Every security director we talk to has the same two problems, and they don't sound related until you put them next to each other.
The first is that they can't fill posts. The second is that the posts they do fill cost more every year and turn over before the training is amortized.
Those are one problem, and it is structural.
The supply side is flat, and it is projected to stay flat
The U.S. Bureau of Labor Statistics projects approximately zero net growth in the security guard workforce through 2034. Not slow growth. Not growth lagging demand. Flat, against a workforce of about 1.27 million people.
Underneath that flat line is enormous churn: roughly 162,000 openings a year, essentially all of them backfill, against annual turnover running 100–200%. At 150% turnover, a 20-post contract hires 30 people a year to stand in the same 20 places. Each of those hires costs $1,500–3,500 in recruiting, screening, licensing, and training before they are useful, and a meaningful share leave before they get there.
Meanwhile the demand side is growing. The manned security market runs about $49 billion today toward a projected $70 billion by 2030.
Demand for watching is growing. The supply of watchers is fixed. There is no version of that where the price of a post goes anywhere but up, and no version where the fill rate improves.
What a post actually costs
The loaded cost of one 24/7 post is $150,000–220,000 a year. That's not the wage — it's the wage plus burden, plus relief coverage for breaks, vacation, and no-shows, plus the churn cost above, blended across the year. Divided out, it corresponds to roughly $17–25 per staffed hour against 8,760 hours.
The number surprises people who have only seen the bill rate. It shouldn't: covering a position continuously requires substantially more than one person, and the industry's turnover means you are permanently paying to replace part of that set.
A post does three different jobs, and only one of them is scarce
This is the distinction that we think should drive every renewal conversation, and it usually doesn't get made.
Deterrence. A visible human changes behavior. This is real, it is well documented, and no camera has ever replaced it. A uniformed presence at a lobby desk is doing work that a sensor cannot do, because the work is being seen.
Response. Someone physically intervenes — opens a door, escorts a person out, renders aid, holds a scene until police arrive. Also irreplaceable, and the response-time requirement is what determines whether a human has to be on-site or can be dispatched.
Watching. Someone looks at things — monitors, a corridor, a gate — and decides whether what they're seeing matters.
Only the third one is scarce labor being spent on something a machine now does better, and it is usually the majority of the hours. A person watching a wall of tiles at 3 a.m. is doing the one job where human attention is at its worst and where continuous machine attention is at its best.
The honest boundary — and we'd rather state it than have a customer discover it — is that a site whose threat model requires a physical response in under two minutes keeps a human nearby no matter what the cameras know. If the requirement is deterrent presence at a specific location during specific hours, that's a post, and it stays a post. Anyone who tells you otherwise is selling you something that will fail on the night it matters.
How to price a post at renewal
Take each post and ask what it's for. Not what the contract says — what would actually go wrong if it were empty.
| If the post exists for… | Then… |
|---|---|
| Deterrent presence in a specific place and time | Keep it. It's doing work nothing else does. |
| Physical response under a hard time bound | Keep it, and price the time bound explicitly. |
| Watching cameras or a quiet perimeter | This is the one to convert. It's where the labor market is failing you worst and where machine attention is strongest. |
| Some mix, which is most posts | Split the hours. The 10 p.m.–6 a.m. block is usually almost entirely watching. |
Most portfolios we see are carrying watching hours inside posts that were scoped for presence, because that's how the contract was written in 2016 and nobody has re-derived it since. That's the arbitrage, and it doesn't require anyone to be laid off — it requires the hours you cannot fill to stop being the hours you most need filled.
What this looks like when it works
The pattern isn't a headcount cut. It's a reallocation.
The overnight watching hours move to continuous machine attention with a human in the loop for judgment calls. The presence hours stay, and often get better staffed, because the same recruiting pipeline is now filling half as many positions. The officers who remain spend their shift on the parts of the job that are actually skilled, which — not incidentally — is the single most effective retention lever anyone has found in this industry.
The guard doesn't disappear. The post does, one shift block at a time, as each renewal gets priced against what it's really buying.
Where this argument is weakest
Conversion is not 1:1. Removing watching hours from a post doesn't reduce cost proportionally, because relief coverage and minimum-shift rules don't scale linearly. A post that goes from 24/7 to 16/5 saves less than the hour count suggests.
The turnover figures are survey-based. The 100–200% band comes from industry surveys, not audited payroll data, and the high end is concentrated in low-wage contract work. A well-paid in-house team does not look like that.
Wage inflation could in principle change the arithmetic. If guard wages rose enough, supply would respond. It hasn't over the past decade, and BLS doesn't expect it to through 2034 — but a projection is not an observation, and this is the assumption we'd most want to be wrong about.
Machine attention has its own failure modes. It doesn't get tired, and it also doesn't get suspicious. It will miss the thing that looked ordinary and wasn't, in ways a good officer with ten years on a site would not. The argument for converting watching hours is an argument about coverage — most sites currently have no attention at all on most cameras most of the time — not a claim that a machine is a better guard than a good guard.