Ataccama unveil new AI agent feature

Ataccama held a virtual event on October 24^th 2024, with several speakers outlining the latest news and plans of Ataccama as well as various fireside chats on different topics. This blog reflects some observations that I made on the sessions that I attended.

Unsurprisingly, AI was a recurring theme. The new “One AI agent” feature, available now in early access, allows data products to be created quickly by business users quickly and easily with natural language prompts.

Snowflake and Ataccama have a partnership including a native app. Snowflakes offers an API enabling lots of different apps in assorted use cases (600 customers use these). There are three classes of Snowflake apps: managed, connected and native. The latter is the tightest, executing within Snowflake. Traditionally apps require data movement, which is costly and introduces security risks and may be slow. By executing apps directly in Snowflake these issues disappear, as no data movement is needed. Ataccama Data Quality Snowflake Native App is one such application, allowing Snowflake customers to use Ataccama data quality features directly within Snowflake. It has 50 data quality rules like postal codes and thing associated with specific regulations. Data can be checked against these data quality rules within the Snowflake instance. A demo was shown with an example of validating an email address, and a SQL statement was run that checked a table of email addresses for validity, the data staying within Snowflake but the set of invalid email addresses flagged by the Ataccama facility. The same function can be called from within a dbt interface.

The annual Ataccama customer advocate award was given to Chantale Boulanger at iA Financial Group in Quebec, who has contributed to a case study of their Ataccama experience.

Tia White spoke about generative AI based on her experiences at AWS. AI has been used for years e.g. in finance to determine a credit decision, but there were models specific to a particular task like deciding the credit-worthiness of a customer. Generative AI is much more generic in approach, being deployable in a wide range of fields. However LLMs may make up information in its answers (“hallucinations”) though this may be reduced by augmenting the base model with company-specific data, such as the customer database and customer transaction history of a company. This augmentation is known as RAG (retrieval augmentation generation). One reason driving AI adoption is the explosion in data over recent years, making it hard for traditional technologies to process vast amounts of data in a reasonable timeline. In her view customers need a modern data foundation in order to succeed with generative AI, with a basis of trusted data. Automation of existing services are obvious areas for the effective use of generative AI, such as customer service, contract reviews and supply chain. In a survey she quoted, virtually every company listed the quality of their data as the biggest barrier to effective deployment if AI. She felt that effective data governance in an organization was a key enabler and indicator of likely success for deployment of AI. She also discussed some use cases of AI in production, including fraud management *25% reduction in loss due to quicker reaction time and higher accuracy). A logistics company used AI to improve inventory, achieving a 15% improvement over their previous approach. In healthcare, a clinic decision companion was constructed to build treatment plans. This reduced readmission by 12%.A manufacturing company used AI for predictive maintenance, a 20% reduction in downtime.

In one of the fireside chats, there was one very interesting observation. A company had tried machine learning and explored generative AI when ChatGPT came out. They found that the generative AI version of the same application was much more expensive in terms of processing power, but was also less good at the particular application. This is an important point, since people often apply the “AI” label to both generative AI but also other AI technologies such as machine learning, which has been around for many years.

Ataccama started as a data quality vendor, extending to master data management and then to data governance. It has emerged as one of the leading vendors across the data management spectrum and received a $150 million investment from Bain Capital in August 2022.