Happy female computer programmers giving high-five over desk in office

ServiceNow Universal Request: When and How to Put it to Work

This is a primer about ServiceNow Universal Request—what it is, how it works, which businesses are good candidates for it, and how to deploy it properly. Universal Request is for sophisticated ServiceNow users with high volumes of requests, a wide variety of requests, or a complex knowledge base of information. Universal Request could be the answer if you’re getting numerous mystery tickets with a long time to resolve.

What is ServiceNow Universal Request?

A new feature designed to enable cross-departmental collaboration, the ServiceNow Universal Request application can save many organizations time and money—but only if it’s implemented in the right ways. To understand how the Universal Request tool works, consider this basic example.

Many end users go to a portal website with questions and requests. But of about 10,000 people that submit queries on the website daily, 1,000 are requests that either go to the wrong place or just don’t fit any of the site’s existing routing options. Perhaps it’s a request that spans departments, is poorly worded, or is outside the usual flow. Organizations can scale to the point that these ‘outliers’ demand a dedicated solution.

These organizations can place a Universal Request widget on the portal as a sort of “catch-all” for queries any of those 1,000 people can use to submit their inquiries. Universal Request is powered by predictive intelligence, which reads each issue and matches it with what users have asked many times before. This allows the widget to match the user to the correct department to solve their problem.

This weeds out a number of tickets, but not for issues that predictive intelligence can’t clearly sort into an existing box. ServiceNow’s Universal Request widget uses trigger words in the ticket and natural language processing (NLP) to make suggestions and automate requests to departments. For example, it may respond: “This sounds like a technical problem, here’s the IT guidance people with similar issues have viewed before,” linking to the right form; or, “This sounds like a procurement issue, here is the form people have needed when they have said this before,” linking to the correct resource.

Best Use Cases for Universal Request

ServiceNow created this new feature to help break down interdepartmental silos and foster cooperation. Universal Request allows for cross-departmental teamwork on tickets.

For example, say ServiceNow generates a Universal Request and automatically assigns it to HR. The HR team realizes as they work on this ticket that they need the help of the IT team to solve the issue. Universal Request allows any team to see and work on the ticket at once.

Another excellent example of this use case is onboarding. A hiring manager may tag the HR team to set up employees on payroll and benefits, while the IT team shares the same ticket information to ensure new hires receive computers and other equipment when they start. The Facilities team can be tagged to ensure new hires have building access passes. If there are any changes, say the new employee can start a week early, the HR, IT and Facilities teams are all alerted at once.

Because it is cross-functional, Universal Request is also a great place for questions and issues that need a safety net because they don’t have clear, single-department answers. So it can come at the end of an automated agent flow that handles the most common questions and backstop tickets where there is an explicit need for cooperation, coordination, and multiple departments.

The educational sector is a classic use case for Universal Request. This is because the end users (students, prospective students) are typically very specific about what they are looking for, and a predictive algorithm that looks at prior inquiries is helpful to interpret specialist tickets. These tickets would typically languish because they are so specific and challenging to route to a human agent with the necessary domain expertise. That’s no problem for an algorithm – in fact, it’s a strength.

But any company can benefit from Universal Request. It really comes down to how willing your end users are to put in the work to give the system a complete description of their requests, since that is the data the predictive algorithm uses to make its recommendations. If the user requests are poorly worded or incomplete, the dataset may not yield an accurate recommendation.

We should also touch on the timeline and implementation needs for Universal Request while we’re discussing where it’s a fit. The good news is that Universal Request is included within the ITSM subscription, so organizations that have already purchased ITSM also have Universal Request.

It does require configuration, however, and the timing it requires varies depending on the goals. Broadly speaking, if the goal for the Universal Request widget is a way for multiple teams to work on one request simultaneously, that is a fairly standard setup and would probably take a month or less. That’s because it relies more on workflow definition than a robust corpus of request data to analyze.

But to implement the predictive intelligence piece as a safety net for the most complex tickets, the timing depends heavily on data quality. Until the system can achieve a confidence index of at least 80%, for example, it will not perform as well as a human agent. Namely, manually looking at the tickets will be more accurate and yield a better outcome than the algorithm below that 80% threshold.

The Challenge of Data Quality

If a user provides data about an issue that is vague, it can confuse the system. For example, a user might ask, “Where is the salary form?” as they look for an IRS W-4 form. But the organization’s HR portal is home to multiple forms, many with “salary” in the title. The amount and quality of the data the system has will largely determine the approach to this query.

Historically for this company, it may be that most employees are indeed looking for the W-4 form, but others may be looking for a W-2 or W-9, so they may be given the wrong information.

If the data set is poor, one solution is encouraging users to be more robust and detailed in their service requests. Perhaps there are more mandatory fields or additional stages to the request tickets to clarify their intent. This is good for the data quality but puts more burden on the user.

The other solution is to train the predictive intelligence model to get better at picking up on even vague requests. Some feature engineering is available in Universal Request to improve the confidence index of a correct recommendation. But it’s relatively limited, as you’ll see below.

How We Help Users Capture More Data and Streamline the Process

How many records does it take to fine-tune a predictive intelligence model enough to produce accurate and confident results every time? This is a good question, but in truth, it isn’t easy to answer. The reality is that between 10,000 to 300,000 data records may be needed—depending on how good the data is.

Even more than 300,000 poor-quality records will not yield a good recommendation—whereas 10,000 accurate, full-bodied case descriptions that have been correctly assigned will allow Universal Request to produce far better results.

To unpack this further, it helps to understand a little more about how Universal Request works. There are three predictive intelligence models in the ServiceNow world, and two are relevant here: classification and similarity.

For example, if an end user writes a Universal Request like the one above about the salary form, the system will parse the “Where is the salary form?” language. It may discover that the most often pairing in queries in the past has been between “salary form” and “W4” and that all of these requests went to the HR service desk.

If it can identify this kind of pattern, it may resolve the request that way. It may also identify knowledge articles with “W4” and “salary form” in the title and ask the end user if that resolves their issue.

Although much of the Universal Request solution’s accuracy rests on existing data quality, Acorio an NTT DATA company can fine-tune the predictive intelligence model. In fact, typically 20 to 30% remains for improvement.

For example, a training model test of the phrase “W4 salary form” routes the user to the IT help desk and the correct form with a confidence threshold of 68%. Depending on the situation, you might want to recalibrate the system to see 68% as a positive result. In other words, based on the data you have from the “W4 salary form” query, the system will direct users to that result, but below that threshold, the system will need more information from the user.

Positive interactions between users and virtual agents are another feature that can be overlooked. Technically, these sessions are a much better way for users to refine their search — because they are responding to questions they weren’t aware they needed to answer. This way, virtual agents can serve as the go-to, honing each query to be much more specific. That might help to resolve them, or it will give richer data to Universal Request as the backup.

This is an excellent step for organizations already using ServiceNow that find themselves burning many human employee hours on amorphous requests and tickets. Let virtual agents refine these tickets and assign them out first before pooling them all as unresolved and in need of human eyes.

We can also help by parsing out the right implementation combination of a virtual agent, AI search, and Universal Request. Using our example from before, if a user enters “W4 salary form” in the search box, AI search can autocomplete with the proper tax form based on responses to similar queries. This reduces frustration and eliminates some tickets while also building up the data set of good outcomes for predictive intelligence to use on other queries.

Another option is to create a “genius result” for searches that aren’t intuitive for end users — for example, one in which AI consistently fails to suggest the correct result in the top 10. Back to our “Where is the salary form?” example from above, let’s say that end users are often confused about whether they really need the W4, W2, or W9 because all three come up. Where the organization knows that searchers using this specific term are looking for a particular result, in this case the W4, they can pin it to the top of the search box results, eliminating an obvious, repeated source of frustration — and an ongoing source of tickets.

We can help you tune your existing system to file tickets exactly where they need to be, and to ensure the tools function as a stacked, interactive funnel for tickets with the agent out front and the Universal Request widget as the final stop.