Top tips for RPA implementation
ASOS is always experimenting with new tech, and whilst Robotic Process Automation (RPA) is not so new, we believed it could support one of our business strategic pillars ‘to develop an efficient organisation’ as well as add another useful capability into our automation armoury.
We’ve been running a pilot during the last six months and have delivered several process automations across multiple business functions, learning lots of lessons!
Here are my top tips to consider:
#1 — Design with security in mind
Robots need access to systems to automate business processes, with security controls like Multi-Factor Authentication (MFA) becoming the norm as Enterprises migrate to the Zero trust security model, future proof infrastructure designs from the outset.
Apply the ‘least privilege’ approach to minimise the level of access given to any single robot to avoid making it a target for a future exploit.
#2 — The source system team are your friends
RPA predominantly re-uses existing infrastructure to deliver value, so a successful implementation is highly dependent on the source system.
Build a strong working relationship with the team(s) accountable, to ensure RPA doesn’t negatively impact a business process or present new technical risks.
Example considerations:
- A legacy system may be about to retire or have a lack of investment, so what are the known constraints?
- What if the robot doesn’t gracefully exit a process, can it lock a record?
- Are any timeouts in sync between the RPA solution and the source system?
- Is excessive CPU being consumed when the bot executes?
- What monitoring is available?
#3 — To performance test or not?
Some RPA processes are high in transaction volume and/or will require a high number of concurrent bots. A level of performance testing may be required to provide appropriate assurance before go-live, some considerations:
- What non-production environments exist and are they available to use?
- If the end-to-end candidate process for automation connects multiple systems are those environments, and data, in sync?
- Is there a mechanism to generate enough test data?
- Can controlled testing occur in Production? Supported by a gradual ramp up of transaction volumes as an alternative?
- If the tactic to performance test in production is to be deployed, what is your attitude to risk vs benefit? and can the blast radius be contained?
#4 — Avoid uber bots
Get the most from your investment by maximising the utilisation of your bots, but try to avoid any single robot gaining super-user privileges to lots of different systems. You need to design your bots to use the lowest level of privilege, otherwise it’s likely to run into issues with advanced anti-virus or your enterprise cyber team. If a bot was compromised they could gain enterprise-wide access from one account. A good policy to adopt is to group bots by closely linked functions or systems.
#5 — How will your users respond?
ASOS’s culture is fast paced, and we always have more ideas, ambition and demand than available resource! So ASOSers so far have loved the RPA solution because it comes with a promise to save them time and take away mundane admin tasks.
However it is important not to get complacent with the change curve your stakeholders and users will progress through. Ensure users are introduced to what RPA is with supporting real world examples, that they are given the opportunity to contribute from the start for example by helping the team understand the “as is” process and to seek their help to design a better process, will all help the adoption of change enormously. Some users may even want to learn more and try configuring part of the solution.
#6 — Don’t trust the bot too quickly!
Sometimes an end-to-end process still requires a human step, e.g. to approve a change at the end, when this happens implement governance to ensure that step is being followed and that a user hasn’t become complacent and just assumes that the bot will always work with 100% accuracy. Make sure the bot earns the right to be trusted first.
Check the quality of implementation during early go-live support. Error queues help catch missed business exceptions or system errors and offer a good resiliency tactic to progress development at pace whilst not designing for every edge case.
How empowered is a bot? If the answer is a lot, who governs how the bot’s work is triggered?
#7 — Use local domain expertise to help implement
Of course you will need to use subject matter experts to understand your candidate process. But do engage and create a cross-functional team with any local teams within the function, be it business change teams or local tech support teams. They will help you navigate business stakeholders and any existing business or technical change roadmaps. Also alignment with any local nuances that your team will just not have the depth of understanding and expertise to know about.
ASOS is an environment that changes regularly, so we found this cross-functional alignment and local sponsorship was crucial to successful implementation and high levels of adoption.
This may also generate interest and start you on the journey of creating advocates within that function, presenting the option of a federated delivery model or identifying citizen developers who want to learn how to implement RPA themselves.
#8 — Don’t forget to maintain the runway
An automated process is equivalent to an application, but there are lots of surrounding elements to a solution that require focus:
- Stay aligned with company cyber security standards.
- Does a pipeline exist to build a virtual machine for each new bot?
- Is there an automated path to production deployment route?
- How and where are access credentials maintained?
- Who supports the infrastructure?
#9— You are creating a digital workforce
Just like employees, robots have a work lifecycle from hire, induct, train, assign work, performance manage and retirement.
As you scale you’ll want these processes to be streamlined and not cause friction. You never know this might convert into its own RPA opportunity as well!
A bot is still a user, they will need virtual machines, MS office licenses etc. so consider these costs.
What is your onboarding plan? Is it aligned with all the different stakeholders you have to engage? Are lead times understood? e.g. Set-up virtual machines, purchase bot license keys, create MS office & legacy system accounts etc.
What happens when a specific process retires?
#10 — Measure more than time saving
Time saving is a key benefit of RPA. Measure how long a human takes to perform a task and the number of times your bot runs that process to generate an opportunity view.
But don’t just measure time saving, every process is unique and will offer many more secondary benefits.
Was there a positive consequence?
- What did that person do when they were re-deployed?
- Did sales increase when you gained the ability to scale that process?
- What was the return on investment of that new opportunity pursued only because RPA gave the team the bandwidth to do so?
Some other benefit types to consider:
- Speed up processes, shorter time to value
- Increased employee satisfaction (work-life balance, involvement in innovation and creative tasks, skill development)
- Improved customer experience and satisfaction because more responsive
- Error reduction, reducing time spent on rework
ASOS are hiring across a range of roles in Tech. See our open positions here.
Scott Mathers is a Change & Transformation Tech Lead at ASOS.com. In his spare time he enjoys keeping fit, juggling a 2nd career as a Dad taxi and following SpaceX!
