A Fortune 50 company had an overall goal to increase evidence-based decision making across the company and to generate new insights systematically, ultimately changing behaviors, impacting business outcomes, and gaining a competitive advantage.
For this project, they focused on analyzing procurement contracts. The analysis allowed them to see a holistic view of all their suppliers to understand the entire commercial value of the vendor over time. Some of the outcomes included optimizing contract negotiation, vendor management, and spend by leveraging unstructured and structured data available in multiple systems.
Parse and extract core entities such as vendor name, contract, duration, and cost of engagement from unstructured procurement data and deliver a unified view of these results for further analysis.
The company did not have the level of visibility required to make informed negotiation decisions regarding professional services and procurement contracts offered to an internal organization.
How Maana Helped
The Maana Knowledge Platform provided a user-guided, machine-assisted iterative “search & discovery” experience for analysts and executives to be able to ask questions related to key concepts like vendor, contracts, etc., such as:
- Given a vendor, what are the ongoing projects?
- Given a vendor, which groups are buying from them?
- Given a vendor, what are the services offered?
- Given a vendor contract, what are the key deliverables?
- Given a vendor contract, what are the Start and End dates?
The knowledge application, DocAssist, used in this project processed over 2K procurement contracts in PDF format. These contracts were connected to a vendor management system containing structured data with appropriate suppliers and contracts and parsed vendor invoices and billing information.
DocAssist performed the following steps to complete the project successfully:
- Entity extraction: Parsing and lifting key entities out of the unstructured PDF contract data forms such as vendor name, customer name, products/services; classified dates like start, completion, and modification (if applicable); summary of monetary amounts encountered.
- Classification: Probabilistic classification of contracts according to customer’s taxonomy, including detection of outliers: contracts not readily classified under the given taxonomy.
- Contextual language models: Using part of speech tags, n-grams, skip-grams and lexicon matches were built to perform entity extraction and classification. The structured data provided was leveraged to generate labels (supervision) to train the machine learning models.
- Linkage of documents: Clustering of documents that belong to the same contract: SOW contract, change orders and amendments.
- Aggregation: Clustering contracts by type, date range, and vendors.
Using the DocAssist Knowledge Assistant, the Maana Knowledge Platform automatically parsed and extracted core entities such as vendor name, contract duration, and cost of engagement from unstructured procurement data.
This approach allowed the company to get a holistic view of all suppliers across organizations and enable better quality decisions around areas such as vendor effectiveness and pricing.