Hold Off on Buying Robots; Break Them Down into a Spreadsheet First

I tinkered with a humanoid robot implementation assessment sheet over the weekend and stepped on quite a few pitfalls. In his G20 video address, Elon Musk said that in the next 10 years, the number of humanoid robots globally will exceed 1 billion, with each producing at least 5 times the output of a human. That's a bold statement. What really matters is where the money comes from and whether delivery can run stably. This sheet is beginner-friendly.

Day 1: Create the table. Open Excel, WPS, or Feishu Bitable, click New Blank Spreadsheet, and name the first worksheet Robot Implementation Assessment. You'll see column labels A, B, C at the top. Row 1 contains fields (column names). Enter seven columns: Scenario, Single Task, Human Baseline, Robot Target, Unit Cost, Evidence, Risk. Single Task is the specific action the robot performs, like moving a box, sweeping once, or grabbing a bowl. Human Baseline is how many times a human does this per hour and the error rate. Robot Target is how many times the robot does it per hour and if it can run continuously for eight hours.

Fill in three rows. Enter Factory Loading/Unloading for Scenario, Move a Box for Single Task, Boxes per hour for Human Baseline, Boxes per hour for Robot Target, Depreciation + Electricity + Labor Supervision for Unit Cost, leave Evidence blank for now, and enter Cardboard Size Variation for Risk. Expectation: The table has 3 rows. Pitfall: Many people start with "General Household Service," which is too broad to verify. Change it to small tasks like "Taking out trash at night" or "Picking up vegetable leaves in the kitchen."

Day 3: Collect evidence. Don't write "Impressive" in the Evidence column; write verifiable items. Click the cell and enter Demo Video, Customer Interview, On-site Record, or Unverified. If suppliers only provide PPTs, mark it as Unverified. In my testing, many booths look smooth, but changing lighting or cardboard types causes jams. Watching live demos of Unitree H1 or Haier CR3, being able to run doesn't equal stable delivery. Beginners can add dropdowns: Select the Evidence column, click Data Validation or Dropdown List, and enter the four options. You'll see a dropdown arrow. Expectation: People filling out the form won't write randomly.

This is where things often go wrong. Musk said each robot produces at least 5x humans, but many interpret this as robots being permanently 5x faster than humans. This misunderstanding mistakes short-term peaks for normalcy. As output increases, failures, charging, and maintenance also rise. There is a gap between predicting visuals and predicting the consequences of actions—that gap is engineering.

One week later: Review and score. Create a new column Score, rating 1 to 5. Criteria: 1 is concept only, 2 has a demo, 3 has real-scenario trials, 4 has stable customers, 5 has a cost model. Create another column Downtime Hours, estimating daily downtime. Click Sort, ordering by Score from high to low. If high-scoring items lack electricity, safety, or data feedback in their Risk column, downgrade them. Musk also mentioned power shortages; robots don't just plug in and run. Queuing for charging and downtime for repairs cost money.

Metric	What Musk Said	Common Beginner Misinterpretation	How I Suggest Breaking It Down
Quantity	>1 billion in 10 years	Everyone gets one immediately	Watch industry adoption ramp-up first
Output	At least 5x per unit	Always 5x	Look at continuous duration per task and failure rates
Economics	Expand 10x+	All companies earn 10x	Look at replacement costs and gross margins

Startup advice: Find a narrow scenario first, run it for three months with real data, then talk about generality. Domestic supply chains, scenarios, and data might enable faster implementation. But don't mistake production capacity for demand. If suppliers brag about scale, first ask how many real workstations they've actually deployed on.

Next step: Take a moving or cleaning task in your company or factory and fill in one row according to the seven columns. Don't ask if the robot looks human; first see if the cost can be brought down.

Don't treat "1 billion units" as the future; break one task down until it's verifiable first.