DORA Metrics
37 minutes reading time

The Financial Impact of Continuous Improvement: Maximizing Software Delivery Performance

The potential return of continuous improvement in software development

Continuous improvement is an objective that we all want to include in our daily lives. However, many of us "de-prioritize" continuous improvement when we have fires to put out or concrete value-added tasks to perform for our colleagues. But did you know there is real financial value to continuous improvement in software development? In this blog post, discover the potential return of improving your software delivery performance.

The methodology behind the findings

To estimate the potential return of continuous improvement in development teams, we will primarily use the four key metrics from the State of DevOps report by DORA (DevOps Research and Assessment), a team at Google that studies the performance of technology industry professionals.

The four DORA metrics are separated into two categories: speed and stability. After surveying more than 32,000 professionals and organizations, the DORA team was able to distinguish four performance groups (elite, high, medium and low) with associated benchmark results.

Delivery performance aspect Elite High Medium Low
Deployment frequency
(how often the organization deploys in production to the end-user)
On-demand
(several deployments per day)
Between 1x per week and 1x per month Between 1x per month and 1x every 6 months Less than once every 6 months
Lead time for changes
(the time between the first commit and delivery to production)
Less than an hour Between 1 day and 1 week Between 1 and 6 months More than 6 months
Mean Time To Recovery (MTTR)
(the time required to repair the service following an incident)
Less than an hour Less than a day Between 1 day and 1 week More than 6 months
Change failure rate 
(the percentage of changes in production that cause an incident)
0 to 15% 16 to 30% 16 to 30% 16 to 30%

Source: State of DevOps 2021, Google. Note: Analysis of the results in the 2022 edition of the State of DevOps report detected only three groups: high, medium and low performance. To adequately demonstrate the magnitude of the potential return from a high-performing team, we'll use the 2021 data featuring the original four groups.

The impact of improved performance and continuous improvement on potential return

In terms of added value

To create maximum value for the organization, the best teams understand that they must optimize the speed of delivery. Value is lost in opportunity costs or when the effort is spent on non-value-added work (e.g., unnecessary rework or manual testing) that could be spent on value-added work (e.g., new features). There is also a loss of value when the deployment of a new product or feature is delayed, including a loss of revenue or customers that the company is not generating but could have attracted by delivering sooner.

The ability to discover and deliver faster to your customers is a substantial competitive advantage that the Lean and Agile paradigms provide. They will remain advantages for years to come.

In terms of avoided costs

The costs avoided by the company are considered a return because they directly impact the revenue and expenses of the original budget. So, for example, if you have an initial budget of $100 million in expenses, and by improving a technology, you now have expenses that amount to only $80 million, you have $20 million available that was not initially planned.

To calculate the potential return of improvement, we will add the value recovered in the unnecessary rework avoided, the potential added value of reinvesting this time in new features and the cost of the failures that can be avoided.

A concrete example of the potential return from improved software delivery performance

Consider Michael, an engineering manager who manages a team of 50 people for a $10 million product. According to the DORA metrics he actively tracks, his team is currently in the medium performance group. What is the payoff if he can get his team to perform at a high level? Let's do the math to find out!

The value recovered in unnecessary rework avoided per year

We need to recognize the value of the time people spend working, which is recovered by reducing inefficiencies. For example, organizations get additional capacity without recruiting and hiring by improving processes. 

Google's research also shows that improved DevOps practices lead to higher employee satisfaction and that employees on high-performing teams are 2.2 times more likely to recommend their organization as a great workplace.

We will find the value of the unnecessary rework avoided using this formula:

Capture d’écran, le 2023-03-14 à 09

Source: The ROI of DevOps Transformation, Google

Let's analyze each element of the formula in more detail.

Technical staff size

You can use the number of technical people in your teams. For Michael's example, we have already determined that his team has 50 technical employees.

Average salary

Salaries vary greatly depending on the company's location and the cost of living. You can use your data, but for this example, we will use Michael's average annual team member salary of $85,000. This is a representative example of the market based on data reported by Glassdoor for developers in Canada.

Benefits multiplier

In addition to salary, the organization contributes to the cost of employee benefits (e.g., insurance, vacation, retirement plans, etc.), usually added to the salary. Each organization has a percentage of about 15% for small companies and 30% for larger ones. So, for example, for a salary of $100,000 annually, you will need to add between $15,000 and $30,000 in benefits. This is a multiplier of 1.15 to 1.3 of the base salary. For this example, we will use the data provided to us by Michael, which is a multiplier of 1.15.

Percent of time spent on unnecessary rework

This number represents the amount of time spent on non-value-added work, or in other words, the hours worked that are essentially lost to inefficiency. Our reference here will be the 2018 State of DevOps report, where the research shows that even the best teams have a rework percentage of 19%, and we can aim for an improvement target of 18%. So we will use 1% for the elite teams' formula, which is the difference between the goal and their reality. The high-performance teams reported a 19.5% rework percentage, while the medium and low-performance teams reported a 20% rework percentage. We will therefore assign them values of 1.5% and 2%, respectively.

This variation in the percentages of the four groups (1%, 1.5% and 2%) will allow us to compare the cost of unnecessary rework per year by team performance level, all other variables being equal.

So, if we apply the formula, we get the following results:

Elite performance High performance Medium performance Low performance
50 employees x
$85 000 salary x
1,15 benefits x
1% rework
= $48 875
50 employees x
$85 000 salary x
1,15 benefits x
1,5% rework
= $73 313
50 employees x
$85 000 salary x
1,15 benefits x
2% rework
= $97 750
50 employees x
$85 000 salary x
1,15 benefits x
2% rework
= $97 750

Since Michael's team is in the medium performance group, he would get a potential return of $97,750 annually by recovering 2% of his time from unnecessary rework

The potential added value of reinvesting the recovered time in new features

The idea is to take the time recovered from reducing inefficiencies and turn it into value by using it to generate revenue through new customer features. Team inefficiencies create opportunity costs and put the potential to add more value to the product at risk. While delivering new features brings revenue, not all features are winners. Several experiments at Microsoft in collaboration with Stanford University have shown that only one-third of features succeed. The most successful companies leverage their ability to deliver quickly to production to experiment.

We will base the revenue potential on the company's current revenues for our calculations. Therefore, we will use this formula:

Capture d’écran, le 2023-03-14 à 09 (1)

Source: The ROI of DevOps Transformation, Google

Let's analyze each element of the formula in more detail.

Time recovered and reinvested in new features

We will use the same methodology as in the previous section (e.g., 2% for teams with a medium performance like Michael's). It is the same data since the recovered time is the time we can reinvest.

Deployment frequency

This is the ability to test features on customers. So we reuse the deployment frequency based on the team's performance. Elite teams experiment 2 times a day (or 730 times a year). High-performing teams experiment between 1 time per week and 1 time per month (on average 32 times per year). Medium teams experiment between 1 time per month and 1 time every 6 months (average 7 times per year), and low-performing teams experiment less than once every 6 months (average 2 times per year).

This variation in the number of deployments for all four groups will allow us to compare the value of experimenting more often by making good use of the time saved by eliminating unnecessary rework.
Lines of business

Organizations may have multiple business units where each unit has a central product to serve customers. We can estimate an average of twenty business lines for large organizations (more than 8,500 employees). In comparison, for medium-sized organizations (more than 2,000 employees), we can estimate an average of eight business lines. We can estimate one line of business for small organizations (250 employees or less). This is the case for Michael, and this is the data we will use for our example, but we invite you to use your figures.

Idea success rate

According to the Microsoft study, only one-third (33%) of the features sent to production positively impact key metrics. We will therefore use this figure for the calculations.

Idea impact

Each successful change has the potential to help generate more revenue or profit for the company. Some features can contribute up to 200% of revenue, while others will not (e.g. 0.01%). On average, we will estimate that an idea contributes to 1% of revenue.

Product business size

The potential revenue from new features is estimated based on the current revenue of the product. For our calculations, we will use the product Michael's team is working on that generates $10 million in revenue.

So, if we apply the formula, we get the following results:

Elite performance High performance Medium performance Low performance
1% recovered time x

730 experimentations x

1 business line x

33% success rate x

1% impact on revenue x

$10 M in revenue

= $240 900
1,5% recovered time x

32 experimentations x

1 business line x

33% success rate x

1% impact on revenue x

$10 M in revenue

= $15 840
2% recovered time x

7 experimentations x

1 business line x

33% success rate x

1% impact on revenue x

$10 M in revenue

= $4 620
2% recovered time x

2 experimentations x

1 business line x

33% success rate x

1% impact on revenue x

$10 M in revenue

= $1 320

By improving its performance, Michael's team could experiment more and get a significantly higher potential return with the avoided rework time reinvested in working on new value-added features.

The cost of downtime avoided per year

Applications and infrastructure can suffer incidents that bring service down, and these outages cost the organization. According to a report by Stephen Elliot, vice president at IDC, a business intelligence firm for the technology sector, the cost of an outage per hour can range from $1.25 to $2.5 billion for Fortune 1000 companies. However, these figures vary widely depending on the nature of the business.

To find the cost of these failures, we will use this formula:

Capture d’écran, le 2023-03-14 à 09 (2)

Source: The ROI of DevOps Transformation, Google

Let's analyze each element of the formula in more detail.

Deployment frequency

The deployment frequency can affect how often the team risks introducing a change that causes an incident. Referring to the table above, elite performers deploy 2 times per day (or 730 times per year), high performers deploy between 1 time per week and 1 time per month (on average 32 times per year), medium performers deploy between 1 time per month and 1 time every 6 months (on average 7 times per year), and finally, low performers deploy less than once every 6 months (on average 2 times per year).

Deployment frequency graph in Axify

Deployment frequency graph in Axify

Change failure rate

This is the percentage of deployments resulting in a production incident. To find this number, we will use DORA metrics. Elite performers have a failure rate between 0 and 15% (average 7.5%), high performers have a failure rate between 16 and 30% (average 23%), medium performers have a failure rate between 16 and 30% (average 23%), and low performers have a failure rate between 16 and 30% (average 23%).

Mean Time To Recovery (MTTR)

When there is a failure or degradation, this is the time required to correct the situation. To find the number, we'll use DORA metrics. Elite performers recover in less than an hour (average of 0.5 hours), high performers recover in less than a day (average of 4 hours), and medium performers recover between a day and a week (average of 24 hours). Finally, low performers recover in over six months (average of 180 days, or 4320 hours).

Outage cost

Outages can be costly for the organization. The cost of an outage can vary greatly depending on its nature and level of degradation. You will need to use your data to refine the calculation. According to Steven Elliot's report, an infrastructure failure would cost an average of $100,000, while a critical failure would cost an average of $500,000 to $1M. Michael confirmed that $1,000 was a representative amount for his team. This is the figure we will use for the calculation.

Therefore, it is the level of performance of the team that will determine the cost of an outage within the company.

So, if we apply the formula, we get the following results:

Elite performance High performance Medium performance Low performance
730 deployments x
7,5% failure rate x
0,5 h to recover x
$1 000 cost

= $27 375

(or $38 per deployment)
52 deployments x
23% failure rate x
4 h to recover x
$1 000 cost

= $29 440

(or $920 per deployment)
12 deployments x
23% failure rate x
24 h to recover x
$1 000 cost

= $38 640

(or $5 520 per deployment)
2 deployments x
23% failure rate x
4 320 h to recover x
$1 000 cost

= $1,987 M

(or $993 600 per deployment)

You'll find that even with more frequent deployments (and thus a greater chance of introducing changes that cause an incident), teams that continuously improve toward elite performance experience much less costly outages.

So the solution to reducing the cost of outages is not to reduce the frequency of deployment but rather to reduce the change failure rate and MTTR.

Combining all of the above

Now that we have identified the major costs and value components, we can combine them to find the potential return of having a higher-performing team.

Capture d’écran, le 2023-03-14 à 09 (3)

Source: The ROI of DevOps Transformation, Google

So we will add up the variables found previously:

Elite performance High performance Medium performance Low performance
$48 875 of avoided rework +

$240 900 of rework reinvested in added value +

$27 375 in costs related to avoided outages

= $317 150
$73 313 of avoided rework +

$15 840 of rework reinvested in added value +

$29 440 in costs related to avoided outages

= $118 593
$97 750 of avoided rework +

$4 620 of rework reinvested in added value +

$38 640 in costs related to avoided outages

= $141 010
$97 750 of avoided rework +

$1 320 of rework reinvested in added value +

$1,987 M in costs related to avoided outages

= $2,086 M

We can see the real impact of improving certain aspects of a development team's performance and the potential return from undertaking a continuous improvement process.

If Michael's team gets better at optimizing the development flow, it can make changes quickly and frequently to deliver value more often.

Getting 2% of your time back from unnecessary rework avoided is a worthy goal, but you must also ask yourself what you can accomplish with the reclaimed time. The chart above clearly demonstrates the correlation between better performance and potential return if it is invested in adding value to the product.

The further a team progresses toward elite performance, even if they recover less unnecessary rework time avoided (1% vs. 1.5% or 2%), they can deliver more value with that time, since their outages are less costly and they deploy changes more often. As a result, these teams spend the most time adding value and the least time on non-value-added tasks.

Regardless of where you are currently in the table (and perhaps you are at different levels for each of the metrics), moving a single column to the left for any of the DORA metrics can significantly impact your organization.

Now what?

Improved productivity can bring a return to the organization by providing better capabilities and performance. But where to start? To improve, we need to set goals. And to set goals, we need to understand our initial situation. Axify is a tool that allows you to harvest data from tools you already use to easily view your flow metrics, team morale and, of course, DORA metrics. You can also create goals from your data and track their progress over time to see your team's continuous improvement.

Page sommaire - Objectif - Version anglaise-1

For more information on how Axify helps development teams measure DORA KPIs, start your free trial or request a demo.