If customers of FinTech companies are expected to rely at least in part if not in the whole on the algorithms of the technology of themselves, what do we know about how important trust is? There's a substantial amount of work on trust in financial markets and in the financial advisory space and that would include in the insurance space, in the real estate space, and some of the related spaces we talked about as part of these lectures that are informative about what we might be able to conclude in the case of FinTech. Of course in FinTech, we have a number of other considerations including the quality of algorithms and what the human machine interface might look like. Let us look next at some of the evidence that has accumulated by way of research and anecdote. Behavioralist and psychologists tell us that when human see each other for the first time or perhaps see each other after not having seen one another for a period of time, we quickly assess attributes of safety and competence very rapidly and powerfully. The first two principal components if you will of human interaction. We can often think of levels of trust construed along these dimensions. For example, trust in technical competence. So now, how do I trust that you know what you're doing? Second, especially important in financial transactions and financial service and relationships, trust and ethical conduct and character. Do I trust you not to steal money from me? Important of course in an obvious way. Third, trust in the ability to be human, trust and empathetic skills, maturity, emotional intelligence. If I tell you personal things about myself or my family, can I trust that you, an advisor, will handle that well and can provide counsel? Many who are concerned about FinTech crowding out the traditional methodologies and traditional relationships in the advisory space and elsewhere, often highlight the last component under the idea that three million years of human progress may not be ultimately resulting in an app by itself. However, the open question remains of, what are the trade-offs between or the interactions between trust in advisory services, the kind of services that FinTech providers are actually potentially providing, and the clients need to feel safety. What builds trust? Well traditionally, we hear taking time to explain products and offerings and recommendations; explaining fees, staying up to date with current products and trends, responding quickly to client needs. Although, our research suggests the financial services industry over emphasizes required frequency of communication. We studied this at Wharton partnership with State Street Global Advisors a couple years ago. We did a survey of thousands of financial advisors and their clients. The kind of clients who might be purchasers of FinTech Services at some level, at least from a consumer perspective. We asked what the most important characteristic of a financial advisor is. Far and away, both advisors and their clients which who were matched, indicated that trustworthiness of the advisor was the most important element. Trustworthiness was at the margin seen as more important by advisors then clients but over all the possibilities for both of them, trustworthiness was at the top. Second understanding the family's needs and goals. Again, viewed slightly more important by advisors than by clients. Quick efficient Folsom communication seem more important at the margin by clients and advisors and then performance. Interestingly enough, financial advisors viewed performance as less important than clients. Finally, cost fees. Advisors felt that fees were not as important as trustworthiness which was inherent in the client data as well, but they thought it was less important than clients saw among other possible answers as well. So far and away, the trustworthiness of an advisor appears to matter at the margin more for advisors than clients but for both incredibly important. An additional part of the analysis looked at the most important attributes of the value proposition of financial advisors ranging from knowledge, personal factors including relationship factors, communication styles including proactiveness, stability of the relationship, transparency and responsiveness for clients being knowledgeable was seen as most important. Clearly important for the FinTech industry is the ability to reflect and to carry through knowledgeable applications of their art in their craft. Responsiveness was seen as the third most important thing for advisors followed by personal factors including pedigree, and history and tenure in the profession not seen as very important by clients. Top for clients was knowledge, second was responsiveness, third was transparency. Knowledge responsiveness and transparency are things that the FinTech industry can lead with. On the other hand, financial advisors thought that personal factors, knowledge and responsiveness were most important. Once again areas in which FinTech may have an advantage. What do millennials think of it? Well, in research specifically focusing on the millennials, when they were asked what the most important characteristics of advisors were. Well, first the dependent upon complexity of the scenario or the questions involved, and second was ordered in a manner that was apparent from previous data and previous comments that we've made. The millennials segment also has specific attitudes toward this idea of knowledge, transparency, and communication. In some research published in the Journal of Financial Planning, Johnson and Larson reported results from a survey of millennials on the attributes of financial advisors that were most important and they dichotomized under two scenarios. One, where there is a simple decision to be made and a second where the information was complex and the decision was complex. I've highlighted the statistically significant attributes and the table is sorted in descending order of importance attributes that were rated most important. By way of statistical significance, it was knowledge and friendliness that dominated and of course to a lesser extent the degree advanced training, and so on were important. In other words, one instrumental and one effective characteristic of advisors. Note that in a high complexity scenario, friendliness and approachability which we equate roughly with safety was more important than technical capacity. For a low complexity financial planning scenario, again at more than the margin, latest financial techniques and models were rated most highly. This gives insight into some of the challenges of FinTech when it's distinct from the human element highlighting the importance of the bionic model in FinTech. To summarize it in one more way, when high complexity scenarios were at stake, the human element dominated when low complexity scenarios were at stake. The technical element dominated although they were both important across the board. In addition, Trust and Constant Contact has been born out as being highly important across scenarios for clients of financial advisory firms. Shmeiser and Hogarth, a couple of years ago found that financial advice increases financial well-being broadly construed by way of perception and also has an influential effect on financial behaviors in a positive direction. So financial advice was important. That seems like a predicate in the FinTech world. Second, trust was a function of age and with contact in particular. These results which were developed in an ongoing way as recovery from the financial crisis was happening found that especially among retired people trust levels were low. However, trust levels were especially low when clients had not been contacted by financial advisors recently either because they hadn't been contacted proactively or they hadn't sought out advice. The results suggest here that a proactive element of FinTech advisory just like it is for human advisory will be important. Now, the FinTech platforms that we discussed and you've heard about or will hear about in other lectures or in your travels and travails, are not purely human in orientation by definition you might say. Although some might have bionic approaches, they don't necessarily distinguish themselves by being not tech. They're by definition tech. So the question then arises what do we know about trust not in humans as you just heard with respect to financial advisors but specifically with respect to algorithms? In a fascinating bit of research published in the Journal of Experimental Psychology and elsewhere, Dietvorst, Simmons and Massey in a pen working paper discovered that people displayed algorithm aversion and the article is titled Algorithm Aversion. People erroneously avoid algorithms after seeing them. In this groundbreaking research, they found that people prefer humans over algorithms generally. That may sound strange in a FinTech lecture but it sometimes needs to be confronted for anyone who is interested in the space, who is investing in the space, who is working in the space and so on. They also found that trust in algorithms is lower when humans view them at work. In other words, before they see algorithms work they trust them more than afterward even if they're better than competing human technologies. Finally, they found that when an algorithm was perceived to make a mistake it was especially bad. So in other words, people like people over algorithms. When people watch an algorithm work they trust it less even if the algorithm is producing better results than humans and then when an algorithm makes mistakes it's especially bad. In their study, the authors focused on predictions of MBA application success and on forecast a number of airline passengers in the airline industry. In each case, they asked participants in their study to make forecasts either of application success or of a number of airline passengers and compared that to quantitative models that they manipulated. In their studies of MBA students performance, they allowed participants to witness results of what they called humans forecasts and then forecasts of student performance that they described to participants as coming from models. When the forecasts were described as coming from humans, they saw no difference in the selection of participants to use humans forecasts. You can see that in study one because there's no palpable difference between a control random oriented choice selection and those that saw results of the humans forecasts. However, when participants witnessed results of forecasts for models whether the models were better or not but even when the models were better their selection of them aka their trust of them declined. You can see the numbers in the control went from 65 percent and 63 percent based on the first treatment of those participants choosing the next round to be a statistical model compared to what happened after they actually saw the models forecast or the models and humans forecasts. Same thing is repeated in study number four which incorporated their own forecasts as well as another human and perhaps because this might have been a cooked MBA student example, they looked at forecasting the number of airline passengers in the United States a kind of exterior problem the same exact results. When asked to select a statistical model or a human model when they saw the results of a human's forecasts, there were just over 50 percent likely to choose the statistical model. However, when they saw a statistical model they were more likely to choose the human forecast 56 percent with a 41 percent for choosing a statistical model and that's also true when they saw both humans and statistical model algorithmic forecasts when the model outperform humans. It's an incredibly powerful result across the board. More than that, your belief in advance was a function of whether you saw a model or a statistical algorithm perform even if again was better. So your confidence in advance made a difference. If you look at the far right of the graph in front of you when participants described themselves as being much more confident in a model in advance, they were 82 percent likely if you will to choose a model and that converged across whether you saw the model perform or whether you didn't see it. However, if you were not confident in advance, and you saw the model perform you are significantly less likely to choose it, 13 percent compared to 39 percent were more confident. In the other dimension across confidence, again, if you were more confident of the model you're more likely to choose it. So how could we bring this together for FinTech? Well, number one we see how important trust is and the variance of trust at least in part. Second, in the FinTech context we might translate this notion of trust into trust of algorithms. What do we know there? Algorithmic aversion exists. People don't choose algorithms over humans once they see the algorithm in action even if the algorithm beats people. They may choose the algorithm if they haven't seen any results even if those results are better than theirs. If algorithms make mistakes and people see them they strongly avoid the use of the algorithm. Interestingly enough the authors of the piece we just mentioned followed on and found that if they could change the algorithms let's say participants if people can change the algorithm even a small bit after seeing its performance they're more likely to use the algorithm. So it looks like in some sense combining judgment with the algorithm helped them feel confident and trusting of the algorithm itself. This reflects some pretty interesting work summarized in a well-known paper in science in the late 80's by Dawes, Faust, and Meehl called Clinical Versus Actuarial Judgment in which in a meta-analytic approach the authors described how actuarial approaches called them algorithmic approaches in many different contexts beat clinical or human-oriented approaches including judging people's personalities, making parole violation predictions, diagnosing progressive brain dysfunction, understanding and diagnosing Hodgkin's disease, forecasting college grades, diagnosing psychological disorders, understanding hospitalization times and numerous others. The actuarial approach dominated the clinical approach although they did find that the combination of actuarial and clinical approaches dominated all the others. There were several reasons: consistency or its converse for humans inconsistency across time, across scenarios, across states of mind of the protagonist. Actuarial approaches tended to be more rational and rules-based. They allowed formal rational weighting of inputs. They included all kinds of relevant variables and excluded irrelevant variables. Sample biases of clinical approaches are often large in number, limited samples, skewed exposure to treated cases. Doctors for example see only the sick as opposed to a controlled sample and so on. One area where it was found that the combination of actuary clinical approaches dominated others was in the area of what we call the broken leg problem or the recognition of extreme outliers in which clinicians were shown to have better predictive capabilities. But the idea that the combination of humans and machine might give the best result is highly supportive of the FinTech proposition in some ways especially as it relates to trust in algorithms and ultimate outcome for clients.
