At Databricks, our automation imaginative and prescient is to automate all elements of the enterprise, making it higher, quicker, and cheaper. For the gross sales groups, we’re digitally reworking our vendor expertise by offering genAI brokers that help the vendor throughout the gross sales lifecycle. Our aim is to reinforce the vendor expertise with AI capabilities by seamlessly integrating them into their day-to-day duties and offering an easier, more practical approach for sellers to retrieve data property in addition to orchestrate actions by automating repetitive handbook administrative duties.
Our “Subject AI Assistant” is constructed on the Databricks Mosaic AI agentic framework and offers a approach for sellers to question and work together with knowledge throughout a number of knowledge sources. It integrates with a number of key platforms together with:
- Our Inner Databricks Lakehouse for account intelligence, gross sales enablement content material, and gross sales playbooks
- Our Buyer Relationship Administration Platform (CRM) system
- Our Collaboration Platform collates and indexes most of our nonstructured knowledge
The AI software is used to:
- Conversationally work together with knowledge throughout a number of knowledge sources utilizing pure language (beginning with English)
- Capability to obtain and create paperwork based mostly on the knowledge gathered
- Take actions based mostly on the information insights (replace fields in our CRM, draft a personalised outbound prospecting e-mail, create a tailored buyer proposal, prep for a buyer assembly, and many others.
The sector assistant responds to seeded prompts based mostly on person and web page context and likewise offers a chat-like interface for open-ended queries on the above-mentioned datasets.
Enterprise Impression
Sellers are usually overwhelmed with the quantity of data thrown at them. They want entry to knowledge residing in varied siloed functions, as a part of their regular day-to-day routine. They require quick access to account, alternative, and use case knowledge that resides in our CRM, in addition to buyer market insights and account intelligence, together with account consumption knowledge that resides in our lakehouse. As well as, additionally they want entry to gross sales content material – enablement playbooks, aggressive gross sales collateral in addition to product information base articles and product roadmap paperwork. It isn’t simply restricted to knowledge retrieval, however the true effectivity good points happen when the repetitive handbook duties they carry out every day based mostly on the information insights they retrieve may be totally automated. That’s precisely what the function of the sphere AI assistant is – assist the sellers within the day-to-day duties together with data retrieval, distilling the insights from the knowledge, and performing actions based mostly on these insights.
Answer Overview
Utilizing the Databricks Mosaic AI agent framework, we constructed a discipline AI assistant by integrating each structured and unstructured knowledge from a number of knowledge sources. The answer offers a complete method personalised and tailor-made for our sellers, out there on-demand in our CRM. Among the capabilities supplied embrace:
Buyer insights present a 360-degree buyer account view with:
- Monetary information/insights in regards to the account
- Aggressive knowledge panorama
- Product consumption by product line and cloud
- Buyer assist instances
- Prime Use Circumstances Driving Income
- Vendor Suggestions on different use instances which are supplied to related prospects
Information hygiene alerts
- Use instances which are going dwell within the subsequent week/month/quarter
- Prime use case blockers
- Use instances that lack key data (ie exec enterprise sponsor and many others.)
Gross sales collateral
- Gross sales playbooks
- Aggressive collateral
- Assembly summarization
- Pitch decks
Orchestrate motion
- Replace CRM with the following steps on particular alternatives or use instances
- Draft a prospecting e-mail to a brand new buyer contact
- Create a customer-facing proposal
Our discipline AI assistant answer is constructed completely on our Databricks tech stack. It permits integration into a number of and numerous knowledge sources and offers a scalable infrastructure framework for knowledge retrieval, prompting, and LLM administration. It’s constructed utilizing the compound AI agentic framework and helps the addition of a number of instruments (SQL queries, Python features) which are all ruled by our Unity Catalog governance layer.
Agent / Device Framework
Human inputs are inherently ambiguous; LLMs have now given us the power to make use of context to interpret the intent of a request and convert this into one thing extra deterministic. To service the request, it is perhaps essential to retrieve particular information, execute code, and apply a reasoning framework based mostly on beforehand discovered transformation. All of this data have to be reassembled right into a coherent output that’s formatted accurately for whomever (or no matter) will eat it.
That’s precisely what the sphere AI assistant does to reply to the queries from the sellers. The sector AI assistant has 1 driver agent and a number of instruments and features that carry out the deterministic processing.
- Information basis: That is the set of information sources that the agent interacts with. In our answer, this knowledge basis contains knowledge in our Lakehouse, gross sales collateral, Google docs in addition to knowledge that resides in our CRM (Salesforce).
- Deterministic processing: The set of features and instruments required to supply right, high-quality responses. The LLM can extract fields from a question and move these to a normal operate name to do deterministic processing. Inside the Databricks Platform, the Mosaic AI Instruments and Features capabilities allow this and user-defined features can carry out most actions inside Databricks. These may be usually Python features or easy SQL queries or APIs that combine with exterior apps resembling Glean, Perplexity, Aha and many others. and these may be invoked utilizing pure language.
- LLM fashions: We leverage Azure OpenAI, GPT 4 because the foundational mannequin for the sphere AI assistant answer. That stated, the framework helps a multi-model method the place the particular capabilities of every mannequin is evaluated with respect to the way it offers with particular use instances. For e.g. we’ve evaluated our answer with varied open supply fashions and we selected Azure Open AI – GPT 4 because the mannequin for our answer based mostly on the groundedness of the mannequin, its capability to generate factual and related content material, its capability to select the fitting user-defined operate / instrument for processing every immediate, and its capability to stick to the content material output formatting prompting offered to the mannequin.
That stated, our answer structure is designed to permit for flexibility in adopting new fashions as they turn into out there in our Mosaic AI agent framework.
At Databricks, we’ve leveraged the Mosaic AI Agent Framework which makes it straightforward to construct a genAI software like the sphere AI assistant. Utilizing this framework, we’ve outlined analysis standards and we leverage LLM-as-a-judge functionality to attain the applying responses. The Mosaic AI Gateway offers entry controls, price limiting, payload logging, and guardrails (filtering for system inputs and outputs). The gateway provides the person fixed monitoring of working programs to watch for security, bias, and high quality.
The parts that we leveraged for our discipline AI assistant are:
Answer Structure
Our Learnings
Information is messy – Leveraged Lakehouse, iterative growth of datasets, targeted on data-engineered pipelines and constructing clear, GOLD Single Supply of Fact datasets
Measuring ROI is troublesome – Be ready to experiment with small focus teams within the pilot. Constructing analysis datasets for measuring mannequin effectiveness is difficult and requires targeted effort and a method that helps fast experimentation
Information and AI Governance is a MUST – Interact early with Enterprise Safety, Privateness, and Authorized groups. Construct a powerful governance mannequin on Unity Catalog for the information in addition to the brokers and instruments
Conclusion
By means of this submit, we hope you discovered about our Databricks on Databrick’s GenAI journey and the way we leverage know-how like this to assist our sellers be more practical. Using GenAI for this use case has helped to showcase how AI brokers can considerably remodel and help each facet of the vendor journey, from prospecting and buyer insights retrieval, driving higher knowledge hygiene by automating repetitive handbook duties and actioning these knowledge insights to driving alternatives and bettering gross sales velocity.
Keep tuned for our upcoming posts, the place we’ll proceed to share our experiences on how AI is reshaping the vendor expertise at Databricks.
