Rise of AI Conference 2019 · Berlin · Unicorn Stage

Human Augmentation Through Automated Conversation Analytics

Michael Brehm · Founder & Managing Director at i2x

At Rise of AI 2019 in Berlin, Michael Brehm, founder and managing director of i2x, argues that AI should augment human workers rather than replace them, using his company's real-time conversation analytics for telesales and service agents as the example. He traces the idea to his previous company, rebate networks, where 2,500 of 10,000 employees were telesales agents he struggled to manage. Citing OECD data, he argues the common fear that automation destroys jobs is backwards: countries with more robots and technology have lower unemployment, not higher, contradicting 1990s predictions. He argues the real risk is not job loss but a widening gap between high-tech nations, which benefit from automation, and developing nations that lose the path of industrializing through basic manufacturing. Brehm describes how i2x records and analyzes every sales or service call in real time, using its own speech recognition, to give agents, team leads and managers actionable feedback during, not after, a call, arguing this keeps underperforming-but-trainable agents in their jobs rather than having them let go.

Key takeaways

  • Michael Brehm cites OECD data showing more robots and automation correlate with lower unemployment across countries, not higher.
  • He argues the bigger economic risk of AI is a widening gap between high-tech nations and developing nations that lose the path to industrialize through basic manufacturing.
  • i2x records and analyzes every sales or service call in real time using its own speech recognition, giving agents feedback during the call rather than after.
  • He says i2x's product requires buy-in from agents, supervisors and managers together, not a tool tailored to just one group.
  • He argues full call automation remains risky because deviations from expected interactions can cause brand damage that outweighs savings.

Detailed notes

Origin: a concrete business problem in telesales

Michael Brehm explains he wanted to build software that helps people communicate better, and traces the idea to his previous company, rebate networks, which grew to 10,000 employees in 30 countries before being sold in parts to Google and Amazon and taken public. Of those, 2,500 were telesales agents who, he says, "drove me crazy" by not following instructions, making sales and customer support a concrete, high-value starting point because companies close to revenue would pay for a solution.

Why "no more jobs" is backwards

Brehm argues that despite constant predictions of an AI-driven jobs crisis, the actual crisis today is reversed: there are more jobs than people to fill them. He cites OECD data showing a correlation across countries between more robots and automation and lower, not higher, unemployment, directly contradicting 1990s predictions that robots would cause mass unemployment. He challenges two assumptions behind the fear: that the total amount of work is fixed (he argues sectors like hospitals and nursing homes have far more work than capacity), and that human intelligence and capability will stay at their current level rather than continuing to evolve through augmentation, bionics or gene editing.

Winners and losers: high-tech nations versus developing nations

Brehm predicts more AI leads to higher productivity and more jobs overall, but argues the real winners will be high-tech nations while the losers will be developing nations that have not yet reached the current level of industrialization, because automating basic production such as clothing manufacturing removes the path by which those countries historically industrialized, widening the gap between developed and developing nations.

How i2x analyzes calls in real time

Brehm contrasts the old model of customer communication, where companies could randomly monitor and give feedback on only one to three percent of calls, with i2x's approach of recording and analyzing every call using its own automatic speech recognition system, breaking calls into data points and feeding real-time coaching back to the agent during the call, when it can still be acted on. He stresses that value requires buy-in from three groups together: agents, team leads/supervisors, and managers, since supervisors are freed from listening to bad calls all day and managers get real-time KPIs for fact-based management. He cites a claim that over 40 percent of service and sales staff who quit say they did so because they received too little training.

Q&A highlights

Asked about handling dialects such as Austrian German, Brehm says it is an economic decision: most companies prioritize building for English or standard German rather than smaller dialect or language markets, so smaller-language requests currently have to be turned down. Asked about differentiation against better-capitalized competitors like Google, he argues focused startups tend to beat one-size-fits-all incumbent solutions because large companies spread limited attention across many projects, and that i2x's moat comes from collecting highly specialized data in narrow use cases over time. Asked whether full call automation is the logical next step, Brehm says it is possible but risky, since even small deviations from expected interactions can trigger brand damage that outweighs automation savings, and argues service is shifting from a cost center to a differentiator, making well-supported human agents more valuable, not less.

Quotes

The true winners are the high tech nations and the really big losers I think will be the third world countries.

Michael Brehm

More AI, higher productivity, more jobs.

Michael Brehm

We help people to keep their jobs and actually not to replace them because we augment.

Michael Brehm

Questions answered in this talk

Does more automation and robots lead to more unemployment?

According to Michael Brehm at Rise of AI 2019, OECD data shows the opposite correlation across countries: more robots and technology adoption correlate with lower unemployment, not higher.

How does i2x help call center agents improve in real time?

Michael Brehm explains at Rise of AI 2019 that i2x records and analyzes every call with its own speech recognition system and sends actionable coaching feedback to agents during the call, rather than only reviewing a small sample of calls after the fact.

Who benefits most and least from AI-driven automation, according to Michael Brehm?

At Rise of AI 2019, Brehm argues high-tech nations benefit most, while developing nations lose out because automating basic manufacturing removes the industrialization path those countries have historically relied on.

Mentioned

human augmentation, conversation analytics, automatic speech recognition, real-time coaching, job crisis

Keep exploring

More from Michael Brehm

More from Rise of AI Conference 2019

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 3,757 words

Show the full transcript

I'm very happy to be here I'll speak a little bit about human augmentation speech analytics in our product but I also want to take the first couple of minutes to talk about the why why is this important and how does this fit into the bigger bigger picture maybe quick background so I wanted to do something with speech because for me it's fascinating it's one of the key elements that makes humanity if you think about what makes humans it's our brain it's our hands as tool and it's our speech now if you think and I wanted to build a software that basically helps people to communicate better with each other because how many times did you have an topic or we heard it in a company or in private yeah we had a communication issue I heard this nearly everywhere and but then it's very broad so the question was were to focus and I remember it with my last company that was running rebate networks which at a certain point of time was pretty big

we had 10,000 people in 30 countries before we sold the company to to Google to Amazon and parts we took public and out of these 10,000 people I had two and a half thousand telesales agents and they drove me crazy because they did everything but what I wanted them to do so I thought this is a very good starting point you have a very concrete business problem it's if you do it in sales and customer support it's close to like the buying the product so probably the companies that would be willing to pay for it that was a little bit the background so what are key drivers off today's world for companies were in hyper competition there are substitutes for most products transparency increases everywhere and there's a really real big question how do you actually get and stay competitive also we're not slowing down acceleration is sorry maybe I should say acceleration is increasing quite a bit I want one chart that very or one number that that I found a few that are

very impressive is our medical knowledge over the last 50 in like in 90 50 doubled every 50 years next year our medical knowledge will double every 73 days and it's not gonna slow down also if you look at inventions 99% of inventions of the world have not happened in the past 5,000 years but will happen in the next 50 years so if you think it's fast now we're actually moving very slow compared to what's happening in the next few days a few years now the new H no more jobs nearly every day in the newspapers you read there will be no more jobs and obviously AI is the big driver behind it is that true and and what can we do yes it's true we have a job crisis but you know what it's a reversed one currently there are more there are there are too many jobs and too few people that can fill them and if you look back in the discussion in the 90s it's really interesting newspapers were full robots will

take away the jobs no more jobs will be all unemployed reality is what happened yes we have lots of robots and and there's very nice data from oacd which is harmonized about around the countries and you see I just picked a few ones but there is a very strong correlation the more robots the more optimization the more technology you have the lower your unemployment and vice versa totally contradiction contradiction to what happened what was predicted 20 years ago there were millions of hours spent discussion why we're all gonna run out of jobs in the 90s it was all not not true in the public discussion today there are two main I think assumptions and questions we a we say the work we do today is limited we just assume there's a certain amount of work but then what would happen and I've heard this question a few times what would happen if we autumn eyes everything um that's done today would we not have any more work I think there would be plenty of work

and actually if you think about like what we could do with with the spare time with the creativity if you think about hospitals nursing-homes there is just there's so much work to be done that would be really great the next one is and that's a little bit a scary thought I think we assume that natural evolution will stop which means humans will stay at their current intelligence and capability level so there was an evolution where the humans we assume is now the ultimate game that's not gonna change why should that be the case why should evolution not continue and with all the technology why shouldn't it accelerate dramatically from from augmentation from improvement from bionics to evolution from gene-editing everything that we hear why should we stay at our current level so I think the the right questions to arts are the better questions is short term yes there will be some unemployment buckets in certain sectors and locations and and we need to think about what can we do about that also in the

past if you had a car factory and then you applied some robots yes there was for a certain amount of time there was more unemployment before the and the in the mid term is will machines and humans eventually merge to a level where a distinction is no more possible and you can discuss is this 20 years 50 years a hundred years but so far I had a hard time if I just rationally think what's happening in all the different eras in a medical world in the software world not see how we're more and more merging these two together and then maybe we're creating a new species and and that's like the thoughts I think are much probably much more relevant and realistic than than the current discussion winners and losers in the end personally I think it's very very simple more AI higher productivity more jobs and the true winners and this I think we have a very hypocritical hypercritical discussion here in Germany and in other countries the true winners are the high tech

nations and the really big losers I think will be the third world countries would have have not reached an industrialization level that we have it currently and and the reason is because if you autumn eyes a lot of basic production like t-shirts etc you will not have you will have a very hard time like creating this these for example clothing manufacturing shops were then the company step by step can work itself up because that will be gone and then you increase the gap between the developed nations and the developing nations so even what do you do with it and that's one of the reasons why why I also founded this company so there's a lot of discussion about both worlds human or machine and and I think it's really both together so it's the human machine hybrid and the great thing is basically AI brings job promotions to everybody who applies it without having to work for it and and that's the the idea behind our our company so how has like customer communication on

the phone looked in the past you had lots of people sometimes you have tens of thousands of people who are on the phone speaking with your customers and and then you randomly monitor them you listen in or after they have been recorded you listen to them and then try to give feedback very limited amount maybe one to three percent of the calls you can give feedback on it's it's very actually also hard work to do this now what we do is we record every call so every single call is recorded is analyzed so we can then distract in from it extract information and give personalized coaching and best practices how do we do this we integrate into the telco systems we then take the speech data and the speech we bring it we send it to the cloud we have developed our own automatic speech recognition system so to translate speech into text and into other elements we then kind of break these calls up into millions of data points analyze it and then send

the information back to the in this case the call Center a service agent and give them real-time feedback so that was always very important for me the question provide actionable information during a call not after because then you can't change it anything but during a call where you can still make some changes and what we have with we try to focus really on three key elements I think to view everything together is very important so the agent themselves but then also that supervises the team leads and the managers and only if you have the buying of all three of them you will get good results and it's also something we've learned it's great if you've built some like very specific yeah but we said we want one that really connects the whole in a way flow of information and line of command because if I dress try to tailor to the manager if address try to tailor to the agent nothing is really going to change what's great about it is the the human agents

now get real-time coaching and training by the way over 40% of age of people who work in service or sales say they quit of the people who quit 40% they say they quit because they received too little training they say they feel they would love to do a good job but they can't do because the companies doesn't enable them it's what one of the core things it's not just for for managers it's not just for for like business people it's in every area and very much there sometimes it's crazy if you think about it you have like people hired day to they are on the phone you pay hundreds of euros per lead and they're on the phone they don't know what to say how to pitch your company how to sell your product and then it's no interest 200 euros out the window no interest five minutes later 200 euros bam-bam-bam like you you burn money like like crazy by not giving them the right tools and that's a really huge problem so so

that's what we tried but then also for the supervisors if your job is to listen to bad phone call conversation the whole day that's not a very nice job it's actually nearly sometimes like punishment so now what we do is we take this really hard piece out we provide them the data so that then they can look at the data they have this and then spend their time on coaching and not listening to too bad calls and then the third one on the managers they get in real time all the KPI so for the first time they really can do a fact-based Management react fast also see how how it's working this this all obviously it took took some time to develop very happy about our investors that supported us one of them hot spring to enable us to really invest into technology majority of the people are engineers at our companies we have to build multiple lever levels of technology so first from the integration side which is pretty complex if you think about

how many telco systems that are out there then from the whole AI where obviously we have big strong one in speech but then if you have the speech itself and and the bodies it doesn't help you you also need a big bi data a Department machine learning that you put on top of it to try to understand what's happening there then the whole back and front end you need to be making it like the UI uux in a way that actually people appreciate it you're interfering in their daily work they already have tons of software systems so we're putting another there so quite a complex product here a demo sorry I think I just saw unfortunately the text and overlay is in German but the background is in English I think can you make it louder Samantha to bring and wicked I 2x and annoy loose on first pass analyzer on coaching the tycoon vagina the itraq software allow miss analucia our to name to transcribe yin wound in a site so an elysian daddy

can I to extend again I'm telephone by rights variant ask each place automatic tips game dirty automata anonymous the address translation intelligence Caravan and transcription currently dad neva often a confident man though with the scans I'm a Google Trends to a Canon order a question mark button to zoom in tighter span would like division Cayman in inclusion induction to a Canon Duty I track software common alagesh premium telephone - loyal would vehicle target ribbon in shafts and child went - aunty - pots the other information will versatile and I figured out sang-man name is that actually did the understood Cena's management Odin again in Dhahran best practices schneller - a Kenan convert Ian's Gardens which dragon wouldn't continue and an hour skits Iseman team service quickly here - MA the society X you're losing for a touchback and a loser and coaching yeah so so that's a little bit and outlook what we're currently doing obviously we're working on on a lot more and what what my topic was to come back to the before

to augmentation I'm a big fan autumn eyes as much as possible because I really believe there will be always enough work reality is if you look at the companies the problems they have and where technology is today you still are very very far away in most of the areas with customer interaction that you can automate and it's great if you say hey I have a speech computer or something and in 50% of the cases it's working great or 60% but then actually in 40% of the cases the customer is very unhappy you get very bad results it took you 15 minutes to talk with them you repeat stuff over and over again and then just to figure to not not get the results so the best company for me it's like to try like take the resources they have and really make them learn faster have better interactions and that way increase productivity and then again the coach the agent the manager and so what what we do we we basically make a data asset

that couldn't be used and analyzed suddenly unallowable and actionable we can derive a lot of best practices from it and then that results in automatic and personalist training on the job and it I just hear it and what's really great is last week I was speaking with the manager of a large call center and he said like in the past I had big like people who normally I would have led to go and then there are the ones who don't want to do this and who are bad performers you have to let them go but then there are the ones who can't do it yet because they haven't the knowledge and these ones you can really help so we help people to keep their jobs and actually not to replace them because we we augment that's it thank you very much thank thank you Michael opening for questioning we have a microphone we have to wait sorry he's low so we also tested a similar use case so speech-to-text analyzed the text and and to

some ijen coaching but you know due to Austrian dialects this use case failed dramatically so how can you overcome this hurdle of dialects and and accent and so on so the question is how strong the dialect is and indeed if you look from from a speech analytics perspective Austrian then it's quite different to regular hope dodge or do you say hi German and in the end it's you need to train them I mean you you can build the speech analytics for it unfortunately in this case it's an economic decision and then the question is do I rather push my English system or my German system or do I develop something in like let's say Austrian Deutsch or in switzer ditch and so far I think most companies I know I have always then taken the other approach so that's why it's for a lot of it's actually I I have I have also we sometimes receive calls from smaller countries smaller language and say what do we have to do can you provide us the

solution I have to tell them then unfortunately look maybe in a few years time but I don't know I can't develop a tailored I don't know Romanian Swedish whatever version it's a small market a quick question the demo you short showed us a full UI for the agent and I was just wondering do you also have an integration for example for the regular standard CRM such as sales for Service Cloud or so on yes so that was actually a demo version to show the normal regular product looks very different so you just have actually in order you have your regular CRM and then there's just a very tiny box that sits in the corner that then gives information to the agent that was just a little bit to demo and to show you what the different parties can do this is not how our customer product looks like might do questions like one on on one side how do you what's your long-term vision especially like in competition with so it's like Google duplex and

what's unique like technology like do we build on top of some other technology or what what makes it unique and like where's the mode kind of against other solutions that are better capitalized for example that's that's a very good question so so first of all if you look into the different use cases I mean what Google does is we try tries to build a one-size-fits-all solution and we are very focused on certain use cases certain companies and try to build the best solution in the world there and if you look in history how many times have large companies incumbents no matter in which area dominated a new sector versus small startups that are super focused I personally would say the bet is more on the startup side then it's on the incumbent side and personally I'm a lot less scared or afraid of the large companies I have a lot more if you want so I would say I would say not not panic or but always you have to be attuned and and be

really careful about other startups that's where I think the biggest danger is from and that that's like always repeating itself because all these large companies they have so many priorities yes theoretically they have lots of people they have lots of money but then a lot of these very special projects they just put two three five maybe ten people on it they start at nine or ten at best they have a long break they go home at five on the weekends they also don't think about this stuff so so that's and then what you can build about it my vision is really to build this human augmentation level level like really like a piece of software that upgrades you as a human and helps you to better communicate and at whatever you do do a better job and we start very focused in certain areas I think we can take one more question and then we continue well there the question is you you develop better your over time you collect very specialized data we just

select very very unique data very unique use cases where nobody else has access to and we build software and special algorithms around it and the longer we do this the harder it will be for others to kind of copy this in this in these fields one more question and I think Michael will be around later yeah thank you for it a good talk and isn't it the next step or is it logically the next step to automate the call itself because there's like bad versions of this in the US and so on and wouldn't it be natural to just not have a human in the call anymore but learn how to do it automatically basically yes you can autumn eyes more and more absolutely the question is how long will this take and if you look at a lot of companies or a lot of use cases even something that for humans is perceived standard for machines it's still not and there's a slight deviation and suddenly everything breaks and then nowadays in like a

customers are very valuable you don't want to lose them in the age of social media if you do this too much you suddenly have a huge show around just and you saved a little bit on automation part but the brand damage is enormous and because of that competition one of the key areas where companies still can differentiate is actually service so service has been moved more and more from a cost center to a profit Center and and personally I see more the actually opposite direction yes oppisite part but then they're thinking like how can we deliver an even better experience how actually can we put explicitly human agents so that they deliver the best possible customer experience and a real wow effect to - once they call so I think I still in reality it's still quite a bit out until you can automate everything cool thank you very much Michael much

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.