25minutes
The Busywork Tax: What Workflow Automation Actually Removes
The manual step takes ninety seconds. In the study that timed it, getting back to interrupted work averaged twenty five minutes when it resumed the same day. Here is what workflow automation actually removes from a real estate business, how to find your own version of it in an hour, and the failure mode nobody warns you about.
Ask an agent what they do all day and they will tell you they sell houses.
Then get them to write down every step of one deal, from the first inquiry to the closing table, and hand them a highlighter for anything that involves typing something into a system that another system already knows. The page turns yellow.
That yellow is the busywork tax. It is not one big thing. It is forty small ones, each of which takes about ninety seconds, and none of which feels worth automating on its own.
Ninety seconds is also not what any of them cost.
In short
- The ninety seconds of typing is not the cost. Twenty four information workers were timed to the second at their desks, and the interrupted work that got picked up again the same day took an average of twenty five minutes and twenty six seconds to come back to.
- You are not automating step four. You are automating one through eight, because the chain is where the reloading happens and a chain is exactly what these tools are good at.
- Zapier's own documentation says it pauses a chain that errors ninety five percent of the times it ran in the last seven days. Read that the other way: a chain failing one time in twenty is a chain nobody is going to tell you about.
Watch it work
One lead, by hand and then wired.
Narrated, 59 seconds. The voice reading it is a licensed clone of my own, which seemed worth saying rather than leaving you to wonder. The lead is invented and the timestamps are staged, but both numbers on screen are the cited ones from further down this page: the twenty five minutes is the field study and the ninety five percent is Zapier's own documentation. The chain it draws is the one on the service page.
What ninety seconds actually costs
In 2005 three researchers at the Donald Bren School of Information and Computer Science, University of California, Irvine published the results of shadowing twenty four information workers at their desks with a stopwatch. Seven managers, nine analysts, eight developers. Each one was formally observed and timed for three and a half days, an average of twenty five hours and forty two minutes per person, more than seven hundred hours of observation in total, with every action noted to the second.
It is not a survey of how busy people feel. It is a timed log of what they actually did, and it is the clearest measurement of the thing this article is about that we have been able to find.
They found that people spent an average of eleven minutes and four seconds on one piece of work before switching to something else or being interrupted, and that fifty seven percent of those stretches ended in an interruption rather than in a decision to stop. Then they measured the part almost nobody measures: how long it took to come back.
The evidence
How long it took to get back to interrupted work
Average elapsed time, 24 information workers observed at their desks and timed to the second. All three resumption figures count work picked up again the same day, which was 77.2 percent of everything interrupted. The middle figure is not a third category: nine resumptions in ten were self-started, and 90.1% at 21 min 28 sec with 9.9% at 61 min 37 sec averages out to exactly the 25 min 26 sec on the third bar. Source: Gloria Mark, Victor M. Gonzalez and Justin Harris, No Task Left Behind? Examining the Nature of Fragmented Work, Proceedings of CHI 2005, University of California, Irvine. Over 700 hours of observation; 57 percent of work segments ended in an interruption and 77.2 percent of interrupted work was resumed the same day.
These were managers, analysts and software developers at one technology company, observed in a single field study, and not one of them sold a house. It measures how fragmented desk work is. It is not a measurement of how much of your week is copy and paste, it cannot be turned into one, and no automation should be sold to you on the back of it. What it establishes is the only thing this page needs it for: the interruption costs far more than the task that caused it.
Put the middle bar next to the ninety seconds. The typing is not the cost. The cost is that a ninety second job pulls you out of something, and the something takes an average of twenty five minutes and twenty six seconds to get back to, counting only the work that got picked up again the same day.
The third bar is the one that should worry a real estate business in particular, because it is the shape of every step in your week that only happens when somebody remembers. Work that people picked back up themselves resumed in twenty one minutes and twenty eight seconds. Work that waited for somebody else to prompt the return sat for sixty one minutes and thirty seven seconds, nearly three times as long.
One caution about that study, because it is the honest way to use it. Those were desk workers at a technology company, in one field study, and none of them were selling houses. It tells you what interruption does to knowledge work. It does not tell you what your own week looks like, and anybody who converts it into a dollar figure for your business has stopped citing it and started decorating with it.
What it actually looks like
Here is a real chain, the kind that exists in almost every real estate business, written out honestly:
- A lead fills in a form at 11:47pm.
- It lands in an inbox.
- In the morning, somebody reads it.
- They type the name, email, and phone into the CRM.
- They do not notice this person is already in the CRM, so now there are two of them.
- They write a reply.
- They set a reminder to follow up.
- Three days later the reminder fires and they cannot remember who this was, so they read the original email again.
Eight steps. Maybe twelve minutes of human attention, spread across three days, most of it spent reloading context that was already written down somewhere. Step eight is the twenty five minutes in miniature, three days late.
Now do that forty times a month.
Why nobody fixes it
Three things we tell ourselves.
It only takes a minute.
It takes a minute forty times a month, and forty minutes is the smallest possible version of that number. A minute of typing is never a minute. It is a minute plus whatever it costs to get back to what you were doing, which the study above puts closer to twenty five.
I would rather do it myself, so I know it is done right.
This is the honest objection and the other two are excuses. Wanting to see the work is correct. The answer is not to trust a system blindly, it is to build one you can watch, and that is a different thing with a different price.
It is not worth automating something this small.
Individually, correct. The point is that these steps form a chain, and a chain is exactly what these tools are good at. You are not automating step four. You are automating one through eight, and the saving is in the joins.
What automation actually is, without the jargon
It is plumbing between the tools you already pay for.
Your CRM holds the contact. Your calendar holds the time. Your email holds the conversation. Your forms hold the intake. Your documents hold the deal. All five of them know things the other four need, and the way that information currently travels between them is a human being with a mouse.
Tools like n8n, Make, and Zapier connect them, so that finishing one step starts the next. Put a language model in the middle and it can also make the small judgment calls that used to need a person: is this the same Sarah Miller who inquired in March, is this lead actually hot, does this message need a human being rather than a template.
The system
The same job, with nobody in the middle of it.
Every hop below is a step somebody used to do by hand the following morning. The two timestamps are the point: the first and the last are the same minute.
Scroll to follow the chain
- The form: Site or portal
- The record: Matched, not duplicated
- The details: Phone and address checked
- The reply: A real answer, in seconds
- The route: Scored, then assigned
- The task: Calendared, with the context
Nobody woke up. Nobody typed. The lead was answered while they were still on the site, the second record was never created, and the reminder in step seven was never needed, because the task arrived with the context already attached to it.
How to find your own version of this
You do not need a consultant for the first pass. You need a piece of paper and an honest hour.
Take this, whether or not you buy
An hour, a piece of paper, and four honest questions.
Follow one real job, end to end.
Not the ideal version. The one that actually happened last week, including the part where somebody had to chase a signature twice. Write down every step, including the ones that feel too small to write down. Those are the ones.
Mark every step where information moves between two systems by hand.
That is your list. It is always longer than the person writing it expected, and the worst offenders are invisible because everyone stopped noticing them years ago.
Mark every step that only happens if somebody remembers.
That is your risk, and it is a different list. In the study above, work people came back to themselves resumed in twenty one minutes and twenty eight seconds. Work that waited for somebody else to prompt it sat for sixty one minutes and thirty seven seconds.
Rank by how often it happens, not by how annoying it is.
The most irritating task is rarely the most expensive one. The expensive one is the boring thing you do fifty times a month without noticing. Most people are surprised twice: by how long the list is, and then by how dull the top of it is.
Then rank it, because the order you do them in matters more than the list does. Sort by two things and ignore everything else: how often the step happens, and how little judgment it needs. The top of that list is always something dull and frequent, a field being copied from one system into another twenty times a week, and it is worth more than the impressive-sounding thing at the bottom that happens twice a month and needs somebody to think.
Most people get this backwards, and it is an expensive way round. The interesting problem is the one with judgment in it, so that is the one they want to automate first, and it is the one most likely to be wrong in front of a client. Start with the boring repetitive hop that nobody will miss, watch it run for a fortnight, and let the trust be earned by something whose failure costs an apology rather than a deal.
There is a third column worth adding while you have the page out: who notices when this step does not happen. If the answer is nobody, you have found something more useful than a time saving. You have found a step that has probably already been skipped, more than once, and nobody knows which deals it was skipped on.
When you have that list in front of you, the arithmetic is the easy part.
In your numbers
How much of your year is the busywork tax?
Count from the audit above: every time somebody types in something another system already knows.
The article's example is ninety seconds. Most people guess low here, then find the real number in the audit.
Yours, or whoever actually does the typing.
Hours of typing, a year
104hours a year
- Manual stepsyour 60
- 60a week
- Minutes of typingyour 2
- 120minutes a week
- Over a year52 weeks
- 6,240minutes a year
- In hours60 minutes in an hour
- 104hours a year
- At your hourly valueyour $75
- $7,800a year
This counts the typing and nothing else, which makes it the smallest honest version of the number. It deliberately does not multiply anything by twenty five minutes and twenty six seconds. That figure is real and it is on this page, but it was measured on twenty four desk workers at a technology company and none of them were selling houses, and this article says a few paragraphs up that anybody converting it into a dollar figure for your business has stopped citing it and started decorating with it. That applies to us. So the interruption cost, which is almost certainly the larger half, is missing from the number above on purpose, and the hourly figure is yours rather than an average we picked.
The task worth automating first is almost never the one that sounds impressive. It is the one you have done so many times that you stopped seeing it.
The failure nobody warns you about
Automation you cannot see is automation you cannot trust, and the failure is never the one people brace for. Nothing explodes. A chain quietly stops firing and nobody notices for a fortnight, because nothing visibly broke. It just went quiet.
The platforms are more candid about this in their documentation than any vendor will be in a sales call. Zapier's own help pages state that by default it "automatically pauses a Zap if it hits an error 95% or more percent of the times that it has run in the last 7 days", and its run status reference adds that a chain erroring repeatedly "will automatically turn off".
That is a sensible default and it is not the interesting part. Read the threshold the other way round. A chain that fails nineteen times out of twenty gets switched off, and you find out, because the work visibly stops. A chain that fails one time in twenty stays on, stays green, and is doing exactly what it looks like it is doing ninety five percent of the time. Over forty leads a month that is two people who wrote to you and got nothing, every month, and there is nothing in the system whose job it is to tell you their names.
The same asymmetry is in the tools that do it properly. n8n documents an error workflow that begins with an Error Trigger and runs when an execution fails, so a failure can send a message to a person instead of landing in a log. It is one setting per chain. Almost nobody sets it, because on the day you build a chain it works, and a thing that works does not feel like it needs a smoke alarm.
The part nobody sells you
Three rules that keep a chain visible.
Every run gets logged.
If you cannot answer did it run, and what did it do, you have not built an automation. You have built a black box that mostly agrees with you. Every one of these platforms keeps a per-run history; the real question is whether anybody has opened it since the week it went live.
Failures shout.
n8n's documented pattern is a separate error workflow, beginning with an Error Trigger, that runs when an execution fails and can message a person. It is one setting on each chain and it takes five minutes. Without it, a failure lands in a log nobody reads.
Anything genuinely ambiguous stops and asks.
The line is easy to find: any step where a person would have hesitated is a step the machine should hand back. Hesitation is information, and it is the only signal you get before the mistake.
The alarm, at least
Whatever you are running today, tell us which platform it is on and we will send back the exact setting that makes a failed chain tell a person instead of a log file.
It works whether or not we ever build you anything, which is rather the point.
What it costs, and how long it takes
There is no honest price on this page, because the number depends on three things nobody can guess from an article: how many systems the chain has to touch, whether those systems have a usable interface for software to talk to, and how many steps need a judgment rather than a field copy. Anybody who quotes you before asking all three is quoting a template.
What can be said honestly is the shape of it. The platforms themselves are the cheap part and are billed by how often your chains run, so a small business pays a subscription rather than a project. Simple chains, like intake and an instant reply, are typically live in days. Multi-system workflows with real branching take longer, and the slow part is almost never the building. It is mapping what your business actually does today, which is the hour with the piece of paper, done properly and with somebody arguing with you about it.
The recurring cost people forget is ownership. A chain is software. It will break the day a vendor renames a field, and somebody has to be the person who notices. Budget for that, or the chain quietly becomes one of the mistakes it was built to remove.
The first month, and the two things to do in it
Nearly every chain that dies quietly dies in its first four weeks, and it dies because the go-live was treated as the finish. Two habits in that first month are worth more than anything you can specify beforehand, and both are free.
The first is to open the run history in week one and read it, line by line, for one chain. Not to check whether it worked. To check whether what it did matches what you thought you asked for. This is the week you discover that a duplicate record is being created rather than matched, or that the reply going out in your name says something you would never say, or that the step you assumed ran on every lead is only running on the ones that came through the website. None of that shows up as a failure. It shows up as a green run doing the wrong thing politely.
The second is to break it on purpose. Put a deliberately bad value into one chain, or switch off a credential for ten minutes, and then sit and wait to find out whether anybody is actually told. If nothing arrives, you have learned the most important fact about your own system before it mattered, and you have learned it on a test record instead of on a client. That is the whole argument of the section above, run once, in the only way that proves it.
After the first month, put a repeating reminder in your own calendar to open the run history once a quarter. It takes ten minutes and it is the difference between a system you own and a system that owns a corner of your business without telling you what it is doing in there.
What it does not do, and should not pretend to
It does not fix a bad process. If the manual version of the job loses leads, the automated version loses them faster, at three in the morning, with a log entry saying it worked. Wiring makes a process consistent, and consistency is only an improvement when the process was right.
It does not remove judgment. Every chain worth building has a step where a person would have paused, and the correct behaviour at that step is to stop and ask rather than to guess. A system that guesses when it should have asked will eventually guess wrong in front of a client, in writing, at a time of its own choosing.
It is mostly not artificial intelligence, whatever it gets sold as. Four of the six hops in the diagram above are a field moving from one place to another with no cleverness in them at all, and they are more reliable for it. A model earns its place at the one or two steps that genuinely need a decision. A build that puts a language model in front of a step which was really an if statement has bought unpredictability it did not need and cannot debug.
And it does not hand you back a day. It hands back an afternoon a month and removes a category of mistake. That is a smaller and duller claim than the one usually made for this, and it is the one that survives contact with a real business.
Three ways it is wasted
None of them are the technology.
You automate the impressive one.
The chain that demonstrates well is almost never the chain that runs forty times a month. The boring one at the top of the frequency list is worth more than the clever one, and it is the one nobody volunteers to build.
It automates a bad process, faster.
If the manual version loses leads, the automated version loses them at three in the morning, at scale, with a log entry saying it worked. Fix the process on paper first, while it is still cheap to argue with.
Nobody owns it.
Chains get built during a project and abandoned after it. Six months later a vendor renames a field, and the person who knew what the third step did has moved on. Write down who is responsible for each chain before you switch it on, or it becomes one of the mistakes it was built to remove.
Common questions, answered honestly
What is workflow automation, in plain terms?
It is connecting the software you already use so that finishing one step automatically starts the next. Instead of a person copying a lead from a form into a CRM, writing a reply, and setting a reminder, the whole chain fires by itself the moment the form is submitted.
What is the difference between n8n, Make, and Zapier?
They all wire apps together and they differ in depth. Zapier is the simplest and the most limited. Make handles branching and more complicated logic. n8n is self-hostable and the most flexible, which matters when a workflow needs custom code or the data has to stay in your own environment. The right answer is decided by the workflow, not by a preference, and it is worth asking anybody who tells you otherwise why.
Do I have to replace the software I already use?
No, and that is the point of it. Automation sits between your existing tools and connects them. Your CRM, your calendar and your inbox stay exactly where they are, which is also why this is usually the cheapest improvement available to a small business: nothing has to be migrated.
What happens when an automation breaks?
It should alert, log what failed and why, and not silently drop the work. That is a build decision rather than something you get for free: the platforms will retry and will eventually switch a chain off, but only a chain that was built with an error path tells a human being. Chains should also be versioned so a bad change can be rolled back, and any step needing real judgment should be built to stop and ask.
How long does it take to automate a workflow?
Simple chains, like intake and an instant reply, are typically live in days. Multi-system workflows with real branching take longer, mostly because mapping what your business actually does today is the slow part, not the building.
Is any of this worth it for a one-person business?
Often more, not less. A one-person business has no one to absorb the busywork, so every manual step is taken out of the only calendar there is. Start with a single chain at the top of the frequency list rather than a platform, and judge it after a month against how many times you touched that job by hand.
What to do about it
Go and get the highlighter. One deal, from the first inquiry to the closing table, every step written down in the order it really happens, and a mark against anything where a person types in something another system already knows. Nobody has to approve it, it costs an hour, and at the end of it you are holding the only document that makes any of the rest of this decidable. Until that page exists, every quote you are given is a guess about a business the person quoting has not seen.
You can see the wiring on the RealtyLT AI page, and what gets connected to what is on the workflow automation page. If you would rather somebody sat through that hour with you, that is exactly what the AI audit is: we follow one real job, rank what each fix is worth, and build the first one.
Three of the hops are written out at length on their own: the 11:40pm website conversation, the 9:42 Sunday phone call, and scoring the lead once it arrives.
The tax is not going to itemise itself.
Somewhere in your week is a step you have done four hundred times and stopped being able to see. It is not the interesting one. It is the dull one that has been costing you an afternoon a month since the year you started.
There is no price on this page because there is no honest one: what a chain costs depends on how many systems it has to touch and how much of it needs a judgment rather than a field copy. The AI audit is the hour described above, done with you, and it ends with the first chain built rather than with a document.
