Traditionally digital automation has been something of a Horizon Three project with the IT teams. Now, however, every CXO is grappling with technologies like RPA, Big Data, AI, Intelligent Process Automation, etc. To complicate the issue, there are too many vendors, no clear rules and too much noise.
At the risk of sounding prescriptive, this article provides an algorithm to help you chart your digital automation agenda.
1. Start with the most important workflows
Put all the technologies aside — Start with the assumption that everything can be automated. Think about the most important repetitive tasks, decisions or processes your team must complete. Think about how having these workflows completed within milliseconds instead of days or weeks would impact your business. Include the revenue impact — in most of my client work we have found that disruptive automation usually shows up as a competitive advantage in marketplace, and not just as immediate costs savings.
2. Add workflows not being done today
One of the key advantages of technologies like artificial intelligence is that they make things possible. Especially, things that are either beyond human limits, or are cost prohibitive if done manually. Any strategic approach to digital automation must include such ever-elusive projects. Invest in workshops for out-of-the-box brainstorming on what is possible and then add all those ideas to this list. Sort it by potential impact in terms of costs, revenue potential or quality of outcomes.
Beyond this point, start with the most important workflows and take them through the following steps, one by one.
3. Automate robotic workflows
If the workflow is repetitive and completely rules-based, it is ripe for Robotic Process Automation (RPA). Examples include web data fetch, mail merge, dispatch, alerts, notices or do-it-yourself tasks for users, among many many others. Usually there is no ambiguity in relationship between inputs and the output of such workflows. RPA is a cheap, stable technology. Usually the biggest factor that makes or breaks a project is the ability to surface and integrate relevant data with RPA tools.
4. Use big data analytics for deterministic cases
Many data driven decisions are pretty straightforward but may involve an unimaginably huge amount of data. These workflows should use Big Data Analytics. Supply chain or production environment analytics based on data coming from a huge range of sensors (Internet of Things) is a classic example. Input to output relationship is well defined for these workflows and does not change over time. A lot of software enabling such analytics claims to be intelligent, but this is still not artificial intelligence. The key factor in determining success of big data projects is usually the quality of data — how clean and well-defined it is to begin with.
5. Use deep learning if there is a lot of structured data
Workflows where the input to output relationship is not well understood are ripe for artificial intelligence. This also holds true for situations defined by the so-called “human intuition”.
Deep learning is the most popular and powerful machine learning technique. It uses multiple layers of relatively simple neural networks to find hidden relationships between a vast set of inputs and some outputs. To train effectively, deep learning needs a large amount of clean, labelled data. Availability of this data and/or data scientists that can prepare this data is usually the biggest bottleneck for deep learning.
There are multiple other artificial intelligence algorithms that apply.
Please note that not all workflows are well suited for AI. In fact sometimes using AI can be counterproductive.
6. NLP is harder but is becoming increasingly accessible
There are many workflows, processes or tasks that involve decisions based on natural language inputs. Interactive chatbots are just one example — think about searching through millions of documents for correct answers, putting together templates and summaries, extracting structured knowledge from free text, or redaction/ classification of sensitive information, among others. Perhaps, the best way to think about this is to imagine what you could achieve if you had a team of 1,000 interns somehow telepathically connected.
Technologies like Calibrated Quantum Mesh or others bring the complete might of AI to language, some without the need for annotated training data. Even deep learning or similar technologies like Watson are successful if you have enough data and the budget for Subject Matter Experts to appropriately annotate that data. However, they will not be our first choice for natural language-based workflows.
7. Revisit whatever is left
At this stage, there may be some workflows that don’t appear to be suited for automation. Typically, these are workflows that involve creativity, innovation or judgment, the quintessential human traits. Think about if any of these workflows can be split or redesigned so that the more boring, tedious or time taking parts can be extracted out, and put back in this hopper. Even partial automation goes a long way, in my experience.
There are a lot of other factors to consider before deciding on the best technology or partner for your automation needs, but hopefully this algorithm can define your agenda for the post-covid-19 world.