VAKT’s roadmap for AI-driven operations in oil trading
In oil trading, operational processes are highly complex. Operations teams are responsible for arranging vessels or barges, securing terminal slots, coordinating inspectors, exchanging multiple documents, and ensuring quality and quantity verification at handover and more. Late nominations can lead to demurrage costs; errors may result in disputes that can last for several months.
VAKT’s stated objective is to help customers reduce friction in transaction processing, including legal confirmation and the logistics operations associated with physical trades. As Jatin Bhadra – the Chief Product Officer at VAKT – explained in a recent conversation: “Operators dislike maintaining logistics data in spreadsheets but have no good alternatives. In-house tools tried so far have often been too rigid, and top-down implementations have struggled with adoption because logistics operators face too many day-to-day operational challenges, causing projects to slow down or stall.” He added: “VAKT’s approach is to avoid becoming another administrative burden. The system needs to be smart enough that operators actually want to use it.”
According to the company, the way to achieve this is through the use of agentic AI technology. A robust data foundation is seen as of primary importance for any AI-driven application. VAKT uses AWS Bedrock as its foundation layer, which hosts the models including VAKT’s own without requiring data to be sent directly to external services such as OpenAI or Anthropic. The company has also updated its master service agreement to define vendor selection criteria, with a key requirement being that customer data is not used for future model training and does not enter the public domain. Most data elements are processed using models deployed within VAKT’s own cloud environment rather than third-party hosted services thus keeping the data private and secure.
VAKT is delivering this functionality in three phases. In the first phase, the company already has two AI features in production: unstructured-to-structured data conversion, which is described as relatively mature, and deal-pairing predictions for logistics operators.
In the second phase, agentic platform with playbook-driven automation is powered by multiple agents working together. At this stage, the system is intended not only to assist but also to act by planning and executing with user control.
The third phase focuses on an agentic platform with agents working together acting autonomously and executing workflows across the operational lifecycle.
VAKT has developed an agentic roadmap based on the gradual introduction of AI-based automation: beginning with suggestion mode and later moving to user-controlled autonomous settings for low-, medium-, and high-risk tasks. The roadmap includes the following next steps:
- Combining email-reading agents with deal-information agents to suggest actions for incoming communications, supported by guardrails
- Introducing user-controlled toggles, once suggestion mode has been validated, to define which tasks agents may execute autonomously and which always require human approval
- Starting automation with lower-risk tasks, such as responding to a terminal, while keeping legally binding counterparty communications under user controls.
- Providing an operator dashboard that highlights commitments, overdue actions, and upcoming tasks over the coming weeks, replacing spreadsheet-based workflows and morning coordination meetings
The cautious approach taken by VAKT is intended to address concerns commonly raised by energy companies regarding AI adoption, including hallucinations, lack of transparency, and “black box” decision-making. Trust in such systems is generally built gradually, with agents learning from human workflows and users retaining control where appropriate and let agents perform automated operation when the trust is built.
VAKT plans to offer two deployment models: a guardrails model, in which agents provide assistance while humans retain approval authority, and a full automation model for customers that are prepared to delegate execution entirely.
Jatin believes that VAKT is well positioned to deliver agentic AI capabilities, emphasizing that this involves not only the AI layer itself but also the underlying technology platform that the company has developed over several years. According to him, VAKT’s ability to ingest, normalize, and structure fragmented and unstructured data across the entire physical trade lifecycle creates the conditions required for agentic AI to operate reliably.
As he noted: “This is a critical distinction: generic AI tools lack access to this structured, domain-specific data layer, and without it, they cannot replicate what VAKT delivers. These capabilities are supported by VAKT’s permissioned blockchain layer, which is designed to ensure that customer data remains within the platform and is not transmitted to external services.”
Keep in touch and sign up to our Newsletter
