How AI is Helping to Extract Meaningful Insights from Digital Door Access Data

Our KIWI solution is installed in 20,000 buildings, serving a broad range of use cases. While predominantly found in multi-family residences, our presence is growing in offices, co-working spaces, vacation rentals, and similar environments.

The applications of KIWI are incredibly diverse. They range from installations where KIWI is used solely on a main entrance door by a landlord’s direct employees, to scenarios where it’s implemented across various facility management doors with both internal and external users, all the way to entirely keyless buildings where all doors and users rely fully on KIWI.

To put this into perspective, our customers manage over 1.3 million active access permissions—which, in the simplest terms, can be understood as 1.3 million metal keys being digitized. These digital doors are opened well over 5 million times per month. This translates to an average of 65 permissions and 250 openings per building per month.

However, given the vast differences in use cases, such averages are relatively meaningless. Consider the stark contrast between a heating room door with 3 people having access rights, opened twice a year, and the main entrance door of a 25-apartment building where every tenant exclusively uses KIWI for access.

Therefore, understanding our data in detail is both exciting and relevant for us and our customers – but it’s far from trivial. With new AI tools, entirely new opportunities are emerging, which we believe will become significant for digital cloud-based access companies.

We have only recently begun to delve deep into this potential and are sharing some of our first learnings and ideas in this blog post.

Unlocking Insights: How AI-Powered Queries in Grafana Unlock Access Data

The Start: GDPR-Compliant Data

Before diving into the “how,” it’s crucial to reiterate the “what”: the data. At KIWI, data handling is designed to be compliant with the General Data Protection Regulation (GDPR, DSGVO in German). While system usage is tracked for a variety of reasons, personally identifiable data is omitted, anonymized, or pseudonymized, as required for GDPR conformance.

The Challenge: From Raw Data to Specific Insights

Digital door access systems generate a wealth of data: opening logs with time data, device status like firmware versions, and much more. This data is largely sent back to the cloud in a form that is hard for humans to read. Manually sifting through this data to create meaningful dashboards, find trends, identify anomalies, or track performance metrics can be a time-consuming and complex task. This is where AI and Grafana come in.

Grafana is an open-source platform for monitoring and observability, excellent for creating dynamic and interactive dashboards. However, designing complex queries to extract specific, meaningful insights can still require significant technical expertise. This is precisely the gap that AI tools are now beginning to fill. Grafana can be used to query a live production database (though be cautious about doing this) as well as offline databases used solely for data analysis.

The Process: Making it Work

Imagine being able to ask a simple question in natural language and have an AI tool automatically generate the complex database query needed to display the answer in a Grafana dashboard. This is becoming a reality (but beware of the still omnipresent AI errors/hallucinations).

Here’s how AI can enhance data analysis for digital door access in Grafana in a few simple steps:

First create a CustomGPT/Gemini Gem. A CustomGPT in ChatGPT or Gem in Gemini can be thought of as an GenAI chatbot which is pre-fed with context relevant for the queries you plan to do in the future. In our specific case, this information is terminology, data schemas, technical instructions about the queries, and background information about the system and how it’s used. At KIWI, we make frequent use of relational databases, so schemas are readily available for these out of the box. The chatbot needs to know what kind of database it should query but also that the queries will be used in Grafana later on. Additional, helpful information can be extracted from internal wikis and the company website. Last but not least, instruct the chatbot to act as an expert on Grafana and database queries and instruct it to ask questions if it is not sure about aspects of the user’s question.

Once the chatbot is pre-fed with this information, start a conversion with it and tell it what you want it to create a query for. Be specific about the request and try to stick to the terminology known to the chatbot. For example, you can ask it to “”create a query to visualize the change in transponder vs. app openings in the tenant user group over the last 6 months.”

In Grafana, create a new dashboard and copy the query over to test it. If the query produces errors, feed this back to the chatbot and ask it to explain the issue and provide a fix for it. Repeat this process until the query runs through.

Once you get results, be careful to validate the query itself and its results. To validate the query, proof-read it. Don’t forget that you can ask your chatbot to explain parts of the query back to you, so you’re not alone with this. To validate the results, do a plausibility check on the data, for example by comparing it against a manually created report and/or by checking samples of the data manually.

Once all of this is done, refine the visual appearance of the dashboard and share it with your colleagues.

Once you have built your first dashboards, you will realize that the possibilities seem endless, as AI not only removes barriers for non-engineers to work with complex data sets but also opens up exciting new avenues for insight. A few initial ideas include

• Automated Anomaly Detection and Alerting: For example, when deviations occur (e.g., an unusual number of failed entries from a specific credential, or access attempts at odd hours), Grafana can use GenAI-generated queries to detect this and notify a responsible person.
• Predictive Analytics for Proactive Management: For example, by analyzing historical access patterns, AI can predict future trends, such as peak access times, potential bottlenecks, or even anticipate maintenance needs for access hardware based on usage intensity.
• Optimizing Access Flows and User Experience: For example, AI can identify friction points in access processes, such as consistently slow door openings or repeated invalid credential attempts in a particular area.

The combination of GDPR-compliant data, advanced AI query generation, and the powerful visualization capabilities of Grafana represents a significant leap forward in how we understand and leverage digital door access data. It’s about moving from simply collecting data to actively deriving intelligent, actionable insights that benefit everyone.

Beyond these immediate applications, the evolving landscape of how our customers deploy and manage KIWI is also making data insights even more critical: Our customers are increasingly installing our products themselves or through local partners and are fully managing their hardware and digital access points autonomously. We are therefore slowly but steadily becoming a flexible backbone for digital access and less a provider of a “fixed digital access infrastructure” (think cable or telecom towers). We are dealing more and more with moving infrastructure (think free-floating car rental) or even “virtual infrastructure,” which both also need to be understood from a data point of view. And yes, fixed, moving, and virtual infrastructure in the context of digital access certainly is a great topic for a separate blog!