.png)
Sturdy was built to help businesses better predict results and to improve revenue. Our premise is that the events that impact revenue and retention are walled inside of product, marketing, finance, and sales systems.
By organizing and illuminating revenue events we can help our customers get to a collective truth that will lead to better products and services, more efficient teams, and happier customers.
Our platform collects business information from every silo and organizes it into one interface. It leverages several flavors of Artificial Intelligence (AI) technologies in a variety of ways. We have built our own Natural Language Classifiers (NLCs), and we partner with 3rd party AI providers. We use AI to identify revenue risks and opportunities, to securely anonymize data, to detect data policy violations, and to make people more efficient by reducing the need to manually log issues.
At Sturdy, we have several inviolable principles:
Please read on to learn more about us.
Sturdy collects and analyzes data from corporate data sources and silos. Some examples are support ticket systems, corporate email, customer chats, and recorded sales and support calls.
We collect this data only from business systems and never from personal sources. Our customers configure the accounts, fields, or user data that is collected.
We use our own proprietary Natural Language Classifiers (NLCs) to analyze unstructured data in order to identify topics that are relevant to businesses, with a lens to identify issues that impact revenue . Among other things, the topics we identify help businesses improve retention, customer satisfaction, pipeline reliability, and product design.
Sturdy’s NLCs label items based on the probability that a conversation indicates a risk, opportunity, or data point. For example, if a customer says, “This is sooo buggy,” Sturdy would label this conversation as “Bug” and, depending on context, “Unhappy.”
Another example is the “Response Lag” label. This label appears when a customer says, “No one is getting back to me,” or “I am still waiting for an update.” Your team can use this label to immediately respond to that customer and identify why the issue happened in the first place.
Sturdy’s customers can also craft their own labels to detect specific issues and opportunities that may be unique to their business. For example, if a Marketing Manager wants to understand how customers respond to a new pricing plan, Sturdy might label a conversation as “New Price: Negative.”
What our AI does not do.
Sturdy’s NLCs are non-generative. They cannot generate names or reproduce text.
Our product does not replace the human decision-making process.
We do not use AI to socially score people, make hiring decisions, or copy actors' voices.
Our AI only detects topics that are relevant in a business context. Our use of AI is only to classify and summarize business events so that human beings have the data they need to predict and resolve those issues.
The Information Sturdy Analyzes
Sturdy’s NLCs analyze unstructured information in the bodies of text (corporate email, tickets, chats, etc.) that is generated and stored across a business when it communicates with its customers.
We have a library of pre-configured labels that were trained to classify unstructured data. We have never used public data to train our models. We have partnered with many private organizations who have agreed to let us train our models using their anonymized data.
The information we analyze is intentionally transmitted by a business to its customers or from a customer to a business. By intentional we mean that, for example, when a customer submits a support ticket they expect that someone is going to view that support ticket, and the Support Representative assumes that someone else has the authority to read the response.
How Sturdy Collects Information for Analysis
Sturdy’s AI analyzes information created when a customer communicates with a business or when a business communicates with a customer. Some examples are support tickets, marketing emails, and chats. Some of Sturdy’s customers use it to analyze recorded sales calls and the messages generated by their AI chatbots.
Sturdy collects communications directly and only from business systems. The collection is administered by our customers.
Sturdy will only collect a communication if Sturdy knows that the message is “to” or “from” a customer. For example, if an email does not contain a customer contact, then it is ignored and never processed or saved.
The Information Sturdy Does Not and Will Not Analyze
Sturdy only analyzes the communications that occur between businesses and their customers.
Sturdy does not access, analyze, or store personal emails, personal phone calls, or internal-only correspondence. Sturdy cannot be configured to access someone’s personal email address, and if a customer somehow used Sturdy to surveil private conversations, then we would terminate their service.
However, as an example, if an employee had a personal (i.e. non-business related) conversation with a customer using their work email or ticket system, it could be incidentally collected and analyzed by Sturdy. This message can easily be deleted by a Sturdy Administrator. Since it is a long accepted practice that personal conversations should not occur on work systems, we do not see this as a violation of our commitment to privacy.
Sensitive Personally Identifiable Information (SPII) and Personal Health Information (PHI)
SPII and PHI should not exist in the majority of the systems we collect data from, as that is a violation of long-standing information security practices and data policy agreements.
However, as an example, we understand that many people still email their Social Security Number, or their Birth Date, to a business when they have a question about payroll.
Sturdy detects and destroys PHI and SPII, in memory, before it is collected and saved, at a recognition rate of over 99%. We also do not collect screenshots or attachments.
We can alert your team if we detect this data so that you can remove it from the host system.
Data Training.
Our models are reviewed regularly and retrained weekly.
Our NLCs are trained and improved when our customers provide false-positive and false-negative feedback. Our customers can opt-out if they so desire.
Personally Identifiable Information (PII), Personal Health Information (PHI), and Sensitive Personal Information (SPI), that could be used to identify a person or an account are irretrievably removed before analysis.
Our third-party providers are contractually prohibited from using any data for training purposes.
If a customer cancels, all of their data is permanently deleted, including any anonymized data that may reside in a training set.
Summarization and Semantic Search.
Our platform allows our customers to search across their chosen data silos from one interface. It also allows people to summarize search results. For example, you can ask, “Please tell me what happened with our top 10 customers in the last 30 days,” or “What were the top 5 sources of customer confusion in the last 60 days?”
To accomplish this, we partner with some of the well-known LLMs. They are contractually prohibited from training using our customers' data. You can find the list of partners on our website under our list of sub-processors.
Bias.
Bias, in this context, could be said to be the application of prejudice in a decision-making process.
We continuously review our product to detect potential bias.
Additionally, Sturdy does not use data to rank or classify individuals or groups of people. Sturdy does not make decisions. In other words, Sturdy does not analyze data that could create bias, it does not classify people, and it does not make decisions that could enforce prejudice.
Data Leakage.
Data Leakage refers to the possibility that confidential data can be exposed by an AI when it generates results or that confidential information is used in the training of a model.
Sturdy’s NLCs are non-generative and, therefore, cannot leak data out; they do not generate information about their context. They only generate labels based on the likelihood of a conversation containing a certain type of business-relevant topic. For example, if a customer says, “I really hate this new feature,” Sturdy might think there is a 95% chance that person is unhappy and label that conversation as “Unhappy.”
Where text is generated during summarization, it is only generated using the body of conversations that relate to that prompt. For example, “What are the top 5 bug reports?” only includes content from the conversations from that tenant that were labeled as bug reports. It does not generate text across tenants and is never trained on the data in the prompt.
Privacy and Data Security.
Sturdy was SOC2 Type 2 certified before it had its first customer. It is GDPR Compliant and complies with HIPAA standards.
Other Concerns Relating to the Ethical Use of AI
Based on what we do, here are some other concerns that we run into:
“Our customers don’t want an AI to read the emails and tickets they send us.”
We rarely hear this. We don’t believe a customer would contact a company with a problem and be concerned about a technology that helps get the issue resolved. Our use of AI is like “asking to speak with a manager.” The difference is that the manager listens to every message and ensures it gets resolved, not just the ones from really mad people.
“Our employees don’t want an AI listening to what they say to customers.”
We hear this fairly often. We do not want your coworkers to think of Sturdy as a surveillance platform. That’s not why we’re here.
Today, everyone has a different tool, and a different reality. We’re all migrating data from one tool to another: “Log it to Jira,” “If it is important, save it to Salesforce.” The “silo-fication” of our shared reality has led to vast amounts of data entry, lack of transparency, and customer unhappiness.
We eliminate the need for manual logging. Sturdy gives everyone the data they need to improve product, service, pricing, strategy, customer happiness, and ultimately, revenue.
Imagine if your product team could search every email, ticket, and chat for “what to build next?” Imagine if your CS team could get an alert every time a customer complained to anyone in the company. In our experience, the lift created from a unified customer reality results in incredible lift for our customers. (And we can prove it here).
.png)
