How much of your day is spent on emails? Around half? Three quarters? A 2019 survey by SoftwareONE found that UK office workers typically spend two hours every day browsing their inbox, which over a year adds up to the equivalent of 30 working days.
In light of coronavirus and more people working remotely without face-to-face interactions, we could now be spending around double that time in our inbox.
In March this year email analytics developer PoliteMail found that there had been a near-100% increase in email volume to employees in the US, up from 1.2 million to 2.2 million per day compared with 2019. This included an alarming level of emails sent and received over the weekend – now 1 million compared with fewer than 100 thousand before COVID-19.
It’s not only emails that are taking up our time either. What was once a quick question across the office can now be a lengthy exchange over Teams or Slack.
However you look at it, arguably too much of office-based work hours is spent on communicating through a keyboard. So, what if some of the time you spend typing away to colleagues or employees could be automated?
What if, in your place, an artificially intelligent assistant could respond for you, and help maximise the time you spend on more valuable tasks and face-to-face communication?
By delegating email responses to a computer assistant, the workday could run more smoothly. A study of emails conducted by the University of California, Irvine and the US Army found that without new messages interjecting throughout the day, workers multitasked less, had longer task focus and experienced less stress at work.
We’ve also developed some unhealthy habits when it comes to when and how quickly we respond to emails. According to a 2017 Bupa survey, 75% of British employees said they felt under pressure to respond to emails outside of work hours. A further study from the University of Central Lancashire also found that the majority (58%) of a pool of 1,000 UK employees feel they have to respond quickly when an email comes in.
Though automating emails could mean improvements to the health and productivity of employees, the appetite for this kind of tool in HR is seemingly low. In a snap poll HR magazine ran on Twitter, 69.2% of the community said they would not trust an algorithm to send an email on their behalf. However, almost a quarter said that they would consider it in some cases.
Part of the argument against automated emailing at the moment is an issue of the technology’s maturity, and the perception of what AI actually is, with many of us not realising we already use a version of it.
Predictive text is a kind of AI, and though experimenting with personal messages is more entertaining than practical, it demonstrates that we’re willing to allow machine learning into our communications. Given many emails follow a similar structure from day-to-day, why wouldn’t we do the same for our work email?
Caroline Nugent, HRD of the Financial Ombudsman Service agrees, saying that in some cases an automated response is appropriate. She says: “We absolutely need to free up HR from the transactional non-value-added work so that we can support organisations and individuals with those things which need more empathy or are complex.
“Technology has advanced so that absolutely standard responses should be replied to automatically, for example, if someone wants to know how many days carers’ leave they can have or where a manager can get support for a specific issue for their direct report.”
She admits however that the technology will need further development if it is to win HR’s trust: “We need to ensure that there are no inherent biases in the technology, and I don’t think we are there yet. There has been lots of talk about HR having more AI but until we actually see lots more of it around it will be the art of the possible rather than the reality.”
Writing in the Harvard Business Review, H James Wilson, managing director of information technology and business research at Accenture Research, and Paul Daugherty, Accenture’s group chief executive for technology and chief technology officer, argue that language will be the next big breakthrough in AI.
Speaking to HR magazine, Wilson says: “With the previous generation of AI and analytics, we saw a wide impact on business tasks that were quantitative or statistical in nature. That earlier generation of AI had many applications in financial trading or the statistical control of factory operations. But the majority of business tasks require a more flexible, language-based reasoning that’s context aware.”
The next big possibility he adds is that a new class of ‘natural language’ AI, i.e. one that can communicate in a way that sounds human and makes logical sense, will be used to help workers to augment or automate 50%-90% of their daily language-based tasks or, as Wilson describes it:
“Any tasks that are about persuasion, creativity, or the application of expert knowledge that’s not simply mathematical…or reducible to statistics.” Such tasks include drafting a persuasive job advert, updating code for a website, or responding to emails.
Typing less and talking more
While a tool capable of performing language tasks in a way that is believably human is still a little way off, many of the big tech firms such as Google, Facebook and Microsoft are already working on their own versions.
One of the first examples of a tool that can help do this is a model called GPT-3, developed by AI research laboratory OpenAI which can generate human-like text.
OthersideAI is one of a handful of start-ups that has early access to GPT-3 in order to come up with helpful ways to use the technology. Founded in July this year, the team has developed a plugin that can automatically write emails for you.
It works by generating sentences based on a few brief bullet points of what you’d like to write. For example: “Thx, lmk, talk soon” becomes “Thanks! Let me know when you want to reschedule a follow up call. Talk soon! Best, David”
The time this saves per email may be fractional, but over the course of the day it could add up to potentially hours of time saved to be redirected to other tasks.
Matt Schumer is co-founder and CEO of OthersideAI. He says tools like the one his company has developed can go in one of two ways: “One it’s going to replace human-to-human communication and that’s not a good thing.
“The other way is that it actually supports human-to-human communication and that’s what we want to do, in the sense that we spend too much of our life emailing as opposed to actually talking.”
The way Schumer describes the OthersideAI plugin is as “an assistant to a human.” He says: “You tell it what you want and it’s like a next-generation search engine or tool that gives you the information you need.”
Rebekah Tapping, group HR director at insurance provider Personal Group, also sees potential for AI to be used in contract writing. She says: “If there were systems where you could input contract information, and the technology could adapt to the specific situation, that would save the most typing time for HR teams.
“They could then instead focus on reviewing final versions.”
There are AI tools available now that can help people draft whole articles. Journalist John Seabrook conducted an experiment with GPT-3 to see if it could be taught to mimic his style and effectively predict what might come next in a piece he was writing for The New Yorker.
While GPT-3 ultimately failed, some of the samples of its writing showed potential, producing close parallels with Seabrook’s work, yet needing intervention to be coherent.
The approach then becomes more about human-AI collaboration.
Kessar Kalim, HR director at the London School of Hygiene and Tropical Medicine, says: “If you can get an AI function to write an article that’s almost identical to a human that’s impressive, but I’d imagine there still needs to be some human input.
“The AI will take it to 60% or 70%, and then the human will take it to the full 100% because say an external event happens, politically - the AI function won’t respond to that whereas a human can.”
The missing link is developing AI that can generate its own ideas and understand the content it is writing.
Kalim adds: “One of the differences between humans and AI is the reflection, and it’s the feeling. When you do something that’s quite sophisticated, you need to have a level of introspection, and intuition, nuance – if you can get an AI to replicate all that – which is responsive to changing environments and changing issues – then you’ve cracked the code.”
The ethical line
There are of course many instances in which it would be unethical to send an automated response to an employee or work email.
In some instances, Kalim argues that an automated response would be offensive to the recipient. He adds: “For something that’s factual you don’t mind an automated or an AI-type service.
“But if it’s something a bit more personal, about your feelings, your values, and you want a discussion with somebody to explore options and talk through a business case for example, I don’t think you can replace HR or humans with AI.”
Similarly, Tapping argues that it would be difficult to trust an AI platform to send work emails. She says: “Employees are all individuals, and there will never be a time when communications do not need a human pair of eyes.
“It’s a bit like driverless cars: you may allow the vehicle to have a go, but still keep your hands close to the wheel. There may be instances where AI is useful, but it will never erode the personal touch.”
On the other hand, speaking in relation to sensitive issues like employee health, Nugent believes that there is still some space to explore AI’s potential: “I would not expect it to be used for individual medical responses.
“If for example somebody was having a mental health incident, I would want a human on the end of the call who can gauge how much support is needed and from where. Let’s get the basics right where full confidence is there and then we should in future have the ability to do so much more with what is an exciting opportunity.”
As it is still early days for this technology, Accenture’s Wilson advises erring on the side of caution and responsibility. For HR leaders and practitioners, he proposes thinking about the application of natural language AI as a spectrum. On one end he says are the situations in which you would never use automated responses, i.e. for issues relating to safety, security, or employee health and wellbeing.
He says: “These tools would not be appropriate in the near-term as far as I can see. Similarly, any communications that involve making a final decision, for instance, about whether to hire, sack, or promote someone. We make mission-critical, non-routine decisions through email all the time, and these cannot be automated.”
At the other end of the spectrum are the more routine activities that could always benefit from automation such as scheduling meetings. Then, in what Wilson terms the “broad middle,” it is about “augmenting people” rather than automation and creating human-machine collaboration.
As an example, he suggests GPT-3 could be used to read a colleague’s email and suggest intelligent follow-up questions about a promotion candidate’s background or fit for a new role. Or it could be used to review all recent scientific research on employee mental health during COVID-19 and identify evidence that could help write a business case for a new wellbeing campaign.
The opportunity, he argues, is in highlighting information that people might usually miss, especially when it comes to volumes of text. He says: “The exciting thing is that these new AI tools will augment humans, making suggestions, spotting patterns and opportunities in emails that the HR practitioner didn’t see.
“But it’ll be the HR practitioner’s decision on whether or not – and how – to use the AI’s suggestion within their specific role.”
Though there will be more opportunity for automated emails in the future Wilson adds the caution that there is also a risk that such developments could lead to a net or exponential increase in the number of sent emails, creating more of a burden long-term rather than cutting it down.
A wider conversation about email culture is perhaps needed to take meaningful steps to reduce our reliance on them, and shift expectations around when and how we should respond.
In the meantime, advocating a few hours off from emails each day to focus on the task at hand could be a way to start tackling our unhealthy obsession with the inbox. And trying things like templates may be a better (and safer) option for saving time. At least for now.
The full article of the above is published in the 2020 Technology Supplement. Subscribe today to have all our latest articles delivered right to your desk.