Bezos-Funded Deep Sentinel Tested

By Ethan Ace, John Honovich, and Rob Kilpatrick, Published Mar 28, 2019, 03:39pm EDT

Backed by Jeff Bezos, the Silicon Valley startup, Deep Sentinel, has declared:

No One Does Home Security Like We Do

Our Surveillance Team has eyes on criminals from the second they step on your front yard.

if our AI or personnel detects anything suspicious, we will contact law enforcement in seconds. Guaranteed.

with zero false alarms and verified crime in progress, police are quick to respond.

Those are really ambitious and aggressive claims.

But how well does it really work? We bought a system and tested it.

IPVM Image

We found some surprising positives and negatives for the product, both in how it worked, who should use it and risk for the company's long-term model.

Executive *******

*** **** **** ** Deep ********'* ****** ** that ** ******** * live ***** ** **** at ********* ******** **** happens ** ***'* *****, including *******, ****, ****, yourself, **** ****, ***., ready ** ******** *** police ** * ******'* notice. ***$**(*** $***) *** ***** covering * ******* *** 24/7 ********, ** ** a **** **** *** homeowners *** *** ********* about ***** ******** *** want *** ******* ** responders ******** ** ********* watching.

*** **** *********** ***** of **** ******** *******:

  • *** ********* **** ******. They ********** '*****' *******, cats, ****, ***, *** humans ** ********. **** is *** * ****** problem *** ***** ***** the ****** ** *** up *** *** **** Sentinel ********* ** ****** and ***** *** ** these ******* **** ************. However, ** ** * practical ******* *** **** Sentinel ***** ** **** drives ** ***** ***** and *** ********** ** responding ** ****** ******* rather **** ** ******* stream ** ******** ******.
  • ************ *** *********** ** this '***' ****** ******** a **** ****** ** technical ***** ** ******* outdoors *** *** *********** to ** ******* ***** few ****** ** ******* the ********* ** *** 3 *******.
  • ******* ** *********: ** design *** *******, *** system ** ****** *** monitoring ** *** *****, including **** *** ** home. *** ****** ****** be ******** ** ****** off, ** *** **** be *********** *** ** a *** * **** privacy **** **** *** to ** ******** *** each **** ******* ** wanted. **** ***** ** a ******* *** ***** who *** ************* **** operators ******** **** ***** time **** ** ******* or *** ******* ** moving ** *** ***** covered ** *** *******. Moreover, ** ** * problem *** *** ******'* already ***** ******* **** since ** **** **** when ********* *** ****** (i.e., **** ****** *** middle ** *** ***).

Positive ******* *********

** ***** **** ******** has * ****** ** be * *********** ****** if *** ** **** develop **** ****. ***** it ** ************* **** the '**** ********' ** Deep ******** ***** *********, this *** *******, ***** will ********** *** ********* of ***** ***** ** reducing ******** ******* ******. Moreover, ** **** *** ways ** ******** *** regular ******* ************ (*.*., whether ****** ***** *****, installation *** ******, ******* on * ******** ** on ******, ***.), **** would ****** *** ****** on *** ****. ***** the ********** ** *** team *** *** ******* by *****, **** **** the ********* *** **** to *******.

** ** ** ********, the ******* *** ***** of **** ******** **** not ******* **** **** or ********** *** **** Sentinel *** ** *********** to ******* *** ****** end ** *** ****** that **** *** * premium ** **** * security ****** **** ****** in ***** ************ *** immediate ******** ********.

Installation/Maintenance ******

**** ******** *** *** key ************ *** *********** issues *** ***** ** other *** *******:

  • ******* ************ ********: **** ******** ****** are ******* ****, **** no ****** *** ****** use ****** ** ***** consumer/DIY *****. **** ******** users ** ********** *** to **** ****** *** camera, *** * ******, and ******* *********** * wide ******* *********, ** ************ ******* in ****** **. ***** vs. ***** **. ******, etc. *** ****** ****** for **** ******** ** less ****** ** ******* these ****** ** *** desire ** ******* **** installation, *** ***** ** no ********* *******, ********* **** ********.
  • ******** ******* *******:**** ******** ****** ** average ******* **** ** up ** *** ******, meaning ********** ****** ****** to ****** ********* ** least ** ***** *** year (*.*., * ******* changed ***** * ****** or ****). ************, ***** is ** ****** *** wired ***** ** ***** as ** ***** ******* powered ******* **** ******, ******, ***** ***** ******* this ******** ***********. *******, Deep ******** **** **** are ********** **** ******* for ****** *******.

*******

*** **** ******** ************ Package *** ** ***** on ******* ******** ********** ~$*** *** *** requires *** **** ** buy ** ***** * year ************ ** *** monitoring ******* *** $** (now $*** *** *****).

***** *** ****** ** 2020 ** $***:

IPVM Image

Size **********

******* ** *** **** IQ, ******* ******* ******, the **** ******** ******* are ******* ** ****** but ****** ** ****** due ** *** ***** camera *** *** ******* compartment/speaker ***** **.

IPVM Image

Physical ********

*** **** ******** ******* have ** ******** **** factor *** ************, **** a ****** ***** *** size ** * **** Cam ** ****** ***** Cam, *** * ***** cylindrical ****** ********, ********** the ******* *** *******/***.

** **** * **** at *** ******** ************ of **** ******** ** this *****:

No ****** ****** ** ******

** **** * **** of *****, *** **** Sentinel ******* *** *** miss * ****** ******* into *** ***** ** view, **** ** ***** sent ***** ****, *** and *****.

IPVM Image

*******, ***** **** ******** mis-identifications ** ***** ******* as ******, **** ** animals, *******, ***., ******** later. ****** **** **** these ****** ***** *** be ***** ** ******* users, ** ********* ***** review **** *** ********* no ****** ****** *** in *** *****.

Quick ******** ******** ****

**** * ****** ** vehicle ******** * ************, the ***** **** **** sent ** * **** operator ** ** ******** for ********** ********. ** our *****, ********* ***** check *** ****** ************* within *** ** ****** seconds.

IPVM Image

************, **** ******** **** intervention *** ****** ** each ****** ***** *** test **** ******, *** operators **** ********* ****** ten *******. ***** ** an ******* ** *** live ******** **** **** tested.

Event ************* *** ** ******* / ******** ******** ****

****** **** ***** "***" systems, **** ******** ** fully *********, **** ********* receiving *** ********* ***** event, ** *** ** dispatch ** ********. ******* of ****, *** ***** by ******* ** *** receive ************* ****** ** two *********:

  • ********* ********** ** **** to * ********** ******
  • ****** **** ********

***** *** ********** ******* notifications ** ****** *** vehicle ******:

IPVM Image

*******, *** ** ******** false ******, ********* *****, users *** ****** ** find ***** **** ************* annoying ** ***********.

Many ***** *********

**** ******** *********** ******* nuisance ****** **** * number ** *******, **** the **** ****** ***** animals *** *******. *** example, * ********'* *** was ********** *** ******** detected ** * ****** in ****, ********** ** event **** ****.

IPVM Image

***** ******* **** **** common ******* ** ***** positives, **** ****, *****, or ***** ********** **** as ****** ** ********.

IPVM Image

*******, *** **** ****** source ** ***** ********* was ******* ** *** scene ******* ** ******* branches ** *****.

IPVM Image

**** **** ***** ********** were **** **********, ** well, **** *** ** the ***** ******** ******* 98% ********** **** ***** objects **** ****** ****** or ****:

IPVM Image

Installation ******

**** ******** ******* **** be ******* ********, **** no ****** *** ****** use (********* **** *** monitor ****** ******* *** users **** ** ******** to ******** ****). **** means **** ***** ******* to ***** ***** ***** door, *** *******, **** mount * ****** ******* their ****, ** *** exterior *** *****. ** contrast, ********* *** ******* such ** ****, ****, or **** ***** ***** to ****** ******* *******, where ** *** ** more ****** ****** ** a ***** ** ***** or ******* **** *******.

Outdoor ******** ****** ********

************ ** **** ******** may ** ****** *** many ***** **** ***** hand *****, ** **** trim ** ***** ********* for ******** *** ******, which *** ** **** with * ****** ***** (shown *****), ***** *** generally ** **** ** hand, ******* **** * drill.

IPVM Image

*******, *** ***** **** brick, *****, ******, ** other ******** *********, ************ may ** ****** ***** capabilities. ************ ** *******, *****, or *****,*** *******, * ****** drill ** ********* ********, which **** **** ***** are ******** ** *** or **** *** ** use. ************, **** **** be ***** **** ******** into ***** ******** ** as *** ** ****** them ** ******* *** hole, ********* ** ******* pulling *** ** *** wall.

**** ** **** ********'* own ************ *****, ******** the ****** **** ****** above * ***** **** does *** ****** ****, with *** ********* ********* drilling ***** *** ****, and *** ******** ******* onto *** ******.

IPVM Image

*** ******** ** ******** **** ********-***** ********* ******* ***** remedy ***** ************ ******, as ***** ******* ** pay $**/***** *** **** Sentinel *** **** ****** to ****** * ***** installation ***.

Battery **** ******

*** ******** ******* **** of **** ****** ** two ****** ********* ** Deep ********. **** ***** a ********* ***** **** to ****** *** ******* of **** ****** *** times **** **** ** risk ******* ** **** of *** ***** *******, with ** ***** *** of ***** ******* ******* falling ** ****** ******.

** *** *****, **** the ******* ** *** about * **** ******* at *** ***** ****/****, battery **** *** **** to ~**%, ***** ***** mean ******* **** **** frequently **** *** ******.

IPVM Image

********

*** ********* ******** **** used ****** *******:

  • *** *******: *.*******

Country ** ******

**** *** ******* *** hub *** ******* ** "Assembled ** *****":

IPVM Image

IPVM Image

Comments (73)

It looks like that installer is drilling into his hand.

Agree: 3
Disagree
Informative
Unhelpful
Funny: 11

...and not securely holding onto the ladder.

yeah, but that ladder isn’t going anywhere, since it’s securely leaning on the in-swinging front door...

Agree: 1
Disagree
Informative
Unhelpful
Funny: 15

Perhaps JB is a secretly super tall insTALLER.

Agree
Disagree
Informative
Unhelpful
Funny: 7

He's not just the installer, he's the CEO.

Agree
Disagree
Informative
Unhelpful
Funny: 8

$50/month monitoring is unlikely to be sustainable/profitable, even if they perfect the analytics. That is an exceptionally low rate, particularly for residential areas that tend to be high activity (eg: neighborhood kids running around, meter checkers, lawn or pool maintenance people, package deliveries, and so forth).

Agree: 11
Disagree
Informative
Unhelpful
Funny

We spoke with them and asked exactly about that, i.e., how can you make money if your operators are checking dozens of nuisance alarms per customer per day? They say that is built into their model and while they will not have the gross margins of conventional security providers (e.g., ADT), that they see it as a competitive advantage and moat against rivals entering this space (i.e., they are well funded and don't need to worry about being cash flow positive for many years as long as they grow).

We had 30+ alarms per day, over a month that is 1,000 total. Say 15 seconds to check (not sure), that's like 4 hours but they say operators are checking many alarms simultaneously and maybe at less than 15 seconds, so 1 man hour per account per month, maybe 30 minutes per account per month?

They definitely will need to get the unit economics right and need to improve (i.e., reduce) alarms that they need to monitor.

Agree
Disagree
Informative: 3
Unhelpful
Funny

They say that is built into their model and while they will not have the gross margins of conventional security providers (e.g., ADT), that they see it as a competitive advantage and moat against rivals entering this space (i.e., they are well funded and don't need to worry about being cash flow positive for many years as long as they grow).

You must have been wearing a Patagonia fleece when talking with them, because they clearly thought you wanted the VC answer.

It is not just operator costs, they're going to have a fair bit of AWS/Azure/Whatever costs with a bunch of chattery cameras sending notifications of cats and shadows.

They said their monitoring center is in CA, which has a minimum wage of $11/hr currently. They will likely need to pay people a decent premium above that to attract and retain the kind of person you need for a good monitoring operation, their grossed-up costs are likely to around $20/hr per person.

For false alarms, they can probably dismiss them within a few seconds, you don't need to watch the whole clip to see that it is a cat or a shadow. But some others they will have to watch 2 or 3x if it is shadowy and they can't quite tell what is going on. Assuming 7 seconds per clip on average, and 1 10 minute break per hour (these jobs can be very tedious and tiring), 1 operator can process about 400 clips an hour.

At $50/month with 900 clips in total, plus some AWS costs, and they're probably losing money. 

Enhancing the analytics quality would certainly help, though I question how much processing power they have on the camera. They could do some analytics processing in the cloud, at the cost of a slight delay and more AWS overheard.

 

Agree
Disagree
Informative: 2
Unhelpful
Funny

You must have been wearing a Patagonia fleece when talking with them, because they clearly thought you wanted the VC answer.

Lol, I am reporting what their claim is.

Assuming 7 seconds per clip on average, and 1 10 minute break per hour (these jobs can be very tedious and tiring), 1 operator can process about 400 clips an hour.

One particular claim they made is that operators are watching multiple clips simultaneously. I am a bit skeptical about how well that will work but that is the claim made.

Right now, 90% of their alarms are likely cats, dogs, shadows, etc. which are easy to review / reject. 

You make a good point and I agree about some clips requiring significant far more direct attention, i.e., is it in an intruder or is it the homeowner, etc.?

Agree
Disagree
Informative: 1
Unhelpful
Funny

One particular claim they made is that operators are watching multiple clips simultaneously. I am a bit skeptical about how well that will work but that is the claim made.

Fair point, however you also need to factor in the clips where they actually need to do something, particularly the valid alarms, which probably involve:

1) Watch full clip

2) Do audio talk down warning

3) Call police, relay info, log police response or comments.

Valid events could easily take them 2+ minutes to fully handle. Granted, they shouldn't have too many of those overall, but every 1 real event takes the processing time of ~17 ignorable events.

They'll also have the overhead of management layers to handle the monitoring employees, and some related expenses (office space, internet connectivity, etc.).

To me, the financial model looks like it breaks down with even just casual analysis, and the more you expand the analysis, the worse it looks.

Residential monitoring for alarm systems is about $10/mo for subscriber-owned systems (eg: not the ADT model where they "give" you a junk system and charge $50/mo for  3 years). These systems are generally very low volume events, have minimal overheard in terms of network costs and so forth. Companies offering these rates are making good-not-great profits.

Deep Sentinel is charging 5x the general street price for alarm monitoring, but probably has 50x the overhead comparatively speaking. I'm not convinced that some analytics improvements or related efficiencies of scale are going to fix that.

I do agree they have a "moat" against rivals in the sense that the investor tolerance for this kind of money-losing approach seems to be declining in recent years.

Agree: 1
Disagree
Informative
Unhelpful
Funny

Deep Sentinel is charging 5x the general street price for alarm monitoring, but probably has 50x the overhead comparatively speaking. I'm not convinced that some analytics improvements or related efficiencies of scale are going to fix that.

They might not. In the meantime, seems like a worthwhile experiment to run. If they want to lose money having human beings look at someone's house 40 times a day, so be it. Eventually, they will either improve it or give up.

I am a bit more bullish about their ability to improve. I would be shocked if they do not change to a model of off by default, on by choice. This alone will radically reduce battery usage and operator expenses.

They mentioned they started with a "STOP" protocol with the assumption that everyone was a threat and that they now have moved to a "HELLO" protocol, reflecting the reality that very rarely are adversaries attacking houses. My point in calling this out is that they are obviously still learning about the industry and the use case, so I expect them to make such optimizations.

Will it be enough? I don't know. But it's good for the market for companies trying new approaches.

Agree
Disagree
Informative: 2
Unhelpful
Funny

I agree it is very low based on typical industry pricing and metrics.  However, if the management at Deep Sentinel operate this startup like Amazon then I would not be surprised if they run a loss (or break even) for an extended period of time.  Jeff Bezos has a new market mantra; " your margin is my opportunity".   Bezos has proven he will sacrifice profit today for the long game.  So if Deep Sentinel does as Bezos has shown he is willing to do in other ventures, loose money or break even to gain market share, all the while working to perfect the AI, then I expect this could be very disruptive.  Just because the offering is weak now does not mean it will remain that way.

I'm curious about others thoughts on this? 

Agree: 1
Disagree
Informative
Unhelpful
Funny

Chris, well stated and thanks for your first comment. That is roughly my thoughts as well. Amazon certainly does not succeed at everything, so this may not work out, but if they give themselves enough time and money, they have a chance to build a new model for home security monitoring that utilizes direct sales and video analytics (a combination no one has ever spent serious money on yet).

Agree
Disagree
Informative
Unhelpful
Funny

Do the monitors call your cell phone when they see something to verify that it is you and not the bad guy?  Does it give a notification that someone is watching?  Is the monitoring here in the US or do they outsource it from the US like other low quality groups?

Agree: 2
Disagree
Informative
Unhelpful
Funny

Do the monitors call your cell phone when they see something to verify that it is you and not the bad guy?

If the monitors are not sure if it is you on the cameras they are supposed to speak over the cameras speaker to confirm.

Does it give a notification that someone is watching?

Someone is always watching (checking the notification for suspicious activity), when they confirm there is nothing malicious they put a verification on the notification:

Operator Verified Notifications Within 10 to 20 Seconds of Receiving

Is the monitoring here in the US or do they outsource it from the US like other low quality groups? 

According to their tech support, the monitoring is done in the U.S. in Pleasanton California.

Agree
Disagree
Informative: 2
Unhelpful
Funny

How do they know who you are?  Do they register your face?

Agree: 2
Disagree
Informative
Unhelpful
Funny

They have a learning period after the cameras are put up where they are supposed to learn who the residents of the house are and the normal traffic.How long this is, I am not sure of. I have yet to have an operator mistake me for someone that doesn't belong at the house.

Edit: We have asked Deep Sentinel to further clarify how they know who the residents of the house are (e.g. do they keep images of everyone? facial recognition?). Will update when they answer.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Deep Sentinel says they while they 'remove all personal identifiers', they have 'secret sauce' behavorial analytics to determine whom the homeowners are:

During monitoring- Our Live Sentinel Center does have the customer profile of the homeowner, i.e. to contact them in case of an incident. Our Center also has an awareness of the homeowner, but without the need to keep photos or perform facial recognition, this is part of our secret sauce, and is behavioral based.  To our end user, they should feel safe their home is being monitored by live (human) guards 24/7, and the technology we are developing to scale to enable this monitoring should be transparent.  

R&D- This is where I was referencing we remove all personal identifiers. We assign a unique event ID to each recording, and use this to continually train our data models. 

I can't figure it out. You can't depend on face or even size of the person, since (1) face is often not captured and (2) many people are of similar sizes and (3) clothing changes. I guess you can make predictions based on how someone behaves, like running vs walking calmly but you risk false positives depending on if an adversary 'behaves' 'normally.

Agree: 1
Disagree
Informative: 1
Unhelpful
Funny

Secret sauce my behind. The cameras have pir in them, and the rest it likely done by human. They’ve bullish!ted with their claims so far...doing the arms in the air got me 60 seconds if not more. There is no way that they will be able to sustain this operation in California seeing that the burger flippers all want $15/hr+ and they’re charging 50/mo. And we all know no ex cop is going to watch cameras at $15/ hour...yet they claim all their monitoring agents are ex cop or military. How do they do it when it’s 7am and everyone’s leaving in the morning and every camera is coming in? For all I know they had Arlo special manufacturer a white label system for them and they’re using geovision software or immix to monitor it. Monitoring residential cameras is too much a gray area as to who is supposed to be there vs who is not. And there’s no money in it.

Agree: 5
Disagree
Informative
Unhelpful
Funny: 1

I saw what appears to be a PIR but this might be a first line to trigger a video clip the actual Analytics can be processed and classified on the monitoring side. The company I use called VideoArmed uses a analytics loaded onto a NVR but they also have that second layer  

Agree
Disagree
Informative: 1
Unhelpful
Funny

An operator response time for 10-20 seconds is very impressive. I would be curious to know what their standard operating procedures were and how many alarms they have each operator handling. 

Agree: 2
Disagree
Informative
Unhelpful
Funny

I don't know about this.  I see the negatives as far bigger than the positives.  And when you call yourself "Deep" anything and tout your AI, yet your product identifies cats as humans, I call B.S.

Agree: 3
Disagree
Informative
Unhelpful
Funny

I call B.S.

Verified Amazon Review:  First I called D.S., then I called B.S.”

Agree
Disagree
Informative
Unhelpful
Funny: 3

I guess is that the analytics does not at this point know what a cat is but with time it will learn and classify properly. I don't think the analytics has to be in the camera or NVR i am pretty sure it can be in the monitoring server and this may still be learning. Again I am not a developer but from what i understand this is a method 

Agree
Disagree
Informative
Unhelpful
Funny

I think this will be short lived. I was not impressed what so ever with their response time, and how their "AI" determines a threat.

Its "cute" for the porch pirates, but all it takes is a bag to cover it and its use less. The wireless signal is nothing great. Anyone notice how they mute all comments on their FB ads?

I bought the system to test myself, and Ive called them out on events that my ring cam will catch and not this...IE this system is guaranteed to miss a "smash and grab" type of incident.

But hell, while I have it, it will be amusement when I walk in the back yard naked bahaha

Agree
Disagree: 2
Informative: 1
Unhelpful
Funny: 1

I bought the system to test myself, and Ive called them out on events that my ring cam will catch and not this...IE this system is guaranteed to miss a "smash and grab" type of incident.

#7 thanks. Can you elaborate on what events Deep Sentinel missed? Are you saying someone ran into the FoV of a Deep Sentinel camera, 'smashed and grabbed' something and Deep Sentinel generated no alert at all?

Agree
Disagree
Informative
Unhelpful
Funny

It recorded it but never sent it off for agent review. It was a door to door salesman that could have done a smash and grab

Agree
Disagree
Informative: 1
Unhelpful
Funny

It recorded it but never sent it off for agent review. It was a door to door salesman that could have done a smash and grab

So if it recorded the event, did you get a person notification? How do you know that it didn't send the notification? Was the event on your timeline but just not reviewed? Did you have the system in privacy mode?

Agree: 1
Disagree
Informative
Unhelpful
Funny

I don’t remember the main icon, it was either a person or a leaf, but I remember the smaller icon as an eye that was crossed out.

Agree
Disagree
Informative: 1
Unhelpful
Funny

system was being monitored because when the ring cam picked it up, I was around the corner from the house and stopped home to see what they did.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Please follow up on the battery life.  2 months expected, 1 week @ 53%.  Changing the batteries every 2~3 weeks is a deal breaker, especially if a ladder is involved.

Agree: 4
Disagree
Informative
Unhelpful
Funny

Deep Sentinel acknowledges that some cameras will need to be recharged even more frequently than 2 months. From their FAQ, currently:

Many cameras won’t need to be recharged for up to two months, but you will eventually have to re-charge them. Based on the camera location and the traffic in the area, each camera will need to be recharged at different times. As expected, high traffic areas use more battery power as the camera turns on more often. Batteries last a couple of years; eventually you will need to change your batteries.

Note, previously, their FAQ said up to a year but they changed it after we asked about it during our test.

Yes, agree with your concern about frequent battery changes.

Agree
Disagree
Informative: 1
Unhelpful
Funny

If AI accuracy (I want to say VA) is resulting in so many false positives (i.e. camera turning on), then battery life expectancy is likely going to be a lot less then 2 months.  Time will tell of course. 

Agree: 2
Disagree
Informative
Unhelpful
Funny

Maybe they will have a service where the CEO comes out with his ladder and drill to change the batteries every month or two.  They could get an RV and make a road trip out of it. 

Agree
Disagree
Informative
Unhelpful
Funny: 2

This is for sure the greatest issue but I think they are looking for a DIY and maybe they will require a installer to wire a system or maybe solar might help is certain locations.

Agree
Disagree
Informative
Unhelpful
Funny

Having the system start by running everything through an operator with no significant AI and absorbing the increased operator cost is a good move IMHO, over time, as their analytics learn and get more sophisticated they can gain efficiencies later, but missing alarms in the early days would very hard for the product to overcome later on. 

The battery life seems like a killer to me, kind of like the VCR blinking 12:00 problem. The first time you were highly motivated to set the time on your new toy, but by the 5th or 6th time you started to think 'who cares?'. If you have to drag a ladder out and it's a PIA to change batteries, how many cycles with no real alarms will it be before you start to put it off?

Agree: 4
Disagree
Informative
Unhelpful
Funny

My guess is they have a team of people sitting around waiting right now and as soon as an alert comes in they jump on it, which is why it only takes 10-20 seconds.  If their customer base grows they will quickly become overwhelmed and response times will grow to where the bad guy will be done and gone by the time they are reviewing the footage.   Not only that they clearly have a training problem if Justin classified shadows and goats as people while Marissa thought a cat was a person, as they scale up turnover will accelerate and training will suffer even more.

I cannot understand how this company can be sustainable, the only way I see them succeeding is to follow the Telsa/Uber/Netflix model and take on massive amounts of debt.

Agree: 2
Disagree
Informative
Unhelpful
Funny

Ooops...Deep Sentinel forgot to tell their customers that most customers will pay fines/fees for the mistakes of the “Surveillance Center”.  Deep Sentinel claims no “false alarms”, which is different than UPR-Unnecessary Police Response, but same fines/fees. Could be lots of customer costs from UPR without false alarms.  My comments sound like doublespeak, but are not.

 

Agree
Disagree
Informative
Unhelpful
Funny

Could be lots of customer costs from UPR without false alarms.

So Lee, if DS notifies the PoPo after this talk down incident, is that a UPR?

Agree
Disagree
Informative
Unhelpful
Funny

Where are the time stamps for the video? With the jump-cut/edit there is no way to determine if the response time was 5 minutes, or 5 hours... I am not trying to detract from what DS did. The talk down feature was probably what dissuaded the would be thief, but without the time stamps this video becomes a lot less impressive. 

Agree: 2
Disagree
Informative
Unhelpful
Funny

Lee, I don't get why this is a problem for Deep Sentinel. The whole point is that they are visually verifying the alarm before calling the police.

To the extent that Deep Sentinel has 'false alarms' that is a problem for their monitoring center and their cost structure, but it is only a problem with the police if Deep Sentinel actually calls them with a 'false alarm' which they likely will not since they are spending the time and money to verify it first, yes/no?

Agree: 1
Disagree
Informative
Unhelpful
Funny

Heres something interesting. In your snapshots, Marissa M, and Justin F are very common names that look at my cameras...makes me question camera to agent ratio. Heck, Justin F looked at me in my back yard just last night.

Agree
Disagree
Informative
Unhelpful
Funny

I'm responding to comments by Chris Cunningham, Mark Palka, and John Honovich.

I think we need to also consider the fact that Amazon AWS is strongly invested in AI and deep learning, including as it applies to video. Look at Amazon Rekognition and AWS DeepLens. We also need to consider that deep learning is multi-dimentionally much more than a traditional analytic algorithm. See my recent article in Security Business on AI and deep learning, in which I only scratched the surface on the deep learning aspect. The big thing about Deep Learning is the training aspect. The bigger and better your sample of training video or photo images are, the better the results your deep learning analytics will be.

Consider the #10YearChallenge on Facebook, Instagram and Twitter, which - while they assert they are not an originator of the meme - provided Facebook with an amazing library of 10-year apart facial images with which to train deep learning facial recognition software. As the linked article mentions, this has potentially great beneficial uses as well as nefarious ones: "Last year police in New Delhi reported tracking down nearly 3,000 missing kids in just four days using facial recognition technology."

Once Sentinel achieves a significant customer base, its service could become both a training ground for its deep learning video analytics at the same time as being a provider of highly-trained AI services whose training samples have been "human validated" to a scale that's otherwise hard to achieve.

Is it a coincidence that Deep Sentinel was founded in 2016, the year that Amazon released its video facial reconition services whose customers include law enforcement? 

We'll see how this all unfolds, but there are more technology and monetary dynamics at play in this startup (typical for today's digital age) than for traditional startups in this arena.

Agree: 2
Disagree
Informative
Unhelpful
Funny

Consider the #10YearChallenge on Facebook, Instagram and Twitter, which - while they assert they are not an originator of the meme - provided Facebook with an amazing library of 10-year apart facial images with which to train deep learning facial recognition software.

This theory makes no sense. Facebook has already been around for more than 10 years, with people uploading pics of themselves as profile pics all along. I don't think Facebook needed anyone to specifically tag selfies with a 10 year delta.

Is it a coincidence that Deep Sentinel was founded in 2016, the year that Amazon released its video facial reconition services whose customers include law enforcement?

Probably.

 

Agree
Disagree
Informative
Unhelpful
Funny

Re: Facebook, it is the metadata tag of 10 years and the hiqh quality facial representation of a large number of image sets that makes for a higher-quality deep learning training set regarding aging, than would otherwise be available. How big the incremental quality improvement would be is hard to say, but definitely a low cost way to obtain a better-qualified set of facial images for that particular AI purpose. It may, for example, help them better facially-identify people in group pictures when there is an age difference between the profile photo and other individual photos and the group picture.

Also, an anonymized set of AI facial modeling data would be of monetary value, and could be utilized, shared or sold without violating any individual's privacy. It's now more about the digital data models than it is about the bit-mapped images.

Re: Timing coindicence, I agree that it's likely a coincidence, but also a factor that would likely have at least a little weight as a favorable factor. 

Agree
Disagree
Informative
Unhelpful
Funny

I think that like many others with the 10 Year Challenge thing you are trying to back port an explanation into a stupid Facebook trend. Even with Facebook's generally poor security model, this theory runs into all kinds of issues with people who have limited who can see their posts and so forth. It is also far too out in the open if the originator's purpose were to gather some kind of training set data.

Agree
Disagree
Informative
Unhelpful
Funny

With high maintenance a customer problem, is the business model's secret sauce that as customers bore with battery replacement, alerts will settle down as  cameras wink out. 

It is not clear if the "AI" is getting feedback from the operators and the training is being stored at the edge or at servers at the station.

The videos of operators responding looks like immix is the automation software.

The motivation behind this product support is package theft, yet it is unlikely it will help. This is a much more complicated problem than they seem to understand - if they had spent some time investigated the commercial outdoor video industry beforehand they could have realized this solution will not be scalable. The price point is too low for a quality result, and the customer is not feeling enough pain to justify increasing their fees.

 

Agree
Disagree
Informative: 1
Unhelpful
Funny

The motivation behind this product support is package theft, yet it is unlikely it will help.

Robert, good feedback and I agree that the marketing is fairly heavy around package theft.

From what I've seen from our test, I would promote this for loved one protection - i.e., you are not at home but your spouse is at home, your kids at home, your elderly parents are at home and you are worried that something bad will happen, most particularly someone breaks in. That Deep Sentinel does have operators standing by and they have the video to immediately review and audio to promptly respond is worth $50 per month to a lot of people.

But maybe package theft is more significant than I imagine...

Agree
Disagree
Informative
Unhelpful
Funny

John, im currently pitching my idea to residential home owners with a similar service to deep sentinel and i approach it from the angle you mentioned about protecting loved ones

Agree
Disagree
Informative
Unhelpful
Funny

...similar service to deep sentinel

how similar?

Agree
Disagree
Informative
Unhelpful
Funny: 1

We are currently doing this with another company we add the analytics to any ip camera via a vpn box essentially so we dont have port forward anything for security reasons. We can also add sirens we can manually sound a prerecorded message or speak through it. We just use the prerecorded message much easier. We can proactivley monitor the alerts and notify the policy if need be call them. We currently use humans to look at every alert. We are in the process of switching to a new analytics software that has very few false alerts that we are currently testing in my homes and other clients homes. That will send the alert directly to the monitoring station but we cant do it for 50 a month. We are thinking of chargin between 55 to 75 per camera per month. Currently we charge 100 per camera per month.

Agree
Disagree
Informative
Unhelpful
Funny

That will send the alert directly to the monitoring station but we cant do it for 50 a month

Glen, I think the Deep Sentinel approach is to eventually get to $50 per month being profitable but fund it via venture capital until the business scales to that level.

Charging more now I think will work in the short run but I suspect Deep Sentinel or some other venture-backed video alarm monitoring approach will eventually become the largest provider in this space.

Agree
Disagree
Informative
Unhelpful
Funny

John, 

I agree with you theres a market for that and with venture capital they can definently make it profitable and work if done correctly. 

Agree
Disagree
Informative
Unhelpful
Funny

Glen, being in the residential space myself I'm curious as what you are pitching if your willing to share.

Agree
Disagree
Informative
Unhelpful
Funny

I pitching the idea like John mentioned. I tell people of the the new service we are going to offer where we monitor the outside of the property for them while they are home or gone to work to proactively take care of their family for them. And i explain an alarm only soumds when some one already broke in and with our system we can give them advance notice of a potential breakin if they are inside the home. I tell them the can arm the cameras when they are asleep or awake during the day and are not going in and out of the house. One home owner said she got scared because a man was outside her house in the backyard. A few minutes later she found out it was the cable guy. But to her it was a scary moment and said she is willing to pay for proactive video monitoring. We currently offer this for commercial properties during the night only and hopefully im the next 2 months we can start with residential. Right now ibam gathering local data on how much people are willing to pay and if theres a need.

Agree
Disagree
Informative
Unhelpful
Funny

I think I will not bet against the 2nd richest person in the world and assume he and his resources have done a lot of homework before investing.

Not that he can't fail but that I'm guessing they know what the end goal is and how their going to get there and of course have the resources to weather any storm of inaccuracy in their plan.

Agree
Disagree
Informative
Unhelpful
Funny

The "Bezos" association is a bit overblown. Bezos is not listed as a personal investor, rather his investment fund Bezos Expeditions  (a team of people that manages his personal money for investments) is the investor in Deep Sentinel. They don't even list DS on their website:

Granted, this could just be that they haven't updated the site yet, but I do not think Jeff Bezos is personally highly involved with DS and pushing them to glory.

The lead investor listed for Deep Sentinel is Shasta Ventures, meaning they have more invested than the Bezos fund, and thereby control more shares and exert more influence on the company, at least for the time being.

Investors make speculative bets all the time, and you never truly know where the "end goal" is at the beginning.

Agree
Disagree
Informative
Unhelpful
Funny

I do believe Jeff Bezos was personally involved in his investment in Deep Sentinel. The founder and CEO of Deep Sentinel David Selinger created the data science team at Amazon back in 2003.

Here's a picture of them together in a personal setting around that time: 

 

Agree
Disagree
Informative: 1
Unhelpful
Funny

Arlo has been unable to exceed a 5% paid subscription rate across their users. The DIY model is crippled by the preference of DIYers (in general) to opt out of paid subscriptions. I think it's easier to fit video verification for $50/month into the pro install/monitoring model than to get DIY folks to pay $50/month.

Agree: 1
Disagree
Informative
Unhelpful
Funny

Context for Arlo, from their most recent 10K:

It is currently at ~5% though modestly trending up and subscription services are a later addition. I agree with your general point, though they may get to 10% in a few years.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Small update, Deep Sentinel says:

We also did extend the actual average out a couple weeks with tweaks, so many people are getting more like 3 months from their cameras now.

We are not going to retest it for that but we will keep an eye out for more fundamental changes as they occur.

Agree
Disagree
Informative
Unhelpful
Funny

The idea is great and i currently offer a similar service for commercial properties were we install professional grade equipment both inside and out and we can monitor the cameras for them using motion. However we have a lot of false alarms and currently charge $100 per camera per month. We are in the process of switching to artificial intelligence which has a very accurate capture rate that reduces false alarms. At $50 a month i would close, we need a minimum of $75 to maintain a decent positive cash flow. We are in the process to offer it to residential homes but will be charging $75 per camera our current surveys shows home owners who can afford it will be willing to pay that much.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Glen, who are you using for AI? Ive tinkered with a few...I just dont like to pay for a service that is essentially software on someone elses hardware.

Agree
Disagree
Informative
Unhelpful
Funny

Im in the process of switching AI providers for a better one but its not ready yet for video surveillance monitoring. Call me and i can talk to you about the pros and cons of the one i am currently using and why i am switching. My number is 210 996 1686 you can text me for a time to talk. I dont want to say anything percieved as negative on a open forum.

Agree
Disagree
Informative
Unhelpful
Funny

I dont want to say anything percieved as negative on a open forum.

that’s understandable.  

however I would add that the IPVM discussion facility is a member-only forum, (unless otherwise noted), and relies upon candid informed content as it’s raison d'être.

Agree
Disagree
Informative
Unhelpful
Funny

I currently use provigil for the analytics the issue i am not comfortable with is that they outsource their video monitoring to India due to cheap labor and relies heavily on motion detection. The other issue we are limited who we can sell to. Right now we are only able to have clients monitored who are closed all night with clearly defined business hours. We can't sell to residential, hotels ore other business that can benefit from this kimd of monitoring. We also have limited control over the analytics since they want to protect their intellectual property. We are currently working with ROG Security who created an AI thats was inted to be used in the guard industry. To reduce the workforce needed to patrol large properties. We are currently working with them to get their software integrated with immix cs so a monitoring center can receive the alertsnin real time. It was really good and ROG does not limit who we sell to and we are free to sell it both to commercial and to residential. While they are open or close so ot opens or market to a wider audience.

Agree
Disagree
Informative
Unhelpful
Funny

Coronavirus 40% Price Increase

Copies of a Deep Sentinel price increase letter are circulating online, stating a hefty 40% price increase due to 'limited inventory, combined with cost and tariff increases' on both equipment and monitoring for new accounts:

The memo states price changes take effect on April 1st.

At the current $499 for 3 camera and hub kit/ $49 month monitoring, a 40% increase is ~$700 hardware and $70 month monitoring.

Agree
Disagree
Informative
Unhelpful
Funny

Brian…. We believe there is a large and stable market for Deep Sentinel, even after the price increase. Like Sonitrol, they specialize in verification technologies at both ends, monitoring and customer. Sonitrol focuses on security inside the property, whereas Deep Sentinel focuses on security outside, pre-entry the property. Key word is “verification”, with end-to-end responsibility for the customer… they both have high credibility with law enforcement, which is what security customers are buying. Most of the existing remote-monitored customers in the US are “deterrent” systems, not “security” systems, without verification at both ends, serviced by “third party” subcontractors that are handicapped with low credibility with local police due to the near 100% error/false alarms, for slow or no response. High attrition can be expected for lower revenue deterrent customers, whereas low attrition with higher revenue verification customers. Deterrent customers are the target of the muscular DIY, MIY, DIFY providers.

Agree
Disagree
Informative
Unhelpful
Funny

Remember the debate last year about how they can charge $50 per month? Evidently they can't. Pricing is now doubled to $100:

IPVM Image

Deep Sentinel phone rep says the pricing has been $100 for some time.

That obviously changes their positioning. For people who really care about having a 'livesentinel' probably still worth the $100 but for those undecided / in between, Simplisafe et al at ~$20 is likely good enough.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Speaking of which, I need to cancel my subscription...too lazy to change the batteries in the cameras, and it was hit or miss if they worked in the hot heat.

Agree
Disagree
Informative
Unhelpful
Funny

John... Lots of us agree.

Another basic lesson for the good-ole-boys in the traditional alarm industry. It is not the price, it is the product/service. In this era of $100 Cell bills and $100 Cable bills, $100 can be fair value for the confidence of family/household/business “security”. Key word… security. Millions of RMR deterrent alarm customers are now paying $40-60 monthly for a technology package that is outdated by many decades… with an expectation of “magic security”… but are not told by their supplier that their system does not qualify for public or private police interaction or site-response.

Agree
Disagree
Informative
Unhelpful
Funny

Which begs the question "What can I offer my customers that's better?". We know we'll find customer who are willing to pay the price.

So how could we duplicate this on a smaller scale for our own businesses? Does DS have a minimum-wage person sitting in a room in front of 9 monitors all each with 9 cameras glancing over each and then reading a screen of AI notifications? Or do they just have someone in front of same screens waiting for notifications and then they are pulled up on said screens?

With the capabilities of today's VMS's couldn't we duplicate that on a small scale for our highest priority (correlated probably as paying) customers?

Agree
Disagree
Informative
Unhelpful
Funny

So how could we duplicate this on a smaller scale for our own businesses?

You might be able to. One of the things that these remote video monitoring companies all try to optimize on is video analytics performance. The idea is to use technology to reduce labor costs and thereby make a profit at a more reasonable monthly fee, e.g. alarm companies using Calipsa.

The potential I still see for Deep Sentinel is the pursuit of scale. Being big will give significant competitive advantages in both lowering marginal operational cost and customer acquisition cost. That said, I still don't know if Deep Sentinel can succeed but they at least have money connections to help them that most don't.

Agree
Disagree
Informative
Unhelpful
Funny

So now that they've figured out $50/mo wasn't going to be sustainable, I wonder if the $100/mo number is based on math, or more hope.

--UD2

Agree
Disagree
Informative
Unhelpful
Funny
Read this IPVM report for free.

This article is part of IPVM's 6,912 reports, 924 tests and is only available to members. To get a one-time preview of our work, enter your work email to access the full article.

Already a member? Login here | Join now
Loading Related Reports