Robotic process automation (RPA) is transforming investment banking. New artificial intelligence (AI) applications are helping firms reduce operational costs, improve client satisfaction and decrease operational risk. Accenture’s Post Trade Processing platform alone currently deploys 55 robots to deal with more than 750,0000 transactions across six critical post-trade areas each day. These bots deliver for us continuous process improvements quickly and at scale, unlocking new benefits for our clients.
But just as human employees need to be continually nurtured and measured to ensure they are engaged and productive, so too do their digital counterparts: the bots. This blog brings together our key learnings around how to manage a bot. Whether you’re managing the performance of a human employee or a bot, the principles are the same: train, monitor, measure, re-train and repeat. In the same way that human resources managers need key performance indicators (KPIs) to measure and optimize human talent, IT teams need an effective system to rate and optimize bot performance in terms of two key dimensions: stability and criticality.
How to evaluate and achieve bot stability
We have identified seven critical metrics for evaluating bot stability:
- Frequency of failure: Gather data on bot failures, including how often they occur, what causes them and which team “owns” them. Possible reasons for failure include the presence of a new navigation in the application, file reception errors, data mismatches and insufficient exception handling.
- Recovery protocols: Bots must be designed to roll back and re-start from the point of failure. Poor recovery can lead to data errors further down the line.
- Run-time variation: Bot performance will vary according to the volume of data being processed. Bot management teams must be able to accurately predict the time required for a given activity.
- Scheduled versus manual interventions: Bots must be triggered at the correct point in a business activity to be of use. Those with external dependencies may, on occasion, require manual triggering. Manual interventions are prone to human error, which can increase the risk of bot destabilization.
- External dependencies: If a bot depends on a third party—for files or triggering, for example—that can greatly impact optimal functioning and should be assessed.
- Environment: Changes between the development and deployment environments, or to the resolution of screens, can affect a bot’s stability.
- Rate of change in the base application: Changes to the base application’s user interface will automatically update the bot. Continual changes to navigation can ultimately lead to bot instability.
Once these critical imperatives are met, IT teams could optimize bot stability by addressing three additional areas; firstly, the human interactions: Multiple interactions in bot orchestration can lead to process instability. It’s important to orchestrate bots in a chain of bots, and enable strong control in the order of execution. Second, you could look at support enablement. As with any application, enabling support with the appropriate documentation of handling errors is required. The quality of the support team and documentation will have a direct impact on the overall grading of the bots. Lastly, you want to look at code quality itself. Use tools to measure the quality of your code—it plays an important role in determining bot stability.
How to address bot criticality
Once stability criteria have been met, IT teams could turn their attention to bot criticality if they want to effectively manage bot rescheduling and license usage. Bots with the following attributes will command a higher criticality score:
- Time-sensitive bots that are connected to business processes with cut-offs (e.g., market cut-offs) which, if delayed, can lead to financial penalties
- Multi-entity bots that cater to multiple business lines
- Hard-working bots that have successfully lowered the headcount of human application users and therefore carry the most risk in the event of failure
Why automation is key
It goes without saying that the overall process of measuring and optimizing bot performance should itself be automated. The introduction of manual processes to bot management would undermine the essential purpose of bots: to accelerate business processes through automation. IT teams must build algorithms that can measure performance month to month, identify degradations and their causes, and initiate remedial actions. Interested to learn more? Contact me at firstname.lastname@example.org.