Pharmacotherapy is likely to remain a mainstream treatment for many mental disorders. A great deal has been learned about psychotropic medications in the past 70 years, and treatment efficacy has improved significantly. However, pharmacotherapy is generally limited to symptomatic relief and cannot provide a cure. In addition, only a certain proportion of patients are able to achieve remission and/or recovery, and the complete disappearance of symptoms remains a distant goal. The accurate allocation of patients to the most appropriate treatment option based on a deeper understanding of pathophysiology is now needed, along with the development of drugs with novel mechanisms of action. In other words, we need to realize “precision medicine” within psychiatry. To this end, conducting better clinical trials by solving current problems, thereby enabling the faster delivery of new drugs to patients, is important. The extensive review by Correll et al1 provides very broad and detailed information regarding the above-mentioned issues and carefully explains what is needed to move forward. As they mention, the lack of sample representativeness in clinical trials, the strong (and increasing yearly) placebo response, the high dropout rate, and the varying reliability of severity assessments are of particular concern. Digital phenotypes derived from personal digital devices2 seem to have ample potential to address these problems. This potential could be further enhanced by successfully combining new ways of delivering health care using communication technologies such as telemedicine. Clinical trials often require patients to travel long distances and to make frequent hospital visits, which may reduce the likelihood of trial success. Promoting decentralized clinical trials, i.e., systems that allow patients to participate in a trial without necessarily coming to the hospital, would facilitate patient recruitment and prevent dropouts. The use of digital data to quantify the severity of symptoms in an objective manner could also reduce variations in assessments made at different clinical sites. Frequent assessments are a major burden on patients, but by utilizing ecological momentary assessment via passive monitoring, a method that is becoming increasing feasible3, therapeutic benefits that were previously difficult to detect might become identifiable. Given the potential of such digital technologies, it seems likely that many currently unmet needs will be addressed. However, the story is not that simple, and this is not a task that can be completed overnight. A potentially important question in the use of these digital tools is whether they can assess a patient at a level similar to that of a skilled evaluator meeting the patient in person and taking the time to assess his/her psychopathology. There are many different types of digital phenotypes, ranging from those in which the patient actively provides input on his/her condition (called active data) to passive data, such as sensor data, that do not require the patient's active involvement. The latter provide a wide range of information, including data that can be collected from a smartphone such as geographic range of activity, call logs, text input and search logs, as well as data that can be collected from a wearable device, such as acceleration which can be translated into activity, sleep rhythm, heart rate (or pulse rate), and skin conductance. Furthermore, passive data can be obtained through smart speakers, cameras, or some other devices, for example patient language as quantified by natural language processing, speech rate, acoustics of speech, facial expression, posture and body movement. Even if these data could objectively quantify a patient's behaviour and/or autonomic nervous system activity, they would not elicit the patient's thoughts or moods and could only serve as surrogate markers. Many studies have reported that it was possible to distinguish between patients with mental disorders and healthy volunteers4, or detect early sign of relapse5 with a relatively high degree of accuracy from these data, but there is still large room for improvement. Even when a pathological feature can be identified, it is often unclear whether it is a state or a trait marker6. Many of these predictive models utilize machine learning, but it should be noted that, although this technique may fit the specific population from which the data were obtained, the generalizability of findings may not always be high. In addition, determining how to accommodate differences across patients' lifestyles is especially important: the identification of digital phenotypes common to patients across cultures might be difficult. Nonetheless, the advancement of the above technologies and the accumulation of the relevant knowledge may benefit not only clinical trials but also real-world clinical practice. Gold-standard evaluations may be difficult to perform in time-constrained clinical settings, but “measurement-based psychiatry” could be delivered more easily with those technologies. In fact, commercially available wearable devices can already quantify sleep and activity, and some practitioners may be using such data to treat patients. Specifically, the accumulation of longitudinal data on individual patients would be useful for identifying changes over time. A large cohort study that collects digital data would allow to identify which patients with which digital phenotypes respond to which treatments. As a result, the selection of drugs with the greatest likelihood of being effective for individual patients might become possible. Concerns about the use of digital tools in clinical practice should also be considered. The question is what kind of long-term changes might occur as face-to-face treatment is replaced by the use of information and communication technology and digital tools. One often discussed issue is the digital divide, i.e., the risk that those who are unable to successfully use digital tools will be left out of health care7. Since the COVID-19 pandemic, psychiatric care has been delivered almost entirely remotely in some countries, but it is necessary to investigate whether this has the same therapeutic effect as face-to-face care. A large body of evidence already shows that telemedicine is no less effective than face-to-face care, but it remains unresolved whether this is true even for long-term treatment over multiple years8. Furthermore, there is a chance that the focus will shift to improving digital device-derived outcomes rather than actual patient recovery, if treatment effects are assessed using digital phenotypes rather than humans. As we accumulate digital phenotypic data in the future, it will be important to study how these data are connected to pathophysiology. For example, studies that explore the relationship between brain functional connectivity and digital phenotypes would be useful. If a treatment has been identified that is effective for a specific pattern of functional connectivity, digital phenotyping may be able to identify the patients who are the best candidates for that treatment. Even if the above-mentioned hurdles are overcome and a regulatory-accepted digital methodology is developed, there is no guarantee that such a methodology would be the best way to quantify mental disorder symptoms over the long term. Sensing technology and analytical methods are constantly evolving, and they can quickly become obsolete. The continued use of once-established standards for many years might nullify the advantages of digital technologies9. In conclusion, a great potential seems to have emerged from the use of digital technologies to foster the progress of psychopharmacology. Interdisciplinary research and development with the goal of actually improving the outcomes of people with mental disorders are now needed.