Background
Technology evangelists have a vision and that vision is of a future where repetitive, routine tasks are undertaken by technology leaving humans to focus on value-adding, creative activities. This applies as much to knowledge work (such as that carried out by data and analytics practitioners) as it does to manual work (such as manufacturing or logistics).
According to a Gartner report in late 2016, 30% of large enterprises were forecast to employ some form of “smart machine” (cognitive computing, artificial intelligence, intelligent automation, machine learning and deep learning) by 2021. The report predicted these smart machines would open up an entirely new industry, expected to be worth approximately £23.3bn in three years’ time.
The need to automate is obvious to those in the data industry even when just considered around the issue of scale. In digital channels, for example, multivariate and A/B testing has now grown beyond human capabilities, so machines are necessary to conduct the analysis. The question is then how much human in the loop is necessary.
Even so, a recently-published report by Microsoft, “Maximising the AI opportunity,” found that half (51%) of UK organisations do not currently have an AI strategy in place. Meanwhile, 51% of employees and 49% of leaders admitted they are not currently using any form of AI to perform tasks at work.
Where automation is being addressed, based on insights from the DataIQ Leaders membership, there are three domains being tackled under the broad umbrella term:
- Robotic process automation (RPA): automating routine, repetitive tools currently requiring significant human labour;
- Artificial intelligence (AI): applying machine-based intelligence to tasks where scale, speed or complexity make human-based intelligence impractical, or to replicate human intelligence;
- Machine learning (ML): deploying deep learning techniques in a research and development (R&D) environment to extract new insights, processes or opportunities.
While the need to apply these techniques may be pressing, the business case for their use can be harder to make, especially within legacy organisations or where a transformation is being undertaken that means baselines are hard to establish.
Organisations doing AI outperform others by 5% on productivity
Microsoft offers some useful indicators in its research which shows that organisations already on the AI journey are outperforming others by 5% on factors like productivity, performance, and business outcomes. It also reveals that organisations investing in establishing the right approach to AI technology – specifically, by developing underlying values, ethics, and processes – outperform those that are not by 9%.
Context
To understand how the DataIQ Leaders membership is responding to the opportunities and issues around automation, we convened a roundtable in October 2018. Led by a senior practitioner from a broadcast media company and with representatives of media, retail and logistics, the conversation provided good insights into the scale and scope of what is being addressed through the use of “smart machines”. This table summarises the projects being run by those present:
Key learnings
1. Low hanging fruit and reverse Trojan horses
To get automation projects started within an organisation, a number of strategies can be adopted. The most common is to identify low hanging fruit – a target process which is either currently expensive to operate or not performing as desired. By applying automation, a use case can be made that acts as a proof of concept for an extension of smart technology into the organisation.
One grocery retailer has the benefit of a view at board level that automation is the right thing to do and has already landed some easy wins. In one example, a store trading report was being created 364 days a year and required one full-time employee to deliver. That person was a qualified accountant, but was not able to add any value to the report which was manually emailed to a distribution list.
This report has now been automated with zero human intervention and is achieving 99.9% accuracy. Prior to the creation of its centralised data and analytics function, the retailer had tried and failed to automate this task for ten years. The reason for this failure was the way the challenge was being looked at – automation of a flawed process, rather than scoping out what was needed and automating it as a new process. This has saved 600 hours of human effort each year.
Where there is no executive buy-in or obvious easy wins, other routes may be available. At one broadcast business there is a formal R&D function which has a budget for innovation, for example, so it has been asked to look at automation opportunities around management information. It is not uncommon within larger organisations for some technical functions to be allowed to spend c15-30% of their resources on innovation without the pressure of a specific outcome. Automation can play into this open space.
Automation can also arrive within an organisation as the unexpected consequence of other actions. Such reverse Trojan horses can happen when “rogue marketers” bring in solutions from non-approved vendors that have embedded automation tools (typically AI) – some of these should be supported, but others need to be shut down. This can also happen through acquisition which brings other tech stacks into the organisation.
2. Automation and human capital
Automation has an impact on human capital and resources within organisations that adopt it, either through a reduction in headcount as smart machines replace manual labour, or by changing the skills profile required through increased complexity.
20% of administrative tasks will be automated.
In a recent report by technology consultancy Cognizant, “21 more jobs of the future,” it is predicted that 20% of administrative tasks will be automated, but a new class of job will emerge. Among those relating to data and analytics which it forecasts will emerge are chief purpose planner, algorithm bias auditor and virtual identity defender. Training and recruiting for these could prove challenging, given the current lack of obvious career paths into them, although reselling seems a likely solution.
Reduction in total headcount is not necessarily the only outcome from automation. At one logistics company which has a unionised workforce and operates the largest road fleet in Europe, the intention is to share the benefits of automation by reducing the working week from 40 hours to 35 eventually. It also has a recruitment issue to confront – the average age of drivers is rising and younger workers are not choosing commercial vehicle driving as a career. That makes progress towards automated vehicles essential for its business survival.
For both physical and digital retailers, stock maintenance and shelf stacking are major labour overheads, while ensuring on-time replenishment has a big impact on customer satisfaction. Both types have to manage inventory and predict demand, while also trying to optimise revenues by identifying the profitability of customers and product lines.
3. Golden domains and third-parties
Some automation projects require large data sets for machine learning, such as IVR. Third-parties will tend to have created these across multiple client engagements (such as language bid data), so it makes little sense building these in-house. That leaves analysts free to explore more complex challenges, such as the reasons for call abandonment, which require pulling together multiple different data sources and testing models.
In choosing whether to work with an external agency, organisations need to consider whether there is a “golden domain” that they do not want to share with a third-party because of the risk of losing intellectual property. For retailers, store planning is one of those, since handing the task to a commercial partner would make them too important to the business.
Working with external agencies also changes how an organisation thinks about the structure of its data and analytics teams, such as whether it is necessary to employ data scientists or data analysts, provided the company has data engineers who can create data assets and services.
Among the DataIQ Leaders membership, it was considered to be very important that analysts remain close to the colleagues in the business who carry out the processes being focused on. Analysts need to understand the real-world business issues or the solutions they propose may be rejected. Perfect algorithms don’t always work in real life.
4. Blockers and the problem with Phil Collins
One broadcast business decided five years ago that it needed to avoid the “Kodak moment” of being reliant on linear broadcasting and advertising revenues by making a move into data-driven services and digital apps. While it has invested in this resource, it has actually seen ad revenues rise. What was a driver of change has become a handbrake because EBITDA is affected by investment into data and analytics – this has a direct impact on the CEO’s bonus. Yet the opportunity is still there, not least because smart speakers have put broadcasters back into rooms. Brands have a key role to play as the service which consumers request from their voice-activated assistant.
Automation does not always deliver the desired customer experience. For example, one-third of of listening on Spotify is to human-curated lists created by celebrities and acts. This underlines the value to service users of the “human in the loop”.
According to one broadcast business, applying machine learning to content recommendation has been less effective than expected, not least because they tend towards homogeneity – playlists on mainstream AOR channels which are automatically generated will tend to over-sample Phil Collins, for example. so conventional techniques are still being used alongside human curation.
Automated decisions also need to be capable of explanation under GDPR. Black box solutions may not be compliant and could also lead to decisions or processes which are not behaviours which are desirable – the algorithms developed by both Amazon and Facebook among others to target recruitment ads are both cases in point, since they were found to be prejudicial against women candidates
Another risk is that external agencies tasked with developing automation may not be willing to explain how their solution works because they want to protect their IP. That could put the client in breach of GDPR if an automated decision is challenged by a customer.
Conclusion
While technology advocates see the automation revolution as occurring in the near-term, adoption in the real world – and especially within legacy organisation – is likely to be more measured. Even in the face of supposed market disruptors, there remains a strength and value in established brands. So while Cognizant is predicting the need for new jobs such as flying car developer or vertical farm consultant, automotive manufacturers are more focused on drive-time forecasters for electric vehicles or supply chain optimisers in field-to-fork logistics.
Experimentation and early use case are important and it is clear that failing to build a base of knowledge, skills and experience at this stage will be a handicap in adjusting to an automated future. A combination of upskilling and reselling with the use of third-party partners will be useful in building out this core capability. Reaching the scale of payback which the level of investment into full-scale automation requires may prove harder for legacy organisations than it is for start-ups backed by optimistic investors, however.
