How we built Google for Business users

at Cuddle.ai (A Fractal Analytics Product)

Photo by William Iven on Unsplash

We are so accustomed to having easy access to information in our personal world. We quickly turn to a search engine like Google to get our answers in a jiffy. Life in the enterprise world is not as simple. Business users have many unmet data needs throughout the day just because the number they wanted was buried in one of their weekly reports or they just couldn’t construct that complex query to access it from the database.

At Cuddle, we enable business users to simply ask questions in plain English on their enterprise data. Cuddle gets business users closer to data without having to depend on citizen data scientists or wonder which database/reports contain relevant information to make data-driven decisions. More on this here.

Here are the choices we made in building a Natural Language Querying platform on Enterprise data:

1. Expect incomplete Natural Language questions from business users

To answer any data question, the following aspects are generally required to form a data query:

1. Measure — Metric/KPI required for the analysis
2. Attribute — Area of business for which the KPI is measured
3. Time period — Period of time for which the analysis is required

Cuddle ASK expects any of the above to be missed out in the natural language question. Cuddle uses AI to make intelligent choices to provide answers that become more personalized to the business user with time.

Examples:

Question: Sales this Month

To answer this question, Cuddle uses the context of the data universe that the user has access to. In the example below, the user has access only to the West Region. Cuddle uses this context to provide the answer for — ‘Dollar Sales for West Region this month’.

Question: Sales for West

In this case, time period is missing. Here, Cuddle uses AI to determine the most appropriate time period based on Cuddle’s learning from how different users analyze information in the organization.

Question: West

Here, both measure and time period is missing. In this case, Cuddle scans across all the measures that the user has access to and uses AI to prioritize the ones that are more relevant to the user. Cuddle decides this based on the learning from the user’s past interactions on Cuddle.

2. Business users do not always know which data table to refer to answer a question

We made an early choice not to ask users to select a database/data table upfront to direct their natural language questions. Instead, Cuddle’s ASK engine does the complex task of scanning for relevant information across all data tables that it has access to and presents the most relevant information to the business user. All complexity of converting a question to a query is hidden behind the scenes for a business user. Cuddle even allows business users to ask natural language questions that require bringing in information from multiple tables.

Business users are guaranteed to get relevant answers to their questions as long as the relevant information exists in the data Cuddle has access to.

3. Data access restrictions play a pivotal role in democratizing data across business users

An open-ended search-based tool like Cuddle ASK allows business users to simply type in any question without restrictions. In an enterprise setting, it is critical to validate if the user is authorized to access the requested information before presenting the same.

Cuddle offers comprehensive data table, row/column and cell-level data authorization rules that can be defined by the Cuddle administrator ensuring data security.

4. It is critical to gather enough inputs about data during setup to make answers foolproof

The ASK engine should be intelligent enough to understand user questions that do not make business sense and handle them accordingly. A good example of this is the specific ways a measure or KPI can be aggregated.

Some KPIs follow specific rules of aggregation — some can be aggregated across all dimensions of attributes and time periods, while some follow a partial aggregation. As an example, the headcount of an organization is generally treated as a cumulative number every month. This number cannot be aggregated (summed) across months.

Cuddle allows Administrators to define the business logic for aggregation rules and uses them to decides what to present to the users.

For the above example, if a business user asked the question — ‘Show me the headcount for this year’, Cuddle will understand that this question is not business relevant and show an appropriate communication to the user.

5. An answer to a business question is a gateway to further exploration

A good analyst provides just an answer to a question, a great analyst goes beyond and provides more context behind that number. Cuddle understands the need for allowing business users to explore and understand different aspects of their business.

Cuddle ASK uses AI to identify the related context and presents it automatically. Context includes the following:

• how the area of business performed in the Year Ago or Previous period
• related trends
• performance against targets (if any)
• Metrics driving the area of business
• KPIs that are frequently analyzed together with the business area
• Key events/anomalies that Cuddle identified in the area of business

All great products are built by making some core choices. We believe the choices we have made here make Cuddle ASK very simple to use without the need for business users to understand what goes behind the scenes or learn any specific syntax for getting the information they need from data.

Follow the Cuddle blog for more articles like this.