“The question is when, not whether, organizations will use intelligent tools. The future is inevitable. The sooner you get started, the easier it will be to keep up. Conversely, the longer you wait, the more competitive advantage you will lose.” - John Sumser
HR Tech: AI and Intelligent Software Implementation – Part I
From an operations perspective, the most important thing to know about intelligent software is that the narrower the problem, the better it works. As a result, a good deal of energy is spent in development trying to break big problems into tiny little pieces. This can mean that larger system implications are overlooked in the pursuit of a problem small enough to be successful.
Great individual instances of intelligent software have very tight, specific boundaries. For example, chatbots work best when based on a decision tree and a script. This focus on success in a small instance can produce big unintended consequences. It’s very important to keep an eye on the overall system and its responses to the new tools.
The number of tools available in Talent Acquisition outnumber other tools by almost 10 to 1. New intelligent tool initiatives are most likely to start in the Recruiting silo or as a part of a larger suite initiative. (Ultimate Software appears to have the largest concentration of customers who are actually using their intelligent tools.)
Starting points seem easier when tightly bounded (point solutions) or delivered as a part of an existing system from an incumbent legacy system. But it’s important to consider the big picture. Intelligent tools in a suite will work with other applications, but can make it hard to switch providers. Point solutions seem like a good place to test the waters, but may not be compatible with your other tools and data.
What follows are the critical strategic and operational concerns.
Never Begin Without A Vision And A Strategy
At a minimum, the HR Department needs a strategy development process with a good first draft in hand. Early commitments to vendors my become locked-in relationships. Careful decision making is critical as is the willingness to move without complete information.
The question is when, not whether, organizations will use intelligent tools. The future is inevitable. The sooner you get started, the easier it will be to keep up. Conversely, the longer you wait, the more competitive advantage you will lose.
However, starting for the sake of starting is a mistake. At a minimum you must have some idea of where you are going (vision) and how you will get there (strategy). Plan first, execute second.
A significant component of the strategy involves knowing where you want to go in general. Workforce planning, with a five-year horizon line, is a good way to understand the kinds of skills, traits, and abilities you want to build in the workforce. Strategic workforce planning requires an in-depth understanding of the company’s view of its future. The workforce plan is simply the set of talent acquisition and development outputs required and the execution details for getting there.
That’s simple in the saying and very complex in the doing. If you’ve spent time listening to the various ideas about the future of work and the skills required to get there, you might be experiencing a bit of overwhelm. There’s a good deal of talk about AI replacing workers willy-nilly. There’s even more talk about the remaining workers needing to acquire very important, but completely undefined, ‘new skills.’
Cutting through the noise to deliver a clear set of scenarios for future workforce composition is integral to the larger strategy conversation. The company knows where it is going. The HR input to that strategy is a combination of feasibility analysis and execution plan. It should say, “Here are the kinds and numbers of people we need; here are the obvious shortages; and, here is the plan to overcome the risk or the assertion that it’s impossible.”
Workforce Impact: How To Think About And Implement Reskilling
While it’s true that AI is going to have a powerful impact of the way we work, it’s no longer clear that there will be significant job losses. It turns out that completely driverless cars and trucks are a long way off. And if drivers are no longer managing and maintaining the fleet, someone will have to. At a minimum, workers will need to learn how to process new kinds of information. They will also need to know how to argue with machines.
As I mentioned, silly exercises of administrivia are not long for the world. Repeated tasks that can be replaced with decision trees or scripts that handle every case the same way will sweep through HR Departments over the next four or five years.
In addition, there will be an onslaught of intelligent tools. Every decision will have an accompanying recommendation, suggestion, or forecast. Employees will increasingly be called on to follow, approve, or debate those outputs. At first, the task will feel superhuman, especially since it seems like the machine has all of the data. But the data in the machines is not always the only source of important information or inquiry.
Ultimately, these conversations will become a quality vs quantity debate. Employees will have to adjust to being overruled by their digital compatriots. They will simultaneously learn how to evaluate machine performance based on outcomes. One manages machines by becoming increasingly specific about the requirements for the outcome, while understanding the larger context and strategy the decisions support.
Get Clarity with your objectives. Never Buy Without A Specific Use In Mind.
ROI is the wrong premise for purchasing intelligent tools. While there is a cash return somewhere down the road, current investments are about maintaining agility and learning to operate in the new environment. It is completely predictable that the technology will change so much that you feel like you are constantly starting over.
The one thing to keep in mind with early experiments is that improving a single process with an intelligent machine is profoundly different from managing at scale.
Early experiments with intelligent tools are partly about the specific task and partly about learning how to manage ‘digital employees.’ The current crop of ‘potential digital employees’ are the worst sort of employee imaginable. They require explicit instructions because they are literal. They keep going until you tell them to stop. They have no conscience or compassion. They must be completely retrained when something goes wrong. They think the entire universe is what they have been programmed to know. And their outputs and predictions are limited by the data set they process.
It is unlikely that many jobs involving more than repeatable administrative tasks will be automated any time soon. Instead, job holders will get new digital interns who require a great deal of supervision. Having a clear problem to solve makes it easier to get comfortable with noticing when the machine is wrong and learning how to fix it.
And still, this is the separating point between 20th and 21st century work realities.
Costs Of Management, Maintenance, And Training.
Intelligent tools use new kinds of technology, make new kinds of mistakes, and may produce inconsistent results. Staff members will have to learn how to take care of the tool and help the organization up the learning curve when data models, sources, interfaces, or integrations change abruptly (as they will). It will pay to overestimate the costs of operating a system of intelligent tools.
Many vendor’s claims depend on the client’s ability to deliver specific documentation online. The cost of and schedule for developing the material must be a part of the conversation.
Should you purchase a Suite or Point solution?
The narrower the focus, the deeper the solution. It is not possible for a provider of a comprehensive set of services to provide the depth of a vendor who is focused on a single problem. While many practitioners can imagine a perfectly integrated set of tools built from ‘best of breed’ products, the reality is that aggregated tools inevitably run into prioritization problems that eliminate the possibility.
There are significant risks and benefits associated with choosing a suite solution. This category includes all legacy providers who are fielding intelligent solutions across the subset modules in their solutions. These vendors have tons of historical data with which to train their intelligent tools. They have deep experience across the depth and breadth of HR. They are uniquely suited to providing cross-functional insight. Many are building ‘ecosystems’ of providers who provide supplemental tools for their ‘platform.’
The problem is that buying into a single system locks you into a point of view about what’s important, what should be managed, and what needs attention. As long as your company world view aligns with the vendor, no problem. Shifting vendors becomes more and more difficult as your measurements are aligned.
For example, nearly all intelligent tool providers offer some sort of ‘flight risk analysis.’ There are no standards for what data is best to predict whether an employee is likely to leave. The data model simply has to be able to predict prior year history to be validated. A time and attendance company with historical data will build an entirely different model than a talent management suite. There is no current method for comparing and contrasting. This raises the possibility that if the different programs were run on the same population, employees who would be on one list may not be on the other. Yet the premise of the tool is that who appears on the list is a critical component of management.
When your management practice is built on a foundation of one vendor’s point of view, it is difficult to switch to another.
There is also no central force that mandates data governance concerns for intelligent tools.
And the difference between tools as part of a larger suite and point solutions for specific tasks matters. Management tasks that are simply handled by the suite provider are the day to day headaches of the owners of point solution collections. While point solutions are often hands-down better at the unique tasks, they can also be more expensive for overall HR Management because they are not integrated into other tasks and functions and require separate attention.
The choice is not simple and bears close attention.
Readiness Equals Data Cleanliness Which is Foundational to Governance
On one level, data cleaning and governance is the classic ‘perfection is the enemy of the possible’ question. Still, the consequence of the first generation of cloud configurations is an ocean of unique workflows that do the same thing for different KPIs, which are all named differently. Current machine learning based intelligent technology works better with structured information. That means having the same steps in a process and using standard protocols to define the fields. Unstructured information is useful for mining underlying trends and concepts, but not so great for systems that lead to process improvement.
Part of the appeal of older systems was the fact that they allowed a great deal of freedom in task specific workflows. Rather than centralizing a workflow authorization process, individual users were often allowed to build unique, situation specific workflows. We’ve seen cases where companies developed over 300 detailed recruiting workflows in which identical steps in the process were called different things.
It’s not a good idea to solve this problem by edict. People are comfortable with the way that they work and will resist authoritarian changes to the way they work. The process has to move slowly and win the commitment of users, almost one at a time.
Once begun, the data governance process becomes a part of ongoing operations.
First, you need an inventory of all fields currently in use across all of the systems that you might want data from. This is often harder than it sounds
Second, the comprehensive list needs to be evaluated for data content redundancy, over-complication, duplicate names, and overlapping processes. Each item should include identification of human users, machine users, sources, and update frequency.
Third, visibility, reporting, and machine processes must be assessed to define a clear picture of output requirements. This is where having a clear picture of where you want to go really matters.
Fourth is the hard work. Through a process of interaction and education, talk with each user (who has workflow permissions) about the tradeoffs between task customization and system level management capabilities.
Finally, establish a governance committee responsible for understanding and navigating the issues that will emerge over time. As data volumes increase, it is important for the organization as a whole to be cognizant of the implications. In the 21st Century, data is infrastructure. This is how you manage and maintain it.
Over time, the dynamic role of the data governance process will simply become a part of work. It is the mechanism that enables humans and machines to work effectively together.
Control of Data Model Standards and Process Governance
You will find it hard to imagine the number of discrete data models that will be used in your organization. They wear out, require maintenance, create liability, and imply service level promises to employees. Success requires keeping the big picture in mind.
The tools and techniques required for management of intelligent tools at scale are just starting to be developed. It’s becoming clear that every company will be using multiple data models per employee, customer, division, project, work team, candidate pool, individual candidate, alumni, customer, investor and other stakeholders. Understanding the health of each data model is critical. Knowing how to prioritize repairs and improvements is critical.
It would be hard to spend too much time trying to figure out this aspect of the future of work. It’s a cross between admin, IT, strategic planning, productivity measurement, and the foundation of work going forward.
These data models will be fed from various sources including on the job equipment, monitoring devices, communications tools, and software that uses public data. The organization’s data needs, production, and consumption will become a, if not the, central management preoccupation.
Read the entire two-part series
- HR Tech: AI and Intelligent Software Implementation – Part I
- HR Tech: AI and Intelligent Software Implementation – Part II
