It should be clear that if an organization wants to transform itself with data and technology, it can’t do so with technical professionals alone. Amateurs or “citizens” must be engaged to develop applications, make sense of data, and automate work processes. Many companies have dabbled in citizen development, but AT&T has excelled at harnessing citizen capabilities, particularly in the area of data science and automation.
Let’s face it—there aren’t enough professional data scientists and automation specialists to do all the analytics and AI work that aggressive adopters would need to be successful. AT&T is attempting to embed AI and automation into the heart of its business, according to Andy Markus, the Chief Data Officer of the company. As a telecommunications company (and no longer an entertainment company after its divestiture of WarnerMedia earlier in 2022), AT&T is rich in both data and people who appreciate it. A focus on citizen development might not work for every company, but it is well suited to this one. AT&T employs several hundred professional data scientists and automation specialists, but it has thousands of citizen developers in these areas.
Citizen Data Science
AT&T has a data-focused culture, and the company has tried to facilitate employee efforts to analyze data with machine learning models. Mark Austin, who directs data science, AI, and automation for AT&T, said in an interview that the goal is to support all aspects of the machine learning pipeline, including finding appropriate data, obtaining the data, engineering the data to create the desired features, creating the model, deploying the model into production, monitoring its performance over time, and governing it effectively. As in most companies, the first three activities typically require 80% of the developer’s time, but AT&T is attempting to reduce that percentage.
AT&T has partnered with the AI cloud platform provider H2O.ai to create a feature store of commonly-used source and derived data in order to reduce the need for time-consuming data wrangling. Both data science professionals and amateurs find it incredibly useful. The feature store has existed for less than a year, but there are already over 26,000 features from which professional or citizen data scientists can choose. In addition, data science models are routinely crowdsourced using an internal “Kaggle-like” competition application called Pinnacle; where a given competition, Austin says, typically leads to an average of almost 30% improvement in the model’s performance. The process involves crowdsourcing a host of autoML solutions, combined with innovative algorithms and ensembles from hundreds of data scientists and engineers across AT&T.
AT&T got an early start at automating workflows with robotic process automation in 2015. Since then it has put over 3000 bots into production. After the first year of experimentation it created an Automation Center of Excellence (ACoE) and now employs 20 full time employees and some contractors in the ACoE, but the company could only have scaled so rapidly with participation from citizen automators. Austin tracks each project, and has calculated that 92% of the RPA implementations are undertaken outside the ACoE.
Over time the automation effort has yielded a lot of value for AT&T. Austin says that the combined bot implementations have saved about 17 million minutes of manual effort each year, generated hundreds of millions of dollars in annual return, and achieved a 20-fold return on investment.
Austin’s responsibilities also include analytics and AI, and he’s excited about the possibilities for incorporating machine learning and AI capabilities into RPA automations. His group has already created several implementations of RPA that include natural language processing, OCR, and ML-based decision-making. Many companies use the term “intelligent automation,” but AT&T is one of the few companies that have truly achieved it.
Building Technical Capabilities for Citizens
AT&T has realized that it needs to provide technical capabilities and resources to facilitate the work of citizen data scientists and automation developers. On the data science side, the company has developed an extensive technology infrastructure to support automated machine learning that supports both professional and citizen data scientists. It incorporates the following capabilities:
- Use of up to seven different autoML tools at once, with competitions among them to see which creates the best model;
- Some autoML tools allow going straight into production deployment of the chosen model;
- Many reusable datasets to be analyzed with machine learning;
- An intuitive semantic search tool that returns, for example, all “features related to churn.”
- A machine learning AND operations tool “Watchtower for MLOps+,” which monitors not only the data and AI (traditionally called MLOps), but also the entire set of activities (applications, API calls, etc.) in the overall business pipeline.
On the automation side in terms of technology, AT&T has adopted Microsoft’s Power Automate as its automation tool standard, which is well-suited to citizen development. It can integrate with Microsoft Office tools, PowerBI, and even Azure machine learning models. The company also maintains a “Bot Marketplace” from which citizens can choose already-developed automation solutions, with configuration assistance from the ACoE if necessary. About 75 new reusable automation components are added every month to the marketplace.
Building Human Capabilities
AT&T also focuses on the human infrastructure for citizen data science and automation with community-building. For data science it has created an “AI Democratization Forum” with weekly virtual live demos that educate on particular issues or inform attendees about capabilities AT&T has developed. About two hundred participate on any given week, most of which are not professional data scientists. Many can’t code at all. AT&T also has a set of online training materials on various aspects of data science, with 575 total courses available. There are also certifications available on vendor tools.
Community-building efforts for automation include a 40-hour training program offered to anyone who wants to become a citizen. In addition, the ACoE hosts an annual “automation summit” for groups within the company to present and share their automation projects.
Some companies and professional AI developers are still skeptical about the role of citizens in building models and automation solutions, but AT&T has demonstrated that this is both feasible and economically valuable if citizens are given the right tools and resources. Professional developers in these areas will always be in short supply, but AT&T has shown that alternative talent sources can bridge the gap.