Aegis AI

Problem: Many organizations are seeking low-cost and inconspicuous ways to protect their occupants against gun violence. They often provide extra protection by installing camera systems that monitor key areas. However, security footage from cameras are rarely or never monitored due to budget constraints.   

Solution: Aegis' computer vision software turns any security camera into a gun-detecting smart camera, eliminating the need for human monitoring and alerting key personnel when a threat is detected. Upon detection, the system alerts customer decision-makers, who then review footage and call the police if a firearm is confirmed.   

Impact: Our solution reduces the time it takes to notify police to less than 10 seconds upon gun detection. Upon arrival, police also have access to a visual and the current location of the shooter - information may not have using traditional security measures.

During a mass shooting, bystanders’ priority is getting to safety, not calling 911. Police don’t learn of most shootings until 5 minutes after the trigger is pulled, and it takes squad cars 18 minutes to arrive. 60% of mass shootings end before police arrive. Moreoever, police often have ambiguous information about the incident (i.e. visual/description or location of shooter) upon arrival, impending their efforts to neutralize the threat.  

The odds of any given school experiencing a gun violence incident is 1 in 1,400 per year. This is a serious problem, but not one that schools can invest millions of dollars in mitigating. The most common way that our customers think about providing extra protection is by installing camera systems that monitor key public areas. However, due to their budgets, they can only afford to hire one security person to monitor 100-200 cameras, or they do not monitor the cameras on a regular basis, resulting in an underuse of expensive security hardware.

The three markets Aegis is looking to target are: education, commercial and public venues.

The $2.8B education market is our initial focus segment, due to schools’ understandable focus on gun crime prevention. School leaders are highly concerned and enthusiastic about gun violence mitigation technology, with most customers telling us that school safety is their number one priority.  

The commercial segment has a $4.7B market size and is increasingly concerned with gun violence due to potential armed robberies, recent shooting incidents involving disgruntled employees, or violence against media companies. There is also increased concern about legal liability by building managers or owners when a violent event occurs. Despite the sizable market, overall there is less focus on gun violence among commercial buildings than in the education segment, making commercial customers a secondary focus for Aegis’ initial sales effort.  

Public venues and infrastructure are a third priority currently. Public spaces such as public transport stations have a large number of cameras per location (500-5,000 units) and have strong demand for violence prevention. The segment has a market size of about $1B.

The Aegis solution automatically detects firearms in your existing security camera feeds, providing early warning and dramatically improving law enforcement response.

Aegis’ technology is based on deep learning-based computer vision algorithms with a model that recognizes guns in frames from surveillance cameras. Once a weapon is detected, the system can alert any stakeholder the customer specifies in almost any way: Text, email, or phone call. Alerts go out within 1 second of the system receiving an image from the customer’s VMS.  

Our unique advantages start with our Model Structure. We have built a proprietary deep learning neural network model that is inspired by industry-leading algorithms. We have customized the model for identifying small, dark objects in complex scenes, optimizing it for gun detection. These changes are completely proprietary because they are based on internal research and allow us to out-perform top industry benchmarks on single-object detection.  

The dataset we use to train the model is completely proprietary and building it from scratch was necessary for performance. Most computer vision models are based on web-scraped datasets that provide ~80% accuracy, which is not acceptable for public safety deployments. To improve performance and accuracy, we crowdsourced data and spent months generating photos of ourselves and friends with weapons in specific situations to plug the gaps in our model and publicly-available datasets. This effort resulted in 300k+ labeled images and has allowed us to achieve greater than 99% accuracy in weapons detection with roughly one false positive per week in a standard deployment (20+ cameras).

Aegis’ technology is based on deep learning-based computer vision algorithms. The AI model recognizes guns in frames from surveillance cameras. Our deep learning model and its training data are proprietary and key in setting Aegis apart from competitors. Our system is built on a proprietary machine learning (ML) stack that runs on a cloud backend. The system pulls video feeds from customers’ Video Management Systems (VMS) via API, meaning no hardware or software needs to be installed at customer’s facility. The ML stack consists of three levels: Platform, Model Structure, and Data.

The ML platform is the underlying framework that processes our deep learning algorithms. We use a platform called Darknet (similar to TensorFlow). All ML platforms are similar, and all but the largest firms use an off-the-shelf platform with few customizations.

Once a weapon is detected, the system can alert any stakeholder the customer specifies in almost any way: Text, email, or phone call. Alerts go out within 1 second of the system receiving an image from the customer’s VMS.  

Our backend is based on Amazon Web Service serverless tools, with key AI elements running on Google Cloud. Everything is completely scalable and can support 1000+ cameras simultaneously. As such, future development effort is focused in three areas: Improving backend efficiency, accelerating customer deployment time, and adding additional detection features.

In an emergency situation - first responders arrive faster, you evacuate sooner, and you won't even notice that it's there until it's needed with the Aegis solution.  

Our solution offers:

• Improved responsiveness: It currently takes an average of 18 minutes for law enforcement to respond to active shooter events (Department of Homeland Security). Aegis AI dramatically improves the responsiveness of security, law enforcement, and emergency medical services. We provide first responders with real-time intelligence on the suspect, allowing them to neutralize the threat and render aid to victims much more rapidly than before.
• Immediate early warning: Your designated staff will be immediately notified with the camera location and the detected frame as soon as a weapon is drawn. This provides building occupants with the critical information to make immediate lifesaving decisions.
• Cost effective and non-intrusive: We don't want the places where we work, live, play, and worship to feel like TSA checkpoints. Aegis AI integrates with your existing infrastructure and there is no additional hardware required. That means that it is completely invisible and costs only a small fraction of hiring a single armed guard.

Currently, Aegis is implemented on more than 400 cameras across 9 customers (2 signed customers and 7 pilots). This translates to a coverage of about 5000 people.

In a year, Aegis is aiming to implement at 20 school districts and to cover about 100,000 people.

In five years, Aegis is aiming to implement at 1000 school districts and to cover about 5,000,000 people.

Aegis is currently implemented with 2 paying clients, has pilots with 7 clients and has a pipeline of 50+. Within the next year, we are aiming to convert pilots and leads into 25 to 30 customers. We also plan to pursue a Series A funding round and achieve the following metrics:

• 1M+ run-rate ARR
• Direct sales pipeline <5 month account executive payback period
• Annual contract values above $50k
• Unit economics: 50%+ GM, <12 month payback period
• High net promoter score and customer satisfaction

In the next five years, Aegis aims to be implemented at 1000 school districts. We also plan to have a ubiquitous technology that can detect all kinds of threats (i.e. intruders, abandoned objects) without invading privacy, making every public space safer.

The barriers to achieving our one-year goals are:

• Educating the customer base about the technology to the extend that customers are confident that this technology works, this technology is affordable and this technology can be implemented without infringing privacy.
• Connecting with the decision makers at school districts and aligning with the budget cycle at schools to close contracts with school districts

The primary barrier to achieving our five-year goals is:

• Competition from existing or new players that offer complementary gun detection solutions (i.e. acoustic gunshot detection, microwave radar technology, metal detectors) or alternative gun detection computer vision systems.

The plans to overcome the barriers to Aegis' one-year goal are:

• Educating customers:

  
  

  • Collaborating with a design agency to polish our messaging, presenting the Aegis solution in a way that is easy to understand by decision makers at school districts.
  • Offering pilots to school districts where the decision makers that see the Aegis solution seamlessly implemented and be assured that the technology is real, effective and non-intrusive.

• Deal closure at schools:
  • Hiring a sales person that specializes in selling at schools to align the Aegis sales process to the budgeting cycle at schools and in identifying and connecting with the appropriate decision markets at schools.

The plan to overcome the barrier to Aegis' give-year goal is:

• Competition:
  • Aegis plans to invest heavily into and prioritize the Aegis solution, ensuring that the product is developed and implemented with quality and that the product satisfies the needs of our customers. Additionally, most of the competitive solutions are complements to computer vision solutions and so Aegis foresees our technology being part of solutions with other technology, rather than displacing them

Full-time staff: 6

Part-time staff: 2

The team has demonstrated competence in building the AI platform, landing angel investment, and deploying with customers.  

Sonny (CEO) spent 9 years in the Marines before business school and is an expert in force protection and anti-terrorism. Sonny and the experience he brings have been critical in building trust with buyers, who are generally career law enforcement officers or military veterans.  

Before business school, Ben (Chief Product Officer) led a team at Microsoft that built the first AI-powered sales organization at the firm, reaching $110M annual recurring revenue (ARR) in 18 months; he brings that same focus on AI operations and scale to Aegis.  

The data scientists and engineers have Master degrees in their fields, and have been instrumental in building cutting-edge technology that enables our AI to achieve 99%+ accuracy at low cost.  

Our sales person has years of experience selling gunshot detection solutions at Shooter Detection Systems, but believes computer vision is the future of the industry. He brings deep relationships in the school security space and knowledge of sales processes. Aegis have also garnered additional external advisors and investors to further bridge any knowledge gap the management team might have.

Aegis partners with:

• NVIDIA - Inventor of the GPU
  • Aegis is a member of NVIDIA's accelerator program, giving us early access to their technology.
• NTT DoCoMo - Japan's largest telecommunications company
  • DoCoMo has connected Aegis with their US infrastructure clients, where we hope to implement the Aegis solution to protect the production and distribution of basic utilities to low-income communities.

As a subscription-based SaaS product, we charge customers a monthly fee per security camera. Since our product is software-only and cloud-based, it can generate recurring revenue immediately after deployment. Our current price is $30 per camera per month with annual contracts, half of which is paid up-front to help our working capital. We are exploring moving larger customers onto two-year contracts in exchange for discounts.  

We will achieve economies of scale when we onboard additional customer cameras by leveraging auto-scaling of our cloud service usage. We are currently implementing the auto-scaling and edge compute features mentioned in the technology section, so compute costs are currently our biggest line item.  

Our current gross margin is 47%, with most of our costs going to AI compute. Going forward, as we make our back-end more efficient and adopt new AI processing hardware, cost of sales will begin to dominate our cost per unit, and unit economics will start to react 80%+ gross margin, resulting in very positive economics.  

For a standard school deployment with 100 cameras, we will make $36k in revenue annually. For a direct-sales driven motion, we estimate a customer acquisition cost of $10k. At our current gross margin this is a payback period of just over 6 months. When new AI chips enable us to provide our services with 80% gross margin, payback period will fall under 4 months.

As a subscription-based SaaS product, Aegis generates revenue by charging customers a monthly fee per security camera. While we are currently profitable at scale, we expect to lose money on each camera deployed in 2019 because we have not yet deployed on enough cameras to see the benefits of economies of scale. We expect to achieve $1M ARR by early 2020 and $46M in ARR by 2023.

Additionally, we plan to pursue Series A fundraising round in mid-2020.

One of the barriers for Aegis is educating the customer base about the technology to the extent that school districts are confident that this technology works, this technology is affordable and this technology can be implemented at low cost and without infringing privacy.

Beyond funding that Aegis hopes to invest in product development, we think Solve can connect us with mentors and partners in the education space to help us validate our messaging, our approach to the sales process and our impact.

Aegis is seeking partners in the education space. Not only might these partners be customers and these implementations will help us validate and improve our product, these partners can help us improve our messaging to ensure that it is well received by decision markers at school districts and help us design our sales process to close contracts efficiently with school districts.

Funding from the AI Innovations Prize will be used to continue development of the Aegis solution. Our primary focus is to prime the solution for scale so that we can efficiently deploy as fast and as cheap as possible, helping us provide protection to as many people as possible.

Use of Funds: Funding from the Everytown for Gun Safety Prize will be used to continue development of the Aegis solution. Our primary focus is to prime the solution for scale so that we can efficiently deploy as fast and as
cheap as possible, helping us provide protection to as many people as
possible.

Data Compliance: The Aegis solution analyzes video from security cameras and it identifies only weapons, doesn’t analyze people or faces until a weapon is drawn. We delete all data within one hour, unless a weapon is detected. Moreover, we don't require no external data except monitored camera feeds.

Use of Funds: Funding from the Innospark Ventures Prize will be used to continue development of the Aegis solution. Our primary focus is to prime the solution for scale so that we can efficiently deploy as fast and as cheap as possible, helping us provide protection to as many people as possible.

Data Compliance: The Aegis solution analyzes video from security cameras and it identifies only weapons, doesn’t analyze people or faces until a
weapon is drawn. We delete all data within one hour, unless a weapon is
detected. Moreover, we don't require no external data except monitored
camera feeds.