Eliminate Approvals Using Machine Learning

To Drive Down Request Fulfillment Times by Nearly 50%, Eliminate Approvals Use Machine Learning, Business Process Management tools, and Stronger UI/UX to Make a Difference in Your Organization

In today’s enterprises, we require approvals for everything – literally everything. As resource managers, we are flooded with them. As team members trying to get things done, we are incessantly blocked by them. We have watchers watching the watchers. And then watch-watchers to watch them! I recently did some research (okay, I asked Felipe to help me do some research) into the approval data at several of our Fortune 50 customers. The results were enlightening, but not surprising.

We picked five key customers who have a wide range of request processes running on the platform. Some of them have more complex processes facing a smaller group of requesters (think infrastructure requests) and some of them have simpler processes facing a large group of requesters (think software requests). Most have a blend. Because of this range of process complexity, the average range of days to fulfill was fairly broad: the lowest was 0.87 days and the highest was 9.32 days. There was also a broad range in total numbers of requests over the two year query period: a low of 9795 and a high of over 473,000. However, one thing was astonishingly consistent across the sample group: the group with the lowest percentage of requests approved was 94.85%. That was the lowest. The highest was 98.05% and across the group, they averaged 96.35% of the requests getting approved. Essentially, the approval step is a rubber stamp. And yet – when you look at what is taking time in the overall request fulfillment timeline – for the group, 46% of the fulfillment time was spent in the review/approval process.

No wonder it takes forever to get things done! If we want to succeed at the speed of global business, we have to do better than this.

Machine Learning

When thinking through the real (if mundane) business challenges facing our enterprise customers, particularly IT departments trying to move more quickly and be more responsive, request approvals came to my mind right away as a place where Machine Learning may be able to help. Instead of routing a task of questionable value to an already overwhelmed manager, could we “pre-decide” and/or let the data decide? Yes we can! To “pre-decide,” you can establish a decision management solution to make all but the most sensitive and complex decisions. As unsexy as it may sound, approvals – given the headache and slow down they can cause – are a great opportunity to use machine learning. Based on many past decisions to approve or reject, given any number of data points, you can let your past dictate your future. You can also configure your BPM solution to continue to feed new data points and decisions into the data set so that the system learns continuously over time.

Reduce or eliminate the number of approvals you are asking for

Of course you don’t need to use Machine Learning to make a difference. Your ultimate goal, however you reach it, should be to reduce the approvals you are asking for as an organization. There are many other, even low-tech, ways to do this.

Can you eliminate the approvals required, at least for large swaths of employees or types of requests? Among our customers, there is an emerging consensus that rather than checking up on every decision, you should hire good people and then entrust them to make good decisions! If that concept makes you anxious, maybe be more incremental: do you have a cost threshold? Can you raise it? Double or triple? How about limiting choices to only things that are pre-approved. You can do this by curating a selection of useful options that are all acceptable to request – then no need to approve on the back end. You can also do this by using role-based permissions to only show relevant, approved choices within the appropriate cost ranges.

Another option is to provide periodic reporting after the fact of requests that have been fulfilled. In this way an “approver” could receive a weekly or monthly report for review. The same usability rules apply: make it easy to read, easy to spot an anomaly or mistake, and easy to do something about. For instance, if the reviewer sees a problem on the report, he or she should be able to click the line to follow up for more information. You can configure your solution to ask the requester for further business justification or if need be, you can shut it down right away, for instance removing an entitlement that has been granted if it is not for an appropriate business need.

If you don’t think you will be allowed to remove approval steps or change the guidelines, work with your auditors. In their defense – they are just trying to mitigate risk on behalf of your company and your customers. Many of them are really great and smart people! Give them a chance to help change some policies that are more box-checking than value-add. I have had so many customers surprised when auditors are willing to evolve their rules after considering a strong business case that my customer made.

Reduce the Error Rate

The data from our customers shows a low error rate (less than 1% across the group) but that is because we focus on it very intentionally during the process design phase. If you don’t plan to handle potential errors, they can be a huge cause of delay. Manager approvals are relatively reliable when tied to your Active Directory or central LDAP – most AD systems are configured to not allow “orphan” employees. That is, when someone leaves the company, an active employee must become the manager of that person’s team members.

However, when approvers are not the requester’s manager – this is where the most trouble lies. Approvals routed to named individuals are obviously the most problematic, but it doesn’t help much to assign an approval to a “role” unless that role can be automatically updated when the individual assigned to that role leaves the company. You need an automated, preferably real-time or near real-time integration to the HR system of record that can auto-assign a back-up or alert someone to assign a new individual to a role when the person leaves the company (not when the approval errors out!) Your absolute last resort should be an error handling process so that if approvals cannot be assigned to the configured approver, they are routed to a team who can handle the situation quickly and assign an appropriate approver.

Ideally, you should design processes so that requests for approval CANNOT be routed to user ids of ex-employees.

Improve Your Approvals

Where you really need approval – this is a BUSINESS DECISION you are expecting someone to make. And you should only be asking people to make these decisions where it is extremely important. I believe it is actually unusual that anyone reads the information in the request they are asked to approve. Often it’s a little hard to read because of the language (who writes these things?) or because of the presentation. An obvious but perhaps overlooked tactic is to make it easier for the approver to understand how they are supposed to add value and make the interaction frictionless.

An easy step is to improve the approval message and delivery. Make it easy to see WHO is seeking approval, WHAT they want to do, and what is the IMPACT – why should the approver care? Make sure that the approval message is easy to read, and provides the appropriate contextual information (not too much! Not too little!) Make sure it’s laid out so that information is easy to see, in a well-formatted list or grid rather than buried in prose. If you do need to use a sentence or two to provide context, make sure the person writing that sentence uses clear, direct human language with formality appropriate to your brand. Make sure the approval message is going to be responsive to the device the approver uses.

It is imperative that your approvers can review and decide on simple requests on their phones. If they can’t, stop reading this and make that happen. You may need to work with your vendor and your security team to make sure your approvers can be easily authenticated on their mobile devices but it is a solvable problem.

Given the fact that approvals make up nearly half of request fulfillment time, and if we agree that the step is adding value in very limited cases, it only makes sense to take a machete to them. Hopefully you have some good ideas of where to get started! And if you are a PMG customer and would like help assessing and improving, reducing, and eliminating your approval steps, reach out to me or to your SDM and we will be delighted to help you slash away.

2019-04-08T18:51:10+00:00