Dive Deep

Leaders stay connected to the details and operate at every level. They build mechanisms to observe performance, identify problems, and recognize what is working well. They investigate when metrics and anecdotes differ. No task is beneath them.

Why it matters

What this principle is for.

If you do not understand the details of the business, you will fail. Dive Deep is getting to the root cause. Good leaders peel layers. They ask why until they hit something they can fix so it stays fixed. When a senior person does that, the team gets more rigorous, because sloppy stories stop surviving contact with the facts.

Look for skepticism when the metric and the anecdote disagree. Intuition is a starting point. Then you test it. When the data is not sitting in a dashboard, good people go get it. They do not take a deck at face value, and they do not derail a meeting by asking for the numbers split a dozen more ways when none of them would change the decision.

Diving deep is how you learn what matters. You learn how the process, the system, and the tool actually work. You find the breaking point before it breaks. You understand a defect well enough to build a fix that scales, not a one-off fix that comes back next month. That is also how you train Are Right, A Lot. Judgment without details is a guess you got away with.

It is not micromanagement. Micromanagement is controlling the minutiae and doing the owner's job for them. Dive Deep is understanding the details, then setting up a few regular checks that keep you close to the work without living in the spreadsheet. You test the assumption, then you step back and let the owner own it. No task is beneath you. That does not mean every task is yours.

Calibration

Under, just right, over.

Score whether someone can get to the root cause and then let the owner run with it. Under looks like taking the summary deck at face value. Over looks like staying in the spreadsheet and never making a call. A story about what a product line earns, or what one unit costs, counts. Jargon does not make the answer deeper.

Situation	Under	Just Right	Over
A weekly status that is all green	Treats the green boxes as the update. Does not open a single item to see the work behind it.	Samples the two or three items that would hurt next week if they were false, traces those to the actual work, and leaves the rest once the sample holds.	Questions every line until the owner spends the next day rebuilding the status report instead of doing the work.
A teammate says they already checked	Accepts the assurance and never looks at the source they used.	Asks to see the source once, confirms the check was real, and then lets them own the next ones.	Rechecks their work in the room every time. They stop checking themselves because it will be redone anyway.
Approving a purchase	Signs because the amount looks familiar or the requester is trusted. Does not open the quote.	Reads the quote and the alternatives, asks the one or two questions that could change the decision, and then signs or sends it back.	Rebuilds the vendor comparison from scratch and holds the order. The team misses the window they needed the thing for.
A handoff between teams	Accepts the summary email and never opens the docs or code being handed over.	Reads the docs and code their team will need this week, names the gaps that would stall them, and then lets the handoff go.	Rewrites the other team's docs and code before accepting the handoff. Both teams wait on a cleanup that was not in the plan.
A launch checklist	Marks lines done from memory. Does not look at the test results or dashboards behind each item.	Verifies the few items that would hurt customers if they were false, then ships on the rest of the list.	Makes the team rerun every test and walk through every item on the list live, including ones already checked. The launch window closes while the team redoes work that was already finished.
A vendor claim	Repeats the vendor's number in the internal update. Does not compare it to our own records.	Checks the claim against a week of our own records, names any gap, and then decides whether to keep the vendor, push back, or drop them.	Starts an audit of the vendor's whole history before the next call. The renewal or the switch to a new vendor sits while the audit keeps growing.
A message asking for a quick approval	Replies yes without opening the attachment.	Opens the attachment, checks the part that cannot be undone, and answers the same day.	Holds the message overnight to reread the whole thread. The person who needed the approval misses their deadline.
A deadline that slipped	Takes the new date and does not look at what actually stalled.	Traces the slip to the real blocker, names it, and resets only what that blocker requires.	Audits the whole plan from the first task. The team spends the week explaining history instead of clearing the blocker.

Examples

What it looks like in the work.

When the metric and the story disagree

The dashboard is green. Three customers and two operators are telling a different story. The leader does not pick a side. They pull the raw events, walk the path a customer walks, and find the missing filter that made the metric look healthy. Skepticism when metric and anecdote differ is the whole principle in one move.

Look at each product line, not the category average

The category's profit margin looks fine compared with last year. A weekly review of the best-selling and worst-selling product lines shows that shipping costs wipe out the profit on every product from one supplier, and that a discount the supplier agreed to was never applied. The action is specific: fix the price or the shipping cost on those products, and get the supplier to apply the agreed discount so the next order does not lose money too. A daily review of all 1,000 products would have used the week and fixed nothing.

Audit, then step back

A leader can sit in the data and rebuild the model. They do it once, to learn where the bodies are. Then they put a short audit in place (a weekly list of what broke, a check that the numbers add up, a walk through the floor) and give the work back to the owner. Doing the owner's job every Tuesday is not Dive Deep. It is a different job, unpaid.

In the role

Individual and manager.

Individual

An individual who dives deep can explain the system they own in one pass: where the data comes from, where it breaks, who consumes it, and what happens if load goes up by a factor of four. They ask why until the cause is a sentence. They do not fake fluency, and they do not need a week to answer a question they should already know.

Manager

A manager who dives deep can operate at every level without becoming the bottleneck. They audit. They walk the floor. They know the drivers well enough to catch a bad story in a review. They do not skip the leaders in between, and they do not confuse a stack of extra reports with understanding. After they have the fact, they let the owner own the fix.

Go deeper

Questions that make the principle concrete.

Can you state the problem in one sentence, including what should be happening versus what is happening?
When the metric and the anecdote disagree, which one did you follow first, and what did you pull to reconcile them?
Are you solving a symptom, or can you name the cause specifically enough that a fix would stay fixed?
Which number in the last review did you accept without asking how it was built?
If load on your system went up by a factor of four, where does it break first?
What audit do you run so you stay connected to the details without sitting in the owner's chair?
Where is the data you are missing, who has it, and why have you not asked them yet?
Are you treating this as a problem you have already seen, and applying a familiar fix without testing the assumption?
How does this problem show up for the customer, for the team on call, and for the teams upstream and downstream?
What would you stop measuring so you have time to act on the few drivers that move the business?

Sources and essays on this principle.

Dive Deep != Micromanaging
Understand the details, then let the owner run. That is not living in their shorts.
kindel.com
Jeff Bezos's 2016 Letter to Amazon Shareholders
Resist proxies. The process is not the thing, and the survey is not the customer.
aboutamazon.com

Principles that sit next to this one.

Principles lean on each other, and some pull against each other on purpose (the tension is intentional). Here is how Dive Deep connects to the rest of the set.