Nobody Reads the DARs. Reading All 8,400 of Last Month's Now Costs Less Than One Guard-Hour.

Last month your officers filed something close to 8,400 daily activity reports. Somebody read eleven of them. Nine because a client complained, one because an attorney's preservation letter arrived about a slip in a parking garage fourteen months ago, and one because your account manager was bored on a flight.

That is not negligence. It is arithmetic. A DAR runs 150 to 250 words. A supervisor who reads carefully gets through maybe twenty an hour. Eight thousand four hundred reports is 420 hours of reading a month — two and a half full-time people whose entire job would be reading about doors that were locked. No security company in the United States has ever staffed that, and every owner in the business has quietly accepted that the reports exist for the lawsuit, not for management.

What the price of reading did between 2025 and 2026

Here is the change that matters, and it is not a new feature. It is a price. Running strong AI models fell roughly tenfold from 2025 into 2026, with capable models now priced around two dollars per million units of text, and high-volume work run on-device costing roughly 90 percent less again than sending it to the cloud. Work that was absurd to run on every single record in 2025 — every report, every tour, every pass-down — is now a rounding error against one guard-hour of payroll.

The per-record check a guard company skipped because it was uneconomic is reading every daily activity report and incident report the night it is filed, and flagging the handful that need a human before the client sees them. In 2025 that was a project. In 2026 it costs less than the coffee in the guard shack.

The three things hiding in the reports you skipped

First, copied reports. The officer on the 2200 to 0600 post at the construction site logs "perimeter secure, no issues" at 0301, 0302 and 0304 every night for three weeks, with the wording identical to the character. Your Silvertrac or TrackTik dashboard shows green because the buttons got scanned. Somebody is scanning the NFC tags on a loop from the trailer and writing the same sentence. You will find out when copper goes missing.

Second, missed and clustered tour hits. Not the missed hit the system already flags — the pattern underneath it. Eight checkpoints that are supposed to take 25 minutes get scanned in nine, three nights a week, always on the same shift. The system counts them as complete. A reader that looks at the timestamps as a sequence sees a guard driving the tour instead of walking it.

Third, and most expensive: the language that turns into a claim. "Subject stated she slipped near the elevator lobby, declined medical, left on her own." That sentence sits in a DAR for a year and then becomes the centerpiece of a demand letter, and nobody at your company knew it existed. The same goes for a fire watch log that has a two-hour hole in it while the sprinkler system was impaired — the hourly rounds your authority having jurisdiction and the property's insurer expect are only worth something if the log is complete, and the time to find the hole is the next morning, not at the deposition.

flowchart TD
  A["Night shift files 280 DARs and 6 incident reports"] --> B["Overnight read of every report filed"]
  B --> C["Check 1: tour hits too fast, too clustered, or missing"]
  B --> D["Check 2: wording copied from a prior night"]
  B --> E["Check 3: injury, threat or police-contact language"]
  C --> F["6am exception sheet, 9 items"]
  D --> F
  E --> F
  F --> G["Account manager works the list before the client calls"]

The 6 a.m. exception sheet

The output is not a dashboard. Owners in this trade do not need another dashboard. The output is a single sheet on the operations manager's screen before the day shift briefing, and on a normal Tuesday it has somewhere between four and a dozen lines on it.

Line one: the Riverfront garage account, officer badge 118, three reports this week with wording identical to last week's, recommend a supervisor ride-along tonight. Line two: the cold storage warehouse, fire watch log shows rounds at 2200, 2300, 0100 and 0200 — the midnight round is missing while the system was impaired, get a written statement today. Line three: the medical office building, a visitor incident on Sunday mentions a fall and declined medical; the incident report was never escalated to an IR number and the client's risk manager has not been told.

Nothing on that sheet was invisible before. It was all sitting in reports your company already had. The only thing that changed is that reading them stopped costing 420 hours.

The arithmetic, with the assumptions on the table

Illustrative figures for a company running roughly 280 posts a night across a metro area.

So the running cost of reading everything is under $300 a month, nearly all of it your operations manager's time. Against that: one general liability deductible in the $10,000 to $25,000 range that you did not eat because the fall got documented and reported the same week; one account you did not lose because you found the phantom patrols before the client's own camera review did. You do not need both to happen in a year for the math to be uncomfortable in the other direction.

The reports a machine must never grade alone

A flag is not a finding. If an officer's reports look copied, that is a conversation with a field supervisor, not a termination email — officers on a quiet static post genuinely do write the same sentence because the same nothing happened, and firing a good officer over a pattern match is how you lose a post and a wrongful termination argument at the same time.

Anything that becomes evidence stays human. Incident reports that involve an injury, a use of force, a police response, or a weapon get read by a person, get an IR number, and get escalated on your written timeline, every time, regardless of what the overnight read said. The machine's job is to make sure the report reaches that person by 6 a.m. instead of being found by opposing counsel in eighteen months.

And be deliberate about the record you are creating. Once you read everything, you know everything, and "we did not know" stops being available to you. Most owners decide that is a good trade, because the alternative is that the information existed in your own files and you were the only party in the room who had not read it. Talk to your carrier and your attorney about retention before you start, not after.

Frequently asked questions

Half my sites still hand in paper DARs. Does this work on those?

Yes, though it adds a step. Photos of handwritten reports read reliably now, including the notes crammed in the margin, but you need somebody at the branch to photograph the shift's paperwork at handoff. Most companies find the paper sites are also the sites with the oldest post orders, and end up moving them onto the phone app once they see the difference.

Do I have to tell my officers their reports are being read?

Tell them. Their reports were always company records and always readable, but people behave differently when they learn about it after the fact. Frame it the way it actually works — missed hits and thin reports get a coaching conversation, and the officer who writes a genuinely detailed report finally gets noticed for it, which is a first in this industry.

What does it cost to try this on one account?

Almost nothing in machine time. The real cost is the 25 minutes a day somebody spends working the exception sheet, and the two or three weeks of tuning what counts as an exception so the sheet does not arrive with ninety lines on it. Start with one 24-hour account and one check — tour timing patterns is the easiest one to argue about with evidence.

Will it write the client's monthly report for me?

It will draft it, and the draft will be better sourced than the one your account manager assembles on the 28th, because it has actually read the month. Somebody still signs it. Anything that goes to a client under your license gets a human name on it.

An incident found at 6 a.m. generates phone calls by 8 — the property manager, the tenant, the client's risk contact, sometimes a reporter. CallSphere builds AI voice and chat agents that answer business phone lines and web chat 24/7, capture who called and what they needed, and route the ones that matter to your account manager instead of letting them pile up in a shared voicemail box while your team is still working the exception sheet.