • Extract Type: custom
  • Custom Schema: {"feature_requested": "string or null"}

Add a Template node:

Drag Template onto the canvas. This will build a prompt for the response generator.

Configure:

Write a brief, friendly response to this customer feedback.

Feedback: {{feedback}}
Category: {{category}}
Feature mentioned: {{feature}}

Keep the response under 3 sentences. Be empathetic and actionable.

Connect:

  • feedback → Template’s feedback input
  • Classify’s category output → Template’s category input
  • Extract’s extracted output → Template’s feature input

Add a Generate node:

Drag Generate onto the canvas. Connect Template’s result → Generate’s prompt.

Configure:

  • Model: Pick any model — cloud or local
  • Temperature: 0.7

Step 5: Define Outputs

Add three Output nodes:

  1. category — connected to Classify’s category output
  2. feature_request — connected to Extract’s extracted output
  3. draft_response — connected to Generate’s result output

Step 6: Test With Real Data

Click “Test” in the toolbar.

  • Enter sample feedback
  • Click “Run Test”

Watch the data flow through each node. Check the outputs.

Not quite right? Adjust a node. Test again. Iterate until it’s perfect.

Step 7: Use It

Save your machine. Now use it in your grid:

In a column formula:

=MACHINE("Feedback Processor", A2)

Bulk processing:

Import 500 feedback items. Add columns for category, feature request, and draft response. One formula each. Run.

500 items processed. Coffee’s still hot.

Real-World Machine Ideas

For Marketing Teams

Content Localizer

  • Input: English blog post
  • Process: Translate → Adapt cultural references → Optimize for local SEO
  • Output: Localized versions for 5 markets

Competitive Monitor

  • Input: Competitor URL
  • Process: Scrape → Extract key info → Compare to our positioning → Flag changes
  • Output: Competitive intelligence brief

Campaign Analyzer

  • Input: Campaign results data
  • Process: Classify performance → Extract insights → Generate recommendations
  • Output: Campaign report with action items

For Sales Teams

Lead Enricher

  • Input: Company name, contact email
  • Process: Scrape company site → Extract tech stack → Score fit → Draft personalized opener
  • Output: Enriched lead record with outreach draft

Proposal Generator

  • Input: Customer requirements, your product info
  • Process: Match features to needs → Generate value propositions → Build proposal sections
  • Output: First draft proposal

Win/Loss Analyzer

  • Input: Deal notes
  • Process: Extract reasons → Classify → Identify patterns
  • Output: Categorized insights for product and sales

For Operations Teams

Invoice Processor

  • Input: Invoice document/image
  • Process: Extract vendor, amount, line items → Validate → Route for approval
  • Output: Structured data ready for accounting

Contract Reviewer

  • Input: Contract document
  • Process: Extract key terms → Flag risky clauses → Summarize obligations
  • Output: Contract brief with risk highlights

Support Escalation Router

  • Input: Support ticket
  • Process: Sentiment analysis → Classify urgency → Check customer tier → Route
  • Output: Assigned queue with priority score

For Product Teams

Feedback Synthesizer

  • Input: Raw user feedback (support, reviews, surveys)
  • Process: Classify → Extract feature requests → Cluster similar items → Prioritize
  • Output: Organized feature request backlog

Release Notes Writer

  • Input: Git commits or Jira tickets
  • Process: Classify change type → Extract user impact → Generate friendly description
  • Output: Customer-facing release notes

Bug Triage

  • Input: Bug report
  • Process: Extract repro steps → Classify severity → Check for duplicates → Route
  • Output: Triaged bug ready for engineering

Advanced Patterns

Chaining Machines

One machine’s output can feed another machine’s input. Build small, focused machines and compose them.

Example: Raw Feedback → Feedback Processor → Response Quality Checker → Final Response

Conditional Workflows

Use If/Then and Switch to handle different scenarios:

  • If sentiment is negative AND customer tier is enterprise → escalate to senior
  • If category is billing → route to finance
  • If confidence is low → flag for human review

External Integration

Use HTTP Request to:

  • Enrich data from Clearbit, ZoomInfo, etc.
  • Post results to Slack
  • Create records in your CRM
  • Trigger webhooks in other tools

Loop Processing

Use Map/Loop when a single input contains multiple items:

  • Email with multiple questions → answer each separately
  • Document with multiple sections → process each section
  • Order with multiple line items → validate each item

Best Practices

Start Small

Your first machine shouldn’t replace your entire workflow. Pick one painful step. Automate that. Expand later.

Test With Real Data

Synthetic test data hides edge cases. Use actual customer feedback, real support tickets, genuine leads. You’ll find the weird stuff fast.

Monitor Quality

AI isn’t perfect. Build in quality checks:

  • Classify confidence scores
  • Human review queues for low-confidence results
  • Periodic audits of machine outputs

Document Your Machines

Future you (or your replacement) will thank you. Name machines clearly. Add descriptions. Note any assumptions.

Version Control

When you modify a machine, consider keeping the old version. Name them Feedback Processor v1, Feedback Processor v2. You might need to roll back.

The Bigger Picture

Simple Machines aren’t just about automation. They’re about leverage.

Every machine you build is a multiplier on your team’s capacity. That 4-hour Monday process? Now it’s 4 minutes. Your team gets 3 hours and 56 minutes back. Every week. Forever.

But here’s the real unlock: once the manual work is automated, you can do things that weren’t possible before.

  • Process 10x more feedback → Better product decisions
  • Qualify leads instantly → Faster response time → Higher conversion
  • Monitor competitors daily → Strategic advantage
  • Respond to customers faster → Better satisfaction → Lower churn

The machine does the repetitive work. Your team does the thinking.

That’s the future of work. And you can start building it today.

Getting Started

  1. Identify one painful manual process. Something that takes 30+ minutes per week.

  2. Map it out. What’s the input? What are the steps? What’s the output?

  3. Build the machine. Start simple. One input, a few processing steps, one output.

  4. Test with real data. Find the edge cases. Fix them.

  5. Run it. Watch hours disappear from your calendar.

  6. Build another one.

The hardest part is starting. But once you see your first machine run — taking work that used to take minutes and doing it in seconds — you’ll never go back.

Need Inspiration?

Check out 5 Simple Machines You Can Build in Minutes for ready-to-use examples:

  • Lead Scorer — Qualify leads automatically
  • Content Repurposer — Social content from blog posts
  • Sentiment Router — Understand customer emotion
  • SEO Meta Generator — Instant SEO metadata
  • Alt Text Generator — Accessible images at scale

Each one started as a simple idea. Now they save hours every week.

What will you build?

Get started with Bwocks →