Management of Change (MOC) Effective Use of Reminders

The presence of an “Approvals” state in the Management of Change (MOC) business process may mislead someone to believe that approvals/signatures/sign-offs only occur during this state. That is incorrect—signatures/sign-offs may occur at any point, as indicated by the icons in the diagram. The Approvals state is highlighted, since this is the approval to “go build”, or actually implement the change. Prior approvals dealt with aspects of assessing feasibility. Subsequent approvals validate that the change was performed correctly. Also the Approvals state typically involves multiple approvals, in order to cover the technical specialties demanded the nature of the specific MOC.

Figure 1. Full, permanent, normal MOC lifecycle, showing an Approval state plus supplementary approvals.

In an electronic MOC system, approvals (whether during the Approvals state, or otherwise) are normally requested by an email notification. The approver is supposed to accept or reject the item within some timeframe, often a few days. The approver’s activity, or more correctly, the approver’s lack of activity is monitored by the eMOC system. Once the approval timeframe is exceeded, without an approval, the eMOC system issues a reminder to nudge the approver into completing the approval task.

The use of reminders is common in electronic systems, including purpose-built eMOC applications, as well as ECM systems upon which eMOC applications are often built.

There are several questions to ask about reminders: do they actually work? How well? Under what circumstances?

MOC Reminders Case Study (Part 1)

Suppose an eMOC system tracks approvals. For each approval, the system:

• assigns a unique identifier
• captures the name of the approver
• captures the timestamp when the approval is requested
• captures the timestamp when the approval is granted

The difference between the “requested” and “granted” timestamps is termed the approval duration. Suppose we consider a sample of 100 approvals, of which the first few are shown in Table 1. Note that, in calculating the duration, each day is deemed to have 8 working hours.

Table 1. Time for a single approval, without intervention.

A useful perspective can be gleaned if the dataset is sorted. The first four columns of Table 2 represent the same data as in Table 1, except that the data is sorted by duration.

Hoff, R., Quantifying the Effectiveness of Interventions in Workflows, submitted for publication in ASME Journal of Computing & Information Science, Dec. 15, 2008.

If we are going to look at the impact of reminders, we’ll need to represent the data in Table 2 using appropriate statistics. From a practical perspective, that’s where things get difficult, since:

• this leads to debates about which probability distribution is appropriate for the data,
• even when there is agreement on which distribution is appropriate, there’s still the problem of fitting a set of data to the distribution; except for the normal distribution, this can be time-consuming and/or complex.

I’ve come up with a very simple approach, that, it turns out, tends to be very accurate as well. Here’s how it works:

Let’s define the following variables:

T₅₀ = 50% of the durations are less than this value. This value is easily determined from a sorted list of durations. In Table 2, T₅₀ = 20.75 hr.

T₉₀ = 90% of the durations are less than this value. This value is easily determined from a sorted list of durations. In Table 2, T₉₀ = 36.99 hr.

T₀ = The average approval duration, without reminders. This is quantity that we want to minimize by sending out electronic reminders. For the data in Table 1, the average, T₀ = 21.65 hr.

MOC Reminders Case Study (Part 2)

Now let’s send out a reminder every 8 hours. Here are the rules:

1. If a person hasn’t performed his approval after 8 hours, an electronic reminder sent out.
2. Once a person receives the reminder, he has 8 hours to respond, otherwise another reminder is sent out. This continues, every 8 hours, until the person responds.

Assuming that these reminders are actually effective, the approval time for certain approvals (i.e. some of the late ones) will be reduced. The reduced average approval time is represented by:

t_αThe average approval duration, with reminders. In Table 2, the approval durations with reminders are given in the right-hand column. The average of these durations is t_α = 15.43 hr.

which allows us to define an efficiency parameter:

γ The net intervention efficiency, defined by =1-(t_α/t₀) If the reminders are totally effective, then γ=1 If the reminders are completely ineffective, then γ = 0. However, in this case, γ=1-(15.43/21.65) = 0.29

The net intervention efficiency is 0.29, which means that the average time for an approval to take place is 29% less with electronic reminders, than without.

Key Variables

There are 3 key variables used in the previous analysis. They are worth reviewing, since they will help with predictions in other cases.

α = The effectiveness of an individual reminder. Recall that if a person didn’t perform the approval within 8 hours, then another reminder was sent out, and they kept being sent out until the person responded. That means (proves!) that an individual reminder is only effective sometimes. The proportion of time that an individual reminder is effective is denoted by the symbol α, called the “individual intervention effectiveness”. In the example, detailed in the right-hand column of Table 2, α = 0.5

Δt = The intervention interval. Recall that the reminders were sent out every 8 hours. They could just as easily have been sent out every 4 hours, or every 16. The intervention interval is denoted by the symbol Δt. In this example, Δt = 8hr.

T₉₀/T₅₀ = The skewness ratio. The greater the value of T₉₀/T₅₀ , the more “stretched out” the distribution in Figure 2 appears. A more “peaked” distribution would have a smaller T₉₀/T₅₀ ratio.

Table 2. Time for a single approval, sorted by duration, and a comparison between approvals without and with intervention.

Figure 2. Time to obtain approvals with and without reminders.

Application

Based on your current MOC approval history, if you compiled data similar to Table 2 then the values of T50 and T90 would immediately become apparent. The parameters you have control over are α and Δt. α, the individual intervention effectiveness, can be increased through education and communication—encouraging people to respond to approval requests. Δt, the interval between sending out reminders, is generally set in the eMOC workflow, and is generally under your control. I would strongly discourage sending out reminders any more often than once a day, in order to avoid user resistance.

So, given values for T90/T50, α and Δt, the beneficial effect of reminders can be determined using the data in Table 3. Let’s try it. Consider the following:

• • The current eMOC system, tracks when approvals are accomplished, and this data is available for analysis.
  • The wait times are calculated, being the difference between the approval timestamp and the approval request timestamp. Only 8 working hours per calendar day are taken into account.
  • The median approval duration, T₅₀ = 20 hr.
  • The time when 90% of the approvals are complete is T90 = 36 hr. So, T₉₀/T₅₀ = 1.8.

The average wait time t₀=21.2 hr. Note that the median time and the mean time are generally not equal.

Your organization wishes to accelerate the approvals, by instituting a daily reminder for tardy approvals. How much will the approvals be accelerated, if approvers are likely to respond to about one-half of the reminders? That is:

• Δt/T₅₀ = 8/20 = 0.4
• α = 0.5

From Table 3, γ = 0.2994.

From the definition of γ,t_α=t₀(1-γ)=21.2(1-.2994)=14.85hr

That is, using the parameters as described in the problem description, the average wait time will be reduced from 20 hr to 14.85 hr.

Table 3. Dependence of net intervention efficiency on intervention interval and individual intervention effectiveness, for various values of skewness ratio.