Who's automating the enterprise? Meet Amelia and the future of work

Imagine all enterprise functions automated by software and performed through a single point of access, which happens to be a virtual agent with cognitive capabilities. You can stop imagining and start thinking about the repercussions, because this is much closer than you may think.
Written by George Anadiotis, Contributor

If you are not into enterprise software, chances are the name IPSoft does not mean much to you. IPSoft however has been around since 1998, and its mission is to "power the world with intelligent systems, eliminate routine work and free human talent to focus on creating value through innovation". This may sound vague or over-ambitious, but it's quite real.

While you probably use Siri or Cortana or Alexa and have at least heard of IBM Watson, again, chances are the name Amelia means nothing to you. Amelia is the closest IPSoft has gotten to a claim to fame beyond the world of enterprise software. Amelia is IPSoft's virtual agent, and has a reputation for being the most human-like of its kind.

Those intelligent systems in IPSoft's mission statement are a reference to IPcenter, what IPSoft calls an autonomic IT management platform. Broadly speaking, autonomic systems are about making adaptive decisions using high-level policies. For IPcenter this translates to components called virtual engineers that automate a big part of IT infrastructure management.

Now IPSoft has just announced 1Desk, which is essentially a combination of Amelia with IPcenter that aims to take it to the next level. Think the entire stack of enterprise software integrated and interactions with it automated and streamlined via Amelia.

IPSoft CEO Chetan Dube and ZDNet had an in-depth discussion about how it all works and what it means for the enterprise, the work force, and society at large.

Connecting applications

It all goes back to those virtual engineers. These software components have been developed by IPSoft to automate and optimize as much as possible of the underlying IT infrastructure for enterprises. But network and hardware operations are really just the tip of the iceberg in the enterprise. Dube elaborates:

Let's take any operation, HR for example. For all of them there are software products to support them. How does it work? Let's say you want to register some sick leave -- your request is handled through HR people and it generates some IT operation in the back end.

But any time a request comes in, there is manual processing in the middle layer. The time of resolution in a chain is determined by the weakest link, and that is the manual processing in the middle office. Front and back office convergence has not really been achieved yet, but 1Desk achieves that. It's cognitive on autonomic: it connects the end user with the applications via a conversational front end.


You don't need bots with brains the size of a planet to build bridges between enterprise software, but it's not easy feat either.

Part of that is really not all that new or cryptic: it's called Enterprise Application Integration (EAI) and has been around for years. EAI has come a long way, but as anyone who has worked in the field knows, integrating an array of applications that were not really designed to work with each other can be hard.

The good part, Dube says, is IPSoft already has a number of connectors to popular enterprise applications like SAP or PeopleSoft. The goal is not to recreate these applications, but to make them work together and offer one go-to solution for all your affairs. You can think of it as an enterprise app store meets Business Process Management to create end-to-end solutions.

A brain the size of an enterprise

The really new part in this approach is Amelia. Dube takes pride in saying Amelia is the most human AI agent out there. But how do you measure that? Using IQ for this purpose is debatable even when applied to humans, let alone AI.

Recently researchers have tried to do just that however, to conclude that most AI agents have the IQ of a 5-year-old at best. "How is a 5-year-old supposed to deliver value in the enterprise?" wonders Dube. He continues:

These agents can't do much more than walk you down a set decision tree. The moment you stray, you hit a dead end -- the classic example is switching your order from pizza to ice cream.

Agents are supposed to parallel human intelligence, but they are not much more than a smart IVR system. Sure, they have natural language understanding, but what they essentially do is use a deep learning classifier to chuck away your request in one of their buckets.

The criticism may not be unwarranted, but begs the question: what does IPSoft do differently with Amelia?

Also: Inside the black box: Understanding AI decision-making

Dube says their approach has been to try and understand human neocortex activity and mimic that:

We wanted to see what makes humans more versatile, and mimic the outcomes and the abstractions of our semantic memory, our episodic, process and analytical memory, all of the things that make us human. If you do that, only then will you be able to consider Turing's question 'can machines think'.

"Can machines think" is the essence of the Turing test, but for now emulating behavior the way Amelia does may be enough. Image: IPSoft

That sounds far out, and perhaps scary too. Not to mention, people have been trying to mimic the human brain for some time now, and failing. "We are not trying to re-create god's creation. Our approach is to study the human brain to mimic the way humans are able to make flexible decisions. Nothing to do with IBM's neuromorphic chips, for example," says Dube.

So what is it then? Even though looking under the hood for Amelia was not an option, IPSoft explained that Amelia utilizes advanced machine learning models to make business decisions drawn from conversational data in real-time. IPSoft also said Amelia measures her own business impact across a wide range of metrics and suggests opportunities for improvement.

As for the training part? Dube explains:

Amelia has several learning approaches. She reads documents, she learns by observation and is also taught processes by business analysts. Once she is working she is always learning from observation -- so there is dynamic automation being created all the time at the cognitive level and also the autonomic level. We always recommend a governance process to check what she's learnt.


"We question ourselves. Flying succeeded when we stopped trying to emulate birds and applied the principles of aerodynamics. Your brain can switch contexts because it activates a different set of neural connections that enables you to switch to a new topic while continuing the thread of the previous conversation. This is fundamental to drive net promoter scores," says Dube.

Net promoter scores are important in the context where Amelia has primarily been used up to now, which typically includes operations like customer support. The list of happy clients of Amelia includes names such as Credit Suisse, Electronic Arts and MetLife. Since we have no way of directly evaluating Amelia, we'll have to take these as proxies.

According to IPsoft, the implementations to date establish that Amelia can handle over 60 percent of client requests without external assistance, achieving over 90 percent accuracy and customer satisfaction ranging up to 88 percent. Cases that cannot be handled are handed over to a human agent.

Furthermore, IPSoft seems to be an analyst darling, ranked high by the likes of Gartner, Forrester and Ovum. Ovum for example calls Amelia "more than a chatbot" and notes it spans the areas of process automation, conversational interaction and AI.

Amelia is part of an elite club of players competing in enterprise automation. IBM's Watson is probably the most well-known among other options that also include established service providers like HCL, Wipro, TCS, and Infosys along with niche players like Enterra and Digital Reasoning. Sooner or later competition could also come from other enterprise software solution providers.


Virtual agents have been evolving, and evolution is accelerating. Image: Gartner

IPSoft seems to have established working relationships with a number of heavyweight consulting companies such as Accenture and Deloitte, which should help give it traction. In a recent analyst discussion, a Watson versus Amelia comparison has been likened to comparing IQ to EQ. Albeit not entirely analogous, the comparison draws on the intangibles of interacting with Amelia.

But can intangibles be measured and evaluated, or should they? Dube says that another metric to take into account is financial impact: "You'd be hard pressed to find another company having an impact of half a billion dollars through its AI services. Seven out 10 top insurance companies and six out of 10 top banks are our clients."

1Desk however is at this point still under development, with one beta customer -- Carestream Health. So how has it been developed and tested so far? The answer is, internally. IPSoft eats its own dogfood and is itself an avid user of this technology. Which brings us to an interesting topic.

Wait, let's think about social impact -- said no CEO ever

That's all fine and well if you are an enterprise. Who would not want to automate everything that can be automated, decrease cost and increase customer satisfaction? There's just one problem: what happens to the people whose jobs are made redundant? How does Dube respond to that sentiment, presumably also at some point expressed by IPSoft's own employees?

Technology creates jobs. Most people were farmers before the industrial revolution. Now some people write on ZDNet, others are doing other things, all of that would not have been possible without technology. What worries me however is speed -- the speed of mainstream adoption of cognitive and autonomic technology.

This is not a multiplier of muscle, it's a cognitive multiplier, and it's instant. You can take the work done by tens of thousands of people and have it rapidly switched over. The speed of adaptation will have to greatly surpass what was required for example for a steam engine.

I do fear there is the possibility of job diplacement. If people such as yourself are not able to wake up the masses, there is a possibility they will be caught on the wrong foot. This needs to happen now. In the UK for example we discussed the resurgence of interest in vocational training with the house of commons. We need to re-educate to thrive in the digital economy.

In France they are discussing reduced working hours. There are also proposals to tax automation agents -- we should do that, we know productivity is going to increase by at least 40 percent. People should have creative freedom, and engage in the kind of things no robot can do.

A digital Darwinistic curve is emerging, companies and nations should determine which side of the curve they are going to be on. If history is any indication of what is about to happen, Digital Transformation will disrupt in a big way. More than half of Fortune 500 in the year 2000 are not there today.


The evaluation of automation as an important trend is not that high: at most 54% of respondents in a recent Deloitte survey ranked this as important or very important. Interestingly, India and China are on top of the list, while the US is in the bottom. Source: Deloitte

And what about people? My son asked me the other day -- dad, are you going to be a robot? It's scary. You may have your own interpretation of the future -- utopian, dystopian, whatever. I ask something else: do we have a choice?

Time flies and technology waits for nobody. I have not met a single CEO, from Deutsche Bank to JP Morgan, who said to me: 'ok, this will increase our productivity by a huge amount, but it's going to have social impact -- wait, let's think about it'.

The most important thing right now, what our top minds should be starting to say, is how to move mankind to a higher ground. If people don't wake up, they'll have to scramble up -- that's my 2 cents.

This is not your business as usual talk, and coming from someone in Dube's position it's not to be taken lightly. There have been analyses on CEOs and corporations and their allegiances. The only metric they are meant to maximize is profit, and one of the lessons of data-driven culture is that aiming to maximize the wrong metric will get you the wrong results.

If things such as social or environmental impact are to be taken into account, they need to be measured and incorporated in business models. Of course, this means nothing less than to rethink and redesign the entire social fabric. And that is precisely what Dube, and other top minds, are calling for now. Time flies and technology waits for nobody.


Automation will take over IT tasks, not jobs

Advisory firm says robotic programs free up tech pros to pursue more innovative work.

Money flows to robotic process automation

Survey finds IT departments are the first to be feeling its impact.

Jobs vs. AI: What happens when everything is automated?

ZDNet sits down with open source columnist Bob Reselman to discuss a question from our not-too-distant future: what happens to jobs when everything is automated?

The first 10 jobs that will be automated by AI and robots

Advances in automation and robotics are putting a lot of jobs at risk. Here are ten jobs first in line for the robot takeover

Editorial standards