Prasanna Gopalakrishnan joined New Jersey-based Automatic Data Processing (ADP) nearly a year ago as its first Chief Product and AI Officer.
With annual revenues of nearly $21 billion, the human resources software and services is looking to artificial intelligence to help it maintain — and build on — an already strong position in the market.
We spoke to Gopalakrishnan recently about her role, and the three “buckets” of ADP’s AI strategy. Earlier this month, ADP held an Innovation Day focused on new AI features and offerings.
• • •
What prompted your company to create the role of Chief AI Officer, and how does it align with your broader innovation strategy?
AI is very central to ADP’s mission. We lead the human capital management (HCM) space with 75 years in business and innovation. The reason why this role has been created is to give a focus to achieve our ambition in the HCM space. We are all about delivering customer client outcomes, and AI sits at the heart of it.
…Our broader strategy is about delivering innovative products, the best services, and leveraging our global scale for our innovation. As you can imagine, my job, my role sits at the heart of how do you leverage AI to deliver this strategy. AI is the means by which we deliver our innovation.
Who do you report to in the company?
I report to the president of product and innovation, who reports to the CEO.
And how often do you meet with the board?
Every time there is an agenda or a strategy with AI, we have regular updates with the board.
How do you prioritize AI initiatives across your enterprise? What criteria do you use to decide where to deploy AI?
We prioritize our AI initiatives and put them under three categories. First, it needs to have a strategic value and align with that. Second, we look at technology, complexity, and risk. You talked a lot about how AI is growing at such a fast pace, so we look at whether it is adding any risk and the complexity. Of course, any risk can be resolved by putting the right controls [in place], but we look at is it easily implementable? What does that mean to our risk and controls framework? That’s the lens we use. The third one is what is it driving as outcomes for our clients and employees when we implement an AI capability.
For example, we have what is called “smart search” in our mid-market product called Workforce Now, where our clients go in and instead of going through the menus and looking at how do I look up my employees, they can actually go to the smart search and say, “Hey, when did John start within our organization?”
…We have a number of client focus groups, and we engage through our customer success division which, is a heavily dedicated division of each business unit who we work with and say, “Can we talk to our clients so we can understand outcomes.” So those are the three lenses that we use: strategic value, technical complexity and risk, as well as the outcome for our clients. And when we do AI for our employees, we definitely talk about [improving the] efficiency and effectiveness of what they do as metrics.
I would frame our AI strategy and the roadmap that we build together into three buckets: empower, enhance, and transform.
Do your clients help your own internal AI initiatives with the feedback that they give you?
Yes. I would frame our AI strategy and the roadmap that we build together into three buckets: empower, enhance, and transform.
The empower category is about how do we bring AI tools to our own employees, who are either customer/client-facing employees, or folks who are in technology, or folks who are in HR, or in the CFO function. As you know, we have 64,000 employees worldwide, and so we are a large enterprise ourselves, so that’s the bucket of empower.
The second category, enhance, is how do we build augmented assistance into our products that we bring to our HR practitioners and our clients. …The third category, to transform, is around disrupting agents and agentic work. …You are hearing this buzzword everywhere, and so we are also in that space trying to build capabilities.
Can you tell me about another high impact AI use case that has helped your internal business processes?
One employee use case is we have close to 8,000 to 10,000 sellers who do such an amazing job. We built an AI tool for sales development associates to prepare for calls to prospective clients. Before they make a call, we help them to figure out what are the right prospects that they should be calling, with a view towards will they take a call from us and is our product of value for the current business they are in, and the current situation they are in. So [we have] tools with models and information about prospects where we collect external data and we prepare what that list looks like for a seller to call. It helps us improve the closing out of a prospect into a sale. Salesforce doesn’t offer it natively, so we have to build our own models.
Another example I would give in our space is in the product space… It’s called ADP Assist. …HR practitioners and employees can go to a conversational assistant and they can get these answers, and that allows them to get to the information very quickly. We are starting to enhance that so that they can start to take action, not just get information…
…ADP Assist has been used in more than five million conversations across pay, benefits, policies, and time off. By automating answers, it reduces the interruptions that pull HR away from higher-value work. On average, HR spends 15 minutes per employee inquiry — time saved when handled through AI.
Our HR practitioners also use ADP Assist to ingest policy documents and help employees with immediate, accurate answers, saving an estimated 19,000 minutes of practitioner time. The system also helps resolve payroll anomalies such as missed punches and unapproved timecards.
These tools that you’ve described, are they using generative AI?
Yes, they are.
Which platform are you using?
I can’t be very specific, but I will tell you just like any other technology products that we use, we want to use the best of breed. Some models are good with language translation, which we are in the global business of, [and] some of them are great for reasoning models. So we use what is best and is fit for purpose… We use all the open source available LLMs as well as private LLMs that are available in the market.
How is the AI function structured within the company? Is it centralized, federated, or hybrid?
My job is to bring all AI together, so…to answer your question directly, yes it is [centralized]. But…a lot of what we do needs to still get embedded into the product, so you have to work with product teams to make sure they are embedding the AI capabilities as they are rolled out…
A lot of our members work at companies that restrict their employees ability to use generative AI. Some companies worried that employees may enter a query and innocently include sensitive information. ADP has employees across the globe. How did you manage your employees’ use of Gen AI when the tool was introduced?
I say that anything we build at ADP we have very good data and an AI governance process that precedes GenAI. Protecting data for employees and what we build as tools absolutely was already in our DNA, so when generative AI tools came about it was easy for us to implement guardrails.
From a technology tier, we can implement networking rules that prevent people from using these tools. We also use communication as a vehicle to tell employees why we are doing what we are doing and then fast forward we are now rolling out employee tools so that we do want to give them the ability to use these GenAI tools and innovate and get themselves skilled, but at the same time we want to make sure that we have guardrails put in. We are stewards of our client’s data, so we take that responsibility very seriously.
You talked about guardrails. Can you tell me some guardrails that you’ve implemented that have helped you with Generative AI use?
We operate in 140 countries. We are GDPR compliant. We have to protect the privacy of data and we also operate in the U.S. The California Consumer Privacy Act, CCPA is similar to the General Data Protection Regulation, GDPR. So the traditional privacy and guardrails we have for data still exists, but we have add-ons.
We look for toxicity as a guardrail. … We also look for hallucinations. We look to make sure that personally identifiable information is not inadvertently sent to any models, and then we look for what is called drift. Drift is if we implement a model and the accuracy is 98 percent [but degrades over time.]
What do you look for in startup partners that may be able to help your AI journey?
Typically, when we look for AI startups we say, for example, there are data quality challenges in every company and are there startups that can actually help us solve some of the enterprise data challenges. We would partner with them to help us accelerate the journey. [Or we look for companies] that have already accelerated their AI journey, and they bring AI capabilities in a particular segment that we already operate in.
What are you concerned about when thinking about the future of AI at the workplace?
Look, skills are at the heart of everything I think about. How do we upskill our employees now at ADP? We are in the human capital management industry. We are pioneers and leaders in this, so how do we help our clients ride this transformation as they need to think about talent…upskilling their employees.
And how are you upskilling your non-technical employees?
We have been rolling out Gen AI tools to all our employees and we are trying to figure out do we need generative AI tools like ChatGPT, or do we need to give them specialized tools, because the marketing team’s needs are very different from the needs of the chief financial officer’s team.
(Featured image by ADPDigital, CC BY-SA 4.0 license.)
You must be logged in to post a comment.