The call at the worst time
It’s a Thursday afternoon. The website is down, and the contact form with it. A prospect writes in to say the page won’t load. The ads are still running, so every click lands on an error page.
Everyone knows who to call. She set up the hosting years ago, she knows which account it lives in, and she’s the only one with the login. She’s also on a plane somewhere over the middle of the country, with her phone in airplane mode.
So the team waits. Somebody tries a password from an old email. Somebody else calls the hosting company and learns it can’t talk to anyone who isn’t on the account. Somebody pauses the ads, which takes longer than it should because the person who usually does that is out too.
She lands, sees the messages, finds the airport Wi-Fi and has the site back before her bag comes off the carousel. The next morning, people thank her in the team chat. Someone calls her a lifesaver. Everyone means it.
Nobody asks the other question: why did the site need her?
That question is what this guide is about. The person who fixed it did excellent work, and she deserves the thanks. The trouble is that the business had been arranged, by accident and over years, so that excellent work by one specific person was the only way back. Her skill belongs in the design of the system, where it protects the business on the days she’s on a plane, and not only in the rescue.
The short answer
Heroics are a symptom of a system nobody designed. Count the rescues. For each thing your business can’t go a week without, name an owner, write the first steps down, automate the routine part, watch it so you hear about a failure before a customer does, and rehearse the recovery. Automation doesn’t remove accountability. Someone still owns what happens when a system breaks, and the design should say who before anyone has to be a hero.
Someone still owns what happens when a system breaks.
Why do businesses come to depend on heroes?
Nobody sets out to build a business that depends on one person. It happens through a loop, and every turn of the loop feels like a win.
- A capable person handles a risk others never see
- The problem disappears
- People learn it just gets handled
- Extraordinary effort becomes the ordinary expectation
It starts when a capable person handles a risk the rest of the team doesn’t see. A certificate is about to expire, a vendor changed a setting, a payment didn’t go through. They fix it before anyone notices. The problem disappears, and because nobody saw it, nobody learns it existed. What people learn instead is that this kind of thing just gets handled. Over time, extraordinary effort becomes the ordinary expectation, and the person who supplies it quietly becomes part of the infrastructure.
Each step is reasonable. The fix was fast, the team stayed focused, the customer never knew. What the loop never produces is a record: no owner of the outcome, no written steps, no alert that would have fired without the hero, no rehearsal of the recovery.
Why does it feel like strength?
Rescues are visible and prevention isn’t. A rescue has a story, a time stamp and a person to thank. The work that would have made the rescue unnecessary has no story at all, because its whole result is that nothing happened. A business rewards what it can see, so it rewards the rescue and rarely funds the prevention.
What does a hero cost a business?
Three costs, and none of them shows up on a report until the day it does.
A single point of failure
When the hero is on a plane, out sick or gone, the thing they hold stops. The business has one path back, and it runs through one person.
A bottleneck
If one person has to approve, fix or touch every instance of something, the work moves at their speed. Growth makes it worse, because every new client joins the same queue.
Knowledge that leaves
What the hero knows lives in their head. When they change jobs, the knowledge walks out with them, and the next person learns it during the next incident.
None of this is the hero’s fault. They did what the system asked of them. The fix is to change what the system asks.
How do you know your business has a hero problem?
You don’t need a consultant to find out. Five questions will do it, and most people can answer them about their own business in a minute. Tap each one that’s true for you.
-
If yes, it needs Owner: That account has a holder, not an owner. Add a second person with access today.
-
If yes, it needs Runbook: Write its first steps down and have someone else follow them once.
-
If yes, it needs Recovery: Pick one backup and restore it on purpose this month.
-
If yes, it needs Runbook: The next fix ends with its steps in writing, where the next person will look.
-
If yes, it needs Monitoring: End each one with a single change: a check, a step, or an owner.
If you tapped one, you have a hero in at least one place. If you tapped several, you can probably already say their name. The meter is a reading aid, not a score. No number of signs makes a business safe, and one sign on the wrong system is enough to stop it.
Each sign also points at a fix. A login only one person holds needs an owner and a second person with access, which is the cheapest change you can make this week. A task that waits for one person needs its first steps written down. A backup nobody has restored needs a rehearsal. An incident that ends with thanks and nothing else needs a change: a check that would have caught it, a step that would have prevented it, or an owner who will make sure it doesn’t come back.
The signs are also hard to see from the inside, for the reason the loop explains. The person who could answer them best is often the hero, and the hero is the one person who never experiences the problem, because they’re the one solving it. Ask the question of the people who would be stuck if the hero were gone.
The five replacements for a hero
A hero does five jobs at once, and nobody writes any of them down. They know who’s responsible. They know the steps. They do the routine parts by hand. They notice when something breaks. They know how to bring it back. The fix splits those five jobs apart and gives each one to the system.
The five replacements for a hero are Owner, Runbook, Automation, Monitoring and Recovery: one named owner for each outcome, its first steps written down, the routine part automated, a check that reports a failure before a customer does, and a recovery that has been rehearsed.
Owner
One named person or role is accountable for the outcome, not only for the task, and has a named second.
Done looks like: You can say who owns it in one breath, and that person has a second who can step in.
In practice: Enrollmentum’s team page names one accountable lead, with every specialist listed under that name.
Runbook
The first steps are written down where the next person will look, so a rescue becomes a procedure.
Done looks like: Someone who has never done the task follows the steps and gets the same result.
In practice: Enrollmentum’s operations run with AI specialists working under written rules, so the rules carry the work instead of one person’s memory.
Automation
The routine part runs without a person, and changes ship from code that can rebuild the system.
Done looks like: The routine part runs on a schedule, alerts on a failure, and a change can be made again from the record.
In practice: Enrollmentum’s security page: “Changes to the lead systems are deployed from code after automated tests.”
Monitoring
A check reports a failure to more than one person before a customer reports it.
Done looks like: The alert reaches the owner and the second, and someone has acted on a real one.
In practice: Every site we manage, build or inherit gets an uptime check, kept in step with the list of sites so a new one cannot be missed.
Recovery
A backup counts once a restore from it has been tested, on purpose, and timed.
Done looks like: You have restored from it, you know how long it took, and the steps are in the runbook.
In practice: Two servers with no backup were found on 2026-09-28 and closed with off-site copies and snapshots.
- Owner. One named person or role is accountable for the outcome, not only for the task, and has a named second.
- Runbook. The first steps are written down where the next person will look, so a rescue becomes a procedure.
- Automation. The routine part runs without a person, and changes ship from code that can rebuild the system.
- Monitoring. A check reports a failure to more than one person before a customer reports it.
- Recovery. A backup counts once a restore from it has been tested, on purpose, and timed.
Owner
The owner is accountable for the outcome, not only for doing the task. That difference decides what happens on a bad day. A task owner does their step and moves on. An outcome owner asks why the site was down, whether it will happen again, and who else needs access. Done looks like this: you can name the owner of each critical item in one breath, and each owner has a named second who can step in with real access.
At Enrollmentum, the clinical trial recruitment service Illuminare runs, the team page puts this in writing. It names one accountable lead, and every specialist sits under that name: “Nick Bostic leads Enrollmentum and is accountable for every study” (team page, read November 15, 2026). Naming the owner is the step that makes the other four possible, because each of them needs someone to keep it true.
Runbook
A runbook is the first steps, written down where the next person will look. It turns a rescue into a procedure. It doesn’t have to be long. Five steps that get a second person to the right account, the right screen and the right phone number beat a perfect document nobody can find. Done looks like this: someone who has never done the task follows the steps and gets the same result.
Enrollmentum’s operations run with AI specialists working under written rules, so the rules carry the work instead of one person’s memory. The team page says they “work in defined workflows, with testing, documented controls and human oversight.” When the rules are written, the same check runs the same way on a Tuesday afternoon and at two in the morning.
Automation
Automation takes the routine part off a person’s plate, and it makes changes repeatable. Google’s Site Reliability Engineering book has a useful test for what to automate. It calls the target toil: work “tied to running a production service that tends to be manual, repetitive, automatable, tactical, devoid of enduring value, and that scales linearly as a service grows.” Renewing, copying, checking and restarting are toil. Deciding what to do when the copy looks wrong isn’t.
Enrollmentum’s security page states the rule for its lead systems: “Changes to the lead systems are deployed from code after automated tests.” We hold our servers to the same rule. A change to a server goes through code that can rebuild it, so the server can be put back the way it was by running the code again, and nobody has to remember what they typed. Done looks like this: the routine part runs on a schedule, it alerts someone when it breaks, and any change can be made again from the record.
Monitoring
Monitoring means you hear about a failure before a customer does. The opening scene had none: the first alert was a prospect. Done looks like this: a check watches each critical item, the alert reaches the owner and the second, and someone has acted on a real one.
Every site we manage, build or inherit gets an uptime check, and the list of checks is kept in step with the list of sites, so a new site can’t be missed. Earlier in my career, at a digital agency, the client sites were watched by a purpose-built monitor that probed every site every five minutes, and the sites in active service held 99.999% uptime. The number came from the monitor, not from anyone’s memory of a good month.
Recovery
A backup counts once a restore from it has been tested. Until then it’s a hope with a schedule. NIST’s contingency planning guide puts the reason plainly: “Testing validates recovery capabilities.” CISA’s guidance for small businesses is blunter: “It’s not enough to schedule all important systems to have a regular backup. It’s critical to regularly test partial and full restores.” Done looks like this: you’ve restored from the backup on purpose, you know how long it took, and the steps are in the runbook.
Case: the backup gap
- Two of our servers are found with no backup.
- NextFull backups first, then off-site copies at storage rates, plus snapshots.
- A nightly database copy that had stopped working for about a week is caught and fixed, with an alert added.
That gap was ours. On September 28, 2026, we found two servers with no backup at all. The order that followed was full backups first, then everything else: monitoring, upgrades and health checks came after the copies existed. The fix used off-site storage at object-storage rates, plus snapshots of the servers themselves. A few weeks later, a nightly database copy that had quietly stopped working for about a week was caught and fixed on October 15, and the fix added an alert for the next time a copy stops. Without the rule that a backup counts only once it works, that copy would have looked fine on the schedule.
A worked example: one lead’s journey
Here is what the five replacements look like on one real path: a lead’s journey from an ad to a clinical trial site at Enrollmentum. Every promise below is quoted from Enrollmentum’s public pages as they read on November 15, 2026. A public promise is the hardest kind of hero dependency to see. When a page says something will happen, someone has to make it happen every day, and if that someone is one person’s memory, the page has a hero in it.
-
“Your IRB approves the packet. Nothing runs outside it.”
“Budgets and launches are approved before they run.”
-
“Validated intake, a consent record, routing to each site, delivery checks and reconciliation, all built and tested before the first ad runs.”
“Each lead carries a record of the consent the person gave.”
-
“Leads arrive in the format the site works from, and every lead can be traced from the ad to the site.”
“Changes to the lead systems are deployed from code after automated tests.”
-
“Campaign pace against the agreed lead range, form quality and rejected leads, read every day.”
“Every message follows the study’s IRB-approved script, and a person takes over anything off it.”
-
“One report with leads delivered against your agreed range, the reasons people do not pass the pre-screen, what changed in the campaign and the leads sent to each site.”
“Walks each page and form end to end with a recording, and rehearses the lead path on a test stack.”
Before the first ad runs, the hand-off is built and tested. The services page promises “validated intake, a consent record, routing to each site, delivery checks and reconciliation, all built and tested before the first ad runs,” and that “every lead can be traced from the ad to the site.” That’s the Runbook and Automation replacements written as a promise to a buyer. The path exists in code and in writing before a single lead depends on it.
The path is rehearsed. Quality checks walk “each page and form end to end with a recording, and rehearse the lead path on a test stack.” That’s Recovery moved earlier: the team finds out how the path behaves on a test copy, not on a real person’s inquiry.
Approval sits where it matters. “Budgets and launches are approved before they run.” That’s the Owner replacement placed on the decisions that carry the most risk, instead of on every step.
Exceptions go to a person. When a study uses follow-up, “every message follows the study’s IRB-approved script, and a person takes over anything off it.” Automation runs the routine path, and the edge cases have a named place to land.
The sentence we changed
The honest center of this example is a sentence of ours. Until November 3, 2026, Enrollmentum’s team page said this:
Nick approves every client message, proposal, price, IRB submission, budget and launch.
He reviews the decisions that matter most to a study: proposals, prices, IRB submissions, budgets and launches.
Read the first version as a buyer and it sounds careful. Read it as an operator and it describes a bottleneck. If one person approves every client message, every message waits for that person, and the day they’re on a plane, nothing goes out. A review of the site in early November pointed at exactly that sentence as a single point of failure. The operator’s own site had a hero sentence in it.
The new sentence keeps a person on the decisions with the most at stake for a study and lets the written workflows carry the rest. It says what is true now, and it’s the same move this guide recommends for any business: put the owner on the outcome, not on every step.
What does it cost to keep a promise?
Every public technical promise needs an owner and evidence, or it drifts. A page says leads are traced from ad to site; somebody changes the routing six months later and nobody updates the page, or the page stays right while the system quietly stops matching it. That drift is the hero problem in documentation form: the promise is kept only as long as one person remembers it exists. The cure is the same five replacements. Name the owner of each promise, write down how it’s checked, automate the check, watch it, and know how you’d put it right. Enrollmentum describes the rest of the work on its services page.
Doesn’t automation make the hero problem worse?
There’s a trap here, and it catches the people who do this work well. Automate yourself out of the daily work and the work gets even harder to see. The renewals happen, the copies run, the alerts stay quiet. From the outside it looks like nothing is being done, so the people paying for it start to wonder what they’re paying for, and the next budget conversation treats the system as free.
The answer is to make ownership, responsibilities and boundaries explicit before anyone depends on them, and to price the ongoing care as a service. We eliminate unnecessary work. We don’t eliminate the expertise or the accountability behind it. A system that runs itself still has an owner, and the owner’s job is real even in a quiet month: checking that the checks still check, rehearsing the recovery, reading the alerts that didn’t fire and asking why.
- Setup and validationBuilt, tested and documented before anyone depends on it.
- Ongoing operationsAn owner keeps it watched, rehearsed and current, on written terms.
- Extra support, scoped separatelyThe work nobody planned, priced when it arrives instead of absorbed.
In practice, the work comes in three parts. Setup and validation, where the system is built, tested and documented. Ongoing operations, where the owner keeps it watched, rehearsed and current. And separately scoped extra support, for the work nobody planned. Service terms, written in advance, replace the open-ended rescue.
This matters more in regulated work. The U.S. Department of Health and Human Services, in its guidance on HIPAA and cloud computing, notes that service level agreements can address “system availability and reliability,” “back-up and data recovery” and “use, retention and disclosure limitations.” Those are written terms, agreed before anything goes wrong. A rescue on a bad night doesn’t create a service level after the fact, however good the rescue was.
Run the Hero Audit
The Hero Audit maps your own single points of failure in a few minutes. List up to eight things your business can’t go a week without: the website, payroll, the ad accounts, the lead inbox, the domain. For each one, answer five questions: how many people can do it, whether it’s written down, whether it’s watched, whether the recovery has been rehearsed, and whether it’s automated.
The result is a map of which items rest on one person, the first one to fix, and which of the five replacements it needs first. It stores nothing. The share link carries your answers in the link itself, so you can send it to a colleague and they see the same map. Below is a filled-in illustrative example; start your own to replace it.
1. List what you can’t go a week without
2. Answer five questions for each
Add an item above to answer its questions.
Illustrative example
Fix the lead inbox first. It needs an owner first, then a runbook, monitoring, a rehearsed recovery and automation.
- Make the lead inbox a role’s job, not one person’s habit: name the owner of the outcome and a second person who can step in.
- Write the first five steps for the lead inbox where the second person will look for them, then have them follow it once.
- Set up a check that tells the owner when the lead inbox fails, before a customer does, and send it to more than one person.
- The website: rests on one person; needs an owner, a runbook, automation, a rehearsed recovery
- Payroll: two or more people; covered
- The lead inbox: rests on one person; needs an owner, a runbook, automation, monitoring, a rehearsed recovery
- Ad accounts: two or more people; needs a runbook, automation, a rehearsed recovery
- The domain: rests on one person; needs an owner, a runbook, monitoring, a rehearsed recovery
Read the example the way you’d read your own. Three of the five items rest on one person, and the lead inbox has every gap, so it comes first, even though the website feels more urgent. The order of the fixes follows the 30-day plan below: an owner and a written first step come before monitoring, rehearsal and automation, because each of those needs someone to own it and steps to follow.
Nothing you type leaves your browser, and the audit gives no score. There’s no grade, only a list you can take into Monday’s meeting with the first item already chosen.
If your map comes back all clear, look again at the list before you trust the answers. The items that are easiest to leave off are the ones nobody thinks of as a system: the domain renewal, the card on file for the hosting bill, the shared inbox where client requests arrive, the one spreadsheet that feeds the monthly report.
What changes from one industry to another?
The pattern is the same everywhere. What changes is what an undocumented rescue puts at risk.
The risk: a fix that contradicts a published commitment
Consent, retention and privacy promises make undocumented rescues riskier. Enrollmentum’s security page commits that “each copy of a lead that we control has a retention period, and an erasure request follows a written procedure.” Picture a late-night workaround that exports leads to a spreadsheet to get them to a site. It works, and it creates a copy no retention rule knows about. A fix nobody wrote down can break a promise somebody did.
The risk: the favor nobody invoices
I’ve hosted sites since Apache and IIS were new, including years of sites for friends and family. It’s the work nobody invoices and everybody notices when it stops. A favor has no contract and no second person, so the domain renewal, the certificate and the backup all wait on whoever set them up. Each one still needs a named owner and a check that runs without that person, even when nobody is paying for it.
The risk: growth that queues behind one person
At LeisurePro, I learned the business and automated it, then built the team that ran it, through No. 1 in its market and the purchase of Scuba.com. I grew it from a largely Northeast business into the EU, then took over US and Scuba.com’s West Coast retail and distribution. Every new market adds a process somebody carries in their head. Automating it, or writing it down for the team that runs it, keeps the growth from waiting on the one person who knows how it works.
The life sciences tab is the one Illuminare’s readers live in. In regulated work the written record is part of the product, so an undocumented rescue is a risk twice over: once to the system, and once to the commitment the system was built to keep.
For a sponsor, a CRO or a research site, the practical test is short. Pick one promise on your own site or in your last proposal, such as how a lead is delivered or how long data is kept, and ask who checks it, how, and when they last did. If the answer is a name and no record, that promise rests on a hero.
Case: the card-testing attack
- FoundA live card-testing attack on a client’s online store.
- Shut downThe response was led to the end and the attack stopped.
- LessonThe checkout needs an owner and a check before the next one.
Eight hero-dependency mistakes
Each of these is common because each one feels like progress at the time.
- Thanking the rescue and changing nothing.Fix: End every incident with one written change.
- A backup with no restore test.Fix: Restore one on purpose and time it.
- One shared login.Fix: One login per person, and a named second for every account.
- Monitoring that alerts the person on vacation.Fix: Send every alert to the owner and the second.
- Runbooks nobody has used.Fix: Have someone new follow it once, then fix what they tripped on.
- Automating a step nobody owns.Fix: Name the owner before the schedule.
- Publishing a promise with no owner behind it.Fix: Every public promise gets an owner and evidence.
- Treating a one-time favor as a service level.Fix: Write the service terms before the next rescue.
Some of the eight look harmless on their own. One shared login feels efficient until the person who changed the password is the only one who knows it, and the account’s recovery email goes to someone who left. Monitoring that alerts one person moves the hero from the fix to the pager: the business still waits for the same person, only sooner. A runbook nobody has used is a guess written down; the first person to follow it finds the step that was obvious to its author and missing from the page. Automating a step nobody owns produces the quietest failure of all, a job that stops running while the schedule still looks full.
Publishing a promise with no owner behind it is the mistake the worked example corrected. A page is read long after the person who wrote it has moved on, so the promise needs someone whose job includes checking that it’s still true.
The first one does the most damage, because it teaches the loop to keep running. Google’s SRE book describes the alternative: a postmortem, “a written record of an incident, its impact, the actions taken to mitigate or resolve it, the root cause(s), and the follow-up actions to prevent the incident from recurring.” It also insists the record be blameless, for a reason that fits this guide: “You can’t ‘fix’ people, but you can fix systems and processes to better support people making the right choices.”
The last one is the quietest. A rescue done once as a favor becomes, without anyone saying so, the expected response time. An incident doesn’t create a service level after the fact. Write the terms before the next one, and the favor stays a favor.
A mistake of our own
Case: the overnight update
- Our scheduled update run starts on a server.
- The sites on it stop answering. A step has no way to recover on its own.
- The server is back. Every public site we host answers as it did before.
- The update is rewritten to recover end to end.
On October 5, 2026, our own scheduled update took a server down overnight for a few hours. The update did what it was meant to do; the trouble was what came after a step went wrong. The run had no path to recover by itself, so the sites waited for a person. We rewrote it the same morning to recover end to end: it brings disks back one at a time, waits, retries, starts the server, checks its health and sends an alert. A check of every public site before and after showed no change once the server was back. It’s the thesis in small. An owned system includes its own recovery, so nobody has to be awake at four in the morning for it to come back.
Run a hero audit in a month
You don’t need a program or a budget line to start. One month, one meeting a week, and the list from the Hero Audit.
- Week 1List the critical itemsEverything the business cannot go a week without. Run the Hero Audit on the list.
- Week 2Name owners, write first stepsAn owner and a second for each item, and the first five steps for the riskiest one.
- Week 3Watch it, restore onceAdd a check that alerts two people, and test one restore from backup, timed.
- Week 4Automate and write termsAutomate the routine part of the riskiest item and write its service terms.
In week 1, the list matters more than its order. Include the boring things: the domain renewal, the payroll login, the account that pays the hosting bill. In week 2, name an owner and a second for every item, and give the second real access the same day. In week 3, add a check that alerts both of them for the riskiest item, and test one restore from backup with a timer running. In week 4, automate the routine part of that item and write its service terms, then run the audit again. The second map should have fewer lines running to one person than the first.
Then do it again next quarter. Treat the audit as a habit rather than a project, because the next hero forms the same way the last one did: one quiet rescue at a time. Keep both maps and put them side by side in the next quarterly review. The lines that moved are the work that was done; the lines that didn’t are the agenda.
Questions people ask
What is key-person risk?
The risk that a business cannot do something important when one particular person is unavailable, because only that person has the access, the knowledge or the habit that makes it work.
What is a single point of failure in a business?
Any one person, account, system or vendor whose absence stops something the business cannot go without. A hero is a single point of failure that happens to be a person.
What is a runbook?
The written first steps for a task or an incident, kept where the next person will look, so that someone who has never done it can follow them and get the same result.
How do you reduce dependence on one employee?
Name an owner and a second for each critical item, write the first steps down, give the second real access, automate the routine part, and add a check that alerts both of them.
How often should you test backups?
Often enough that a broken backup is found before you need it. Pick a cadence, restore on purpose, time it, and write the steps down. CISA advises regular tests of partial and full restores.
Does automation remove the need for an owner?
No. Automation moves the routine part to a schedule; someone still owns what happens when it breaks. Name the owner before you automate.
What is the Hero Audit?
A free check on this page: list up to eight things your business cannot go a week without, answer five questions for each, and it maps which rest on one person and names the first one to fix. It stores nothing.
Sources
- Vivek Rau, “Eliminating Toil,” chapter 5 of Site Reliability Engineering, Google. Cited for the definition of toil.
- John Lunney and Sue Lueder, “Postmortem Culture: Learning from Failure,” Site Reliability Engineering, Google. Cited for the definition of a postmortem and blameless culture.
- Marianne Swanson and others, NIST Special Publication 800-34 Revision 1, Contingency Planning Guide for Federal Information Systems, May 2010. Cited for “Testing validates recovery capabilities.”
- Cybersecurity and Infrastructure Security Agency, Cyber Guidance for Small Businesses. Cited for testing partial and full restores.
- U.S. Department of Health and Human Services, Guidance on HIPAA and Cloud Computing. Cited for what service level agreements can address.
- Enrollmentum, Services, Team and Security practices, read November 15, 2026. Quoted in the worked example.
The opening scene and the audit example are illustrative. The incident dates come from our own operations records; the earlier-career examples come from the author’s work history.