Proving efficiencies from AIOps in federal government
- By Matthew Leybold, Allen Chen, Steve Mills
- Oct 14, 2020
Intelligent automation has changed everything, and as a result citizens expect to interact as easily with the government as they do with commercial online sites. That means filing a tax return, registering to vote or paying a parking ticket should be as simple as ordering a pack of batteries. And just as Americans trust Amazon will maintain the highest standard of security to keep data private, they also expect government agencies to protect their data and operations from external threats.
However, three quarters of the federal technology budget is spent on operations and maintenance for legacy systems, according to IDC, which means citizens are not benefiting from the efficiencies that come from new technologies powered by artificial intelligence (AI) and machine learning (ML). To design a digital experience that matches citizen expectations, enterprise IT must move from a back-office support function into a strategic catalyst that unlocks value and enhances public-sector safety and productivity.
For any government agency, reaching this target state does not mean scrapping legacy IT altogether; digital transformation can be achieved by integrating AIOps, which leverages big data, advanced analytics and ML to enhance and automate IT operations and monitoring.
The business case for AIOps
Any commitment to a more AI-inspired way of working has to overcome three major challenges before an enterprisewide rollout can be considered.
The biggest challenge is the flood of enterprise, operational and mission-centric data that IT must manage, as various public-sector agencies integrate complex datasets from different departments and private partners. By using AIOps, users can build a data model that sources data from disparate systems and authoritative data sources to detect anomalies in the behavior of applications and IT infrastructure. Alerts will continuously monitor the health of the environment and reduce the noise by analyzing alert patterns to filter out false positives, enabling managers to prioritize where action needs to be taken.
Secondly, spending on new technology for mission-critical capabilities, like AI, is still being held back by budgets that require CIOs to do more with less. But through AIOps, agencies can measurably decrease the time to repair IT failure, replace manual tasks with automation and optimize the capacity of multiple business units. The efficient use of resources and increased output leads to cost reductions, more than offsetting the initial investment required to implement AIOps.
Finally, organizations must shift from legacy to next-generation security, even when it may seem too big a lift for most government agencies. AIOps can circumnavigate this by focusing on specific, security-centric use cases (such as threat vectors, vulnerabilities, fraud) that reduce the risk and overall attack surface of the enterprise technology footprint.
Where to begin
As technology becomes a driver of value, so the role of a CIO evolves from a passive manager of technical delivery to an agent of change.
CIOs should begin by performing a holistic review of their enterprise IT ecosystem and identifying the use cases that will advance the agency’s mission, such as anomaly detection to enhance transparency, noise reduction to leverage transparency into insights and activities informed by insights that proactively remediates issues.
With a use case identified, a proof-of-concept for AIOps should be delivered iteratively. This allows agencies to start small and grow their AIOps program as value is delivered. A six-point build process for a particular use case can scale over time to an enterprise solution:
- Establish a set of potential use cases that can be prioritized based on feasibility, level of effort, complexity and their ability to deliver value.
- Build an AI/ML model with initial enterprise data capture, focusing on those sources that drive insights so that a platform can be built to enhance efficiencies.
- Iteratively train the model with predictive AI/ML techniques for future cycles. For instance, agencies might use an isolation forest or clustering approach for anomaly detection or use logistic regression or random forest analysis for failure prediction.
- Generate insights that enable informed decision-making using outputs from the data model as well as inputs into automated action handling for remediation.
- Automate decisive action by building action handlers for cloud and IT infrastructure that can orchestrate production changes and manage the feedback loop to IT service operations, returning information to the data model in the process.
- Establish a feedback loop for continuous improvement by evaluating the results of the minimum viable product against the original value hypothesis of the use case, and then build this to an enterprise-scale service that will return the value to citizens and stakeholders.
Once the process is implemented, the approach is the same for subsequent use cases, even though the applied ML models and analytical techniques may vary.
Where to end
By generating increased efficiency and effectiveness through AIOps use cases, federal CIOs can begin to unlock the value of intelligent IT operations to predict and prevent cyberattacks, process increasing call center volumes and make sense of IT operations data.
Ultimately, AIOps is a strategic imperative for government agencies striving to match the expectation of their citizen stakeholders. It can be deployed at breakneck speed and iterated in real time. Most important, though, the process for implementing an AIOps use case does not change, so once an agency learns the approach, there is no limit to its ability to scale and evolve towards intelligent automation and predictive capabilities.
This might seem like a bridge too far for some agencies, but with a proof-of-concept designed around value realization, CIOs can take the first steps without excessive risk. Eventual enterprise-scale rollout will meet the expectations of citizens who view themselves as government customers and, as such, expect the same levels of service they get when ordering a pack of batteries from Amazon.
Matthew Leybold is an associate director in BCG Platinion out of New York City and leads the cloud and it infrastructure topic as well as public sector for Platinion North America.
Allen Chen is a partner and associate director in BCG Gamma out of Seattle and leads product and engineering for BCG Gamma’s data science platform team.
Steve Mills is a partner and associate director in BCG Gamma out of Washington, D.C. and is the BCG North America lead for public sector AI as well as the AI ethics lead.