top of page

Sign off Central

Goldman Sachs, 2019-20

System summary

Submissions such as CCAR, and FDSF, made to regulatory authorities, are often comprehensive and span the firm's breadth. Business, Risk, Accounting, and Operations all provide evidence of the firm's financial health. Coordinating these exercises across the teams involved is a mammoth and complex task, and we built Sign-off Central to address this problem.


Regulatory Reporting

The team coordinated submissions across businesses and interfaced with regulatory representatives.

Risk Management

~100 people

Product Team

4 people

My role

Product co-lead: I designed the processes to streamline coordination and workflow around regulatory submissions. The primary problem statements that we were looking to solve were working across different teams, different time zones, and asynchronous communication.

User needs

Task handling

  • Ability to generate, assign, verify, and close tasks.

  • Aggregated task views for leads and managers to sign off.

Collaborative documentation

  • Creation of standard language and hashtags to create a consisted language to record invetigations and insights.

  • Interface with commentary suite to generate management packs.

  • Integration with symphony to have active lines of conversation.

Co-ordinator view

  • Integrated Gantt chart to track progress.

  • Creation of status and summary reports and emails by collating information from tasks from the current submission

  • Historic run comparison to identify items requiring attention to be fixed during runs


This system was to mark the shift from traditional spreadsheet-based risk management into a system-based approach. The two primary goals set forth were to reduce the time taken to produce submission grade risk metrics and increase the process's visibility across teams and geographies. We built the following components to bundle into our product to achieve these goals.

Task Object

We envisioned a task as an object that would reside in the data warehouse and could be accessed by users and systems.
The object stored information such as task creator, assignee, level of urgency, metadata about deliverables, and methods to store datasets, queries, responses among others.

Task Handling

There were two methods to generate tasks, via a user interface and via an API. The interface is usually used to ask clarification questions, or request supporting data for a conclusion. The API however is extensively used in conjunction with the anomaly detection system we had worked on earlier. The system generated a list of anomalies for users to handle. The task list was linked to final submission, which could not be made until all tasks were closed at all appropriate levels.

Task Routing

Task routing allowed users to move tasks back and forth or down the supply chain. The router connected to communication channels including email and symphony to alert users (and managers).

User View

This view showed a user the assigned tasks, along with their status, urgency, and deadlines. The view was color-coded as red amber and green to allow for a quick visual recap. Users could customize and pivot their views by urgency, expected delivery, deliverable. This enabled users to tune views to thier own working styles

Aggregate view

This view was created for managers whose teams were participating in the exercise. As the person accountable to the firm's management, it was important that managers had a full understanding of the pack produced. The manager had a view of every task that was assigned to any member of their team, and its progress. As a secondary sign-off mechanism, a manager sign-off on all tasks was required to mark as closed.

Coordinator view

This was an aggregate of the manager view that a run coordinator had insight into. By this point, the number of active tasks could have reached a few hundreds, therefore the view had to be a statistical one with an ability to drill down. An interactive datacube was used to handle this.
The coordinator view also had the ability to customize dashboards that they wanted to use with their submissions. This feature was beneficial as it allowed coordinators with distinctive styles of work use the system with equal ease.



Using a systematic data backed approach, we were able to influence multiple teams to address issues in their infrastructure which lead to iterative streamlining of the process. Over three quarters, we reduced time to submission from six weeks to two weeks.


We were able to generate on the fly summary and commentary packs, which were previously done collating manual inputs. The accuracy of this pack improved manifold as inputs were no longer subjective.


Using the task handling system, we gained increased accountability among people, who in return felt invested in the project.


The output of this system which now had standardized language and used hash tags to bundle up comments led to the creation of our next product – the Commentary suite, which was a standalone product that allowed users to generate management packs from raw numbers.

bottom of page