Los algoritmos como remedios recetados

Debemos tratar los algoritmos como medicamentos recetados

Por Andy Coravos, Irene Chen, Ankit Gordhandas y Ariel Dora Stern
February 14, 2019


Una cosa es cuando las compañías usan algoritmos para personalizar los anuncios de zapatos o etiquetarte en una foto de Facebook, y otra muy distinta es que los algoritmos decidan si te liberan bajo fianza o te envían a prisión. Si bien muchos han expresado su preocupación sobre el uso de algoritmos para informar decisiones tan importantes, pocos han discutido cómo podemos determinar si realmente se pueden usar de manera segura y efectiva.

Como profesionales de la tecnología de la salud, creemos que es útil ver estos algoritmos de alto riesgo como medicamentos recetados o dispositivos médicos. Durante décadas, las compañías farmacéuticas y de biotecnología han probado medicamentos mediante ensayos clínicos meticulosamente ajustados. ¿Por qué no tomar algunas de esas mejores prácticas y usarlas para crear algoritmos que sean más seguros, más efectivos e incluso más éticos?

Los algoritmos son como las drogas.

Tanto los algoritmos como las drogas pueden tener un enorme impacto en las vidas humanas, el truco está en equilibrar los beneficios y los riesgos. Los medicamentos de quimioterapia, por ejemplo, pueden reducir el tumor de un paciente, pero también pueden causar efectos secundarios angustiosos. Los algoritmos son muy parecidos. Un algoritmo utilizado en los servicios de protección infantil para predecir el peligro puede salvar a alguien de la violencia, pero un examen equivocado puede ser innecesariamente intrusivo para las familias.

Pero a diferencia de los medicamentos y procedimientos médicos riesgosos, estamos interactuando con algoritmos sin leer las etiquetas de advertencia, porque no hay ninguna. Un oncólogo probablemente nunca recomendaría medicamentos de quimioterapia invasiva para una persona sana; Es solo cuando se trata un cáncer agresivo que consideramos los riesgos y las recompensas. Al considerar las aplicaciones de tecnologías de alto riesgo, los arquitectos de los sistemas de inteligencia artificial podrían adoptar un enfoque similar que tenga en cuenta tanto los riesgos como los posibles beneficios.

Debido a que los algoritmos y las drogas tienen puntos en común, podemos aprender de los paradigmas regulatorios existentes y exitosos en la atención de la salud para pensar en los algoritmos. Estas son algunas de las formas en que son similares.

Afectan vidas. Al igual que los medicamentos y dispositivos, los algoritmos de alto riesgo tienen el potencial de transformar el mundo de un usuario de manera significativa. Ya se han desarrollado algoritmos para hacer recomendaciones sobre si los acusados ​​deben ser liberados bajo fianza, para determinar los beneficios de atención médica y para evaluar a los maestros.

Pueden ser utilizados como tratamiento médico. Existe un campo completamente nuevo de tratamientos dirigidos por software llamados terapias digitales (DTx), que son programas de software que previenen, controlan o tratan una enfermedad. Por ejemplo, Akili Interactive Labs ha creado un DTx que se ve y se siente como un videojuego para tratar el TDAH pediátrico. Algunos incluso han sido aprobados por la Administración de Drogas y Alimentos de EE. UU. (FDA), como reSET-O, un producto de Pear Therapeutics para pacientes con trastorno por uso de opioides.

Se desempeñan de manera diferente en diferentes poblaciones. Algunas drogas funcionan en una población pero no en otra. Por ejemplo, el adelgazador de sangre clopidogrel, o Plavix, no funciona en el 75% de los isleños del Pacífico cuyos cuerpos no producen la enzima necesaria para activar el medicamento. De manera similar, los algoritmos pueden afectar a diferentes poblaciones de manera diferente debido al sesgo algorítmico. Por ejemplo, la plataforma de reclutamiento impulsada por la IA de Amazon tenía un sesgo sistemático contra las palabras orientadas a las mujeres en los currículos.

Los investigadores han respondido a estos hallazgos mediante la elaboración de estrategias para detectar y reducir el sesgo algorítmico. Outlets como Propublica han investigado algoritmos problemáticos, mientras que los informáticos han creado conferencias académicas y centros de investigación centrados en el tema. Estas iniciativas representan pasos en la dirección correcta, al igual que el impulso creciente para ensayos clínicos más representativos puede conducir a una mejor investigación médica.

Pueden tener efectos secundarios. Así como un medicamento que ataca una condición puede tener efectos secundarios en otro sistema, los algoritmos también pueden tener efectos no deseados. En la guerra en línea por la atención, un sitio web que apunta a aumentar el compromiso puede encontrar retrospectivamente que sus algoritmos de aprendizaje automático aprenden a optimizar el contenido que induce a la ira y el miedo a fin de aumentar el tiempo dedicado al sitio. En la búsqueda para hacer que el producto sea pegajoso, el efecto secundario es un cambio de comportamiento inoportuno (indignación) en su población de usuarios.

Lo que podemos aprender del desarrollo de medicamentos para construir mejores algoritmos.

En el desarrollo de medicamentos, los fabricantes deben demostrar la seguridad y la eficacia de los productos farmacéuticos antes de que salgan al mercado. Pero mientras la FDA ha creado y hace cumplir tales protocolos, los algoritmos permanecen en gran medida sin regulación.

Algunos principios rectores ayudan a ilustrar cómo las herramientas del desarrollo de medicamentos podrían usarse para construir algoritmos más seguros y efectivos:

El manejo de "eventos adversos". Tomar medicamentos recetados conlleva el riesgo de lesiones, hospitalización y otros eventos adversos, como los llama la industria. La FDA tiene una estructura de informes públicos bien documentada para el manejo de tales contratiempos, donde los informes de eventos graves como muertes y hospitalizaciones se registran voluntariamente en una base de datos pública.

Pero, ¿qué hacemos cuando el “aporte” médico es un algoritmo? Actualmente, carecemos de herramientas de informes públicos sólidas para manejar resultados de algoritmos adversos, como la crisis de Cambridge Analytica en Facebook, pero se pueden crear bases de datos públicas para casos de uso comunes.

Saber cómo funciona el producto. Antes de que un fabricante ponga un medicamento en el mercado, debe comprender la interacción bioquímica entre el cuerpo y el medicamento. Pero este nunca ha sido el caso de los algoritmos. Debido a algo llamado "efecto de caja negra", muchos algoritmos de aprendizaje automático son difíciles o incluso imposibles de interpretar. Esto debe ser reconocido y abordado cuando sea posible. Cuando entendemos cómo las entradas se transforman en productos, es más fácil comprender los riesgos potenciales si el sistema no funciona correctamente.

Entendiendo para quién es el producto. Los investigadores clínicos deben definir claramente los usuarios objetivo de un medicamento para que un médico que prescribe pueda confiar en que el medicamento se ha probado con éxito en pacientes similares. De manera similar, los algoritmos bien diseñados deben definir las características de la población en la que están destinados a ser utilizados. Debido a que los algoritmos pueden funcionar de manera diferente en las poblaciones para las cuales no se desarrolló el algoritmo, es esencial asegurar que los algoritmos especifiquen y documenten a qué poblaciones y casos de uso se aplican. Al hacerlo, se confía en que los riesgos y las recompensas para el grupo objetivo se han estudiado lo suficiente y se consideran una compensación aceptable.

Entender cómo se desarrolló el producto. Los ensayos clínicos se basan en registros de ensayos públicos y en informes obligatorios de los patrocinadores para respaldar la transparencia. Tal sistema responsabiliza a los desarrolladores de productos por realizar estudios éticos y publicar sus resultados. Sin embargo, los algoritmos de alto riesgo a menudo no comparten sus métodos de validación públicamente; Debido a que las empresas son tan protectoras de su IP algorítmica, a menudo no está claro cómo se ha probado un producto o si los resultados son reproducibles. Pero comprender cómo se desarrollan los productos puede ayudar a clarificar y mitigar los resultados no deseados.
Informar a los usuarios de los riesgos y beneficios. El Informe Belmont, escrito en 1979, describe los principios éticos básicos que involucran a los sujetos humanos en la investigación médica, como el "consentimiento informado". ¿Pero qué tan consciente está de los experimentos sutiles que se le administran en línea? Se podría argumentar que Facebook A / B está probando nuevos elementos de diseño en su suministro de noticias para determinar qué implementar es una forma de investigación de sujetos humanos sin consentimiento.

La versión del consentimiento informado de tech world son las políticas de privacidad y los términos de servicio que acompañan a muchas aplicaciones y sitios web. Pero los usuarios rara vez los leen, lo que hace que algunos investigadores se refieran a ellos como "la mayor mentira en Internet". Cuando hay mucho en juego, es particularmente importante que estos acuerdos no solo sean legibles, sino también leídos por los usuarios. Con el objetivo de abordar la falta de transparencia de los datos, la Unión Europea creó el Reglamento General de Protección de Datos (GDPR), una de las reglas de protección de datos más estrictas del mundo, que entró en vigencia en mayo de 2018 con el objetivo de otorgarle al individuo más poder sobre sus datos. . Bajo GDPR, las compañías asumen una mayor responsabilidad por la protección de datos y una responsabilidad clara para obtener el consentimiento de los individuos de quienes recopilan información. Las personas ahora pueden solicitar información completa sobre los datos que una empresa ha almacenado en ellos.

Protección de derechos de datos y privacidad. En la atención médica, los pacientes y los médicos tienen derechos claros y reglas de gobierno para las muestras biológicas como la sangre, la orina y los genomas; los investigadores no pueden usar estas muestras para investigación fuera de los estudios y procedimientos que el paciente haya aceptado. Pero no hay equivalentes para "especímenes digitales", que a menudo contienen datos individuales altamente sensibles. Para la mayoría de los productos de tecnología, los derechos de los datos y la gobernabilidad no son claros para los usuarios, y los algoritmos de alto riesgo exigen más que un enfoque único para todos. Los derechos de los datos deben incorporarse al producto en sí y no aceptarse ciegamente a toda prisa para completar el proceso de registro.

Creación de estándares y gobernanza para algoritmos.

La calidad y los estándares éticos internacionales han sido bien adoptados en las industrias médicas durante décadas: existen las Buenas Prácticas Clínicas (GCP) para administrar los ensayos clínicos, las Buenas Prácticas de Fabricación (GMP) para los productos y las Buenas Prácticas de Laboratorio (GLP) para garantizar la consistencia y confiabilidad de laboratorios de investigacion.

¿Es hora de una buena práctica de algoritmo (GAP)? ¿Debemos establecer una FDA para algoritmos?

Existen varias barreras para hacer que estas estructuras de gobierno funcionen. Debido a la ubicuidad de la inteligencia artificial en todas las disciplinas, un organismo regulador global sería poco realista; La supervisión puede y debe adaptarse a cada campo de aplicación.

La industria de la salud, por ejemplo, ya está bien posicionada para regular los algoritmos dentro de su campo. La FDA ha comenzado a publicar contenido detallado, emitir directrices y eliminar productos basados ​​en la inteligencia artificial como la terapéutica digital. Otras industrias con organismos reguladores, como la educación y las finanzas, también podrían ser responsables de articular las mejores prácticas a través de la orientación o incluso la regulación formal. Sin embargo, muchas otras industrias no tienen agencias reguladoras con responsabilidad pública. En entornos no regulados, los consorcios de la industria y los líderes de la industria tendrán que jugar un papel importante en la articulación de las mejores prácticas.

La sociedad está buscando formas de desarrollar algoritmos seguros y efectivos en entornos de alto riesgo. Aunque quedan muchas preguntas, los conceptos y herramientas de la investigación clínica se pueden utilizar como un punto de partida que hace reflexionar.

Read more »

6 consejos para desarrollar una aplicación popular

6 Insider Tips For Developing A Popular App

ROB JONSON, VENTUREBEAT

The press loves an “app millionaire” — entrepreneurs that make big money as their products attract venture capital, or the assimilating hands and deep pockets of a technology giant like Apple or Google.

Most developers will never live in that world.

The majority will make their living as contractors or employees working on other people’s apps, but some make a living writing their own apps and releasing them in the various stores. I have been living off my own apps for a decade now. From the Treo 600, through the iPhone revolution and on to the growing popularity of Android.

At the risk of sounding like a hoary old man of the hills, I’ve seen a lot of changes and been astonished at the ways in which the market has developed. But some things have remained true throughout the 10 years I’ve been running Hobbyist Software, and I expect they’ll go on being true long after I’ve hung up my keyboard. So here are a handful of tips and observations for the solo developer.

Satisfy your own needs.

My most popular apps are ones that I’ve developed for myself to satisfy my own needs with devices that excite me. If you want a capability on your device, chances are there are other people thinking the same thing. Having a problem you want to solve for yourself means that you are more committed to it and actually understand it. It’s also a lot more fun.

Take feedback and act on it.

Without exception it has been feedback from highly engaged users which has allowed my apps to keep developing over the years, to improve and stay fresh. I respond to most customer emails myself and aim to do so quickly. This seems to have a lasting halo effect as customers recommend my apps to their friends. Many people are surprised and very pleased to get an email from the real developer rather than a support minion.

Keep it simple — but not too simple.

You’ll typically get most feedback from the highly technical users who want lots of complicated features and options. These guys are great, they have some killer ideas, but they are not the majority of your users. In order to keep the broader base happy, you need to keep things simple. Palm OS used to talk about "the zen of Palm." They obsessed about letting users act in as few taps as possible. Apple has embraced this desire for simplicity — though, with Apple, making things beautiful can sometimes get in the way of achieving the goal.

Looks matter.

When I started developing, apps were called applications, and we cared more about what they did than how they looked. Times have changed. For your app to be a success, it needs to look good. Spend that bit of extra time (and maybe money, if graphic skills aren’t your thing) to give it a bit of polish.

You can’t predict success.

Apps are like pop songs. You write the app, you polish it and you release it. You don’t know whether it will be a hit or flop. That is true even after your first successful app. Most pop bands are one-hit-wonders, and most developers will struggle to follow initial success. I had low expectations for the app that would become my most successful project, and others that I was super-excited about disappeared without a trace. You do your best, release your app, then move on if you need to.

Small is good.

I have been accused of lacking ambition, but I like my small, low-risk approach. I don’t have employees, I have never spent more than a few thousand pounds to develop, design, and launch an app. Many developers are working towards a big launch on borrowed money, hiring an expensive team of rock star developers and publicists, hoping and hanging on for that ever elusive venture capital or big tech buy-out. I look at most of those app ideas and wonder why they didn’t just build their app in the evenings, launch it, and see what happens. Most will disappear without a trace, but a good idea that fulfills a need will gradually find a market. And probably has as much chance of hitting it big as any other decent app, with a lot less risk.

Venture Beat

Read more »

¿Qué es ser cool en Silicon Valley?

Losing Our Cool

Hey Silicon Valley: Stop co-opting cool.

By Seth Stevenson

Your tablet computer is not cool.

Your rideshare app is not cool.

Your TED Talk on mindful tweeting is—I hope you’ll agree—pretty clearly not cool.

A million dollars isn’t cool. You know what’s cool? Nope, not that.

I am here on an etymological mission. I’ve come to reclaim cool from Silicon Valley. Someone must rescue this once-useful word from the hordes of techies who misuse it.

Long ago, in a time before geeks ruled the earth, the word cool had meaning. It meant, roughly: not giving a fuck. Or, on occasion: not giving two fucks.

Look no further than the original, thermal sense of the word. Coolness implies less-active molecules, a steady state, a chilly reserve. Heat is all about engagement and transformation, overbubbling pots, hopping and colliding.

This distinction has, sadly, been lost. In the tech world, it’s now cool to gush with enthusiasm. It’s cool to be engaged and accessible and to post needy social media messages. It’s cool to get up on a stage and claim that your new device is going to change the world. It’s cool to be in the audience, watching someone on stage claim that his new device is going to change the world.

Don’t get me wrong. I think geek culture is terrific in many ways. I like when people are engaged, enthusiastic, and accessible. Those are admirable qualities. Generally beneficial to society. They’re just not cool.

Coolness, I would argue, is all about disengagement and indifference. Consider: What was the quintessential 20th-century signifier of cool? Correct: a pair of dark sunglasses. Preferably sunglasses worn by Cary Grant but, really, any sunglasses would do.

Why not some other type of eyewear, like maybe Rec Specs? Or another sunblocking accessory—like, say, sunscreen, or a parasol? (I often wish we could live in a world where parasols are the coolest possible accessory. We don’t live in that world.) Why did we collectively settle on a pair of dark glasses as the material embodiment of coolness?

Because sunglasses create a barrier. An unknowability. The sunglass wearer—her eyes obscured—is signaling a distance. A disengagement from her surroundings. A level of icy remove. Everybody knows icy remove is cool.

What else signaled coolness, back when things were still actually cool? (I mean, in those blissful days before we let people convince us that Google Glass and 4Chan and Kevin Smith are cool?) That’s right: a cigarette. Lit. Dangling from a pair of unperturbed lips. Why? Because the cigarette signals supreme indifference. Indifference to the Surgeon General’s warnings, to societal opprobrium, to death itself. So freaking cool!

When marketing company “coolhunters” go a-hunting for cool, are they in search of the kind of people who engage with popular culture and rave about new mass-market products? Of course not. They look for the truly cool kids with their own internal cool compasses—people wholly indifferent to what everyone else is doing.

To be clear, cool is not about hating on stuff. The designer Bruce Mau has lobbied against coolness on these grounds. “Don’t be cool,” he pleads. “Cool is conservative fear dressed in black. Free yourself from limits of this sort.” I hear ya, Bruce Mau. Some people are scared to embrace stuff, worried that they might be uncool if they embrace the wrong stuff. So they stick with safe choices, like all-black outfits. It’s an annoying, defensive pose. A posture that geek culture, to its credit, has fought nobly to destroy.

But this is really just another misunderstanding of cool. Genuine cool does not flow from a place of caution, a fear of endorsing the wrong thing. Cool is always—and fundamentally—indifferent.

To wit: Who’s cool? Prince, duh. Now ask yourself, does Prince dress all in black? No, duh. Prince wears purple velvet leisure suits and fussy ruffled collars. Why? Because Prince doesn’t give like 12 fucks. Indifference emboldens eccentricity. Eccentricity is cool.

Yes, I know, language is fluid, words evolve, meanings shift and meld. Just ask famed lexicographer Paris Hilton. Hilton literally trademarked the catch phrase, “That’s hot.” Yet she uttered it with such obvious disengagement and indifference that, for a brief cultural moment, it was sort of cool. Fluidity in action. Polar reversal. Not bad (meaning bad) but bad (meaning good). Are you up for it? I’m down.

I acknowledge this is a purely semantic and almost certainly futile battle on my part. Please note: I’m not saying I’m cool. And I’m not saying it’s better to be cool than geeky. I might even be guilty, on occasion, of misusing cool in precisely the manner I’ve railed against here. Once in a while, I, too, geek out. (“Coooooool,” I murmured as I swiped a finger across the screen of my first iPhone. And then I felt a hot wave of shame.)

I’m just trying to raise awareness here. We ought to restore the integrity of the term.Cool had a singular and powerful connotation, once. Now we’ve diluted it beyond recognition. It’s become just another all-purpose accolade. Cool’s not cool anymore.

Slate

Read more »

Ser ingeniero en software implica salarios cada vez más altos

Los ingenieros en software están ganando salarios de atletas profesionales

SARAH MCBRIDE, REUTERS

SAN FRANCISCO ( Reuters ) - Entre los ejecutivos mejor pagados de Twitter Inc, el nombre de Christopher Fry se destaca.
El vicepresidente senior de ingeniería recaudó $ 10.3 millones el año pasado, justo detrás de Twitter Dick Costolo el presidente ejecutivo con 11.500.000 dólares, de acuerdo a los documentos de salida a bolsa de Twitter. Eso es más que los sueldos de los ejecutivos como Director de Tecnología de Adam Messinger, Director Financiero Mike Gupta y director de operaciones de Ali Rowghani.

Bienvenido a Silicon Valley, donde la escasez de los mejores talentos de ingeniería en medio de una explosión de capital de riesgo respaldados por cheques de pago de nueva creación se está inflando.

"El número de unidades de los jugadores en Silicon Valley no ha crecido ", dijo Iain Grant, un reclutador en Riviera Partners, que se especializa en la colocación de los ingenieros en el capital de riesgo respaldados por empresas de nueva creación. " Pero la demanda de los mismos se ha ido por las nubes. "

Abundan las historias sobre los extremos a los que los empleadores van a atraer talento de ingeniería - además de los comedores gratuitos, servicio de lavandería y autobuses lanzadera que los Googles y Facebooks del mundo ya son famosos.

Una start-up ofrece un ingeniero codiciado contrato de arrendamiento de un año en un sedán de Tesla, que cuesta en el barrio de $ 1,000 al mes, dijo el capitalista de riesgo Ganesan Venky. Se negó a identificar a la empresa, que su firma ha invertido pulg

En el Hotel Tonight, que ofrece una aplicación móvil para reservas de hoteles de última hora, CEO Sam caña describe puesta en escena de la oficina a aparecer viva adicional que considere contratar. Él cordada de dos empleados de una partida de ping- pong y se coloca otro grupo justo al lado del bar.
Funcionó : el recluta firmó y construyó una pieza clave de software de la compañía.

En el caso de Fry, su retribución fue en su mayoría en forma de adjudicaciones de acciones, por un valor el año pasado en 10.100.000 dólares, de acuerdo a los documentos de salida a bolsa de Twitter registradas en los reguladores de valores. Sacó un sueldo de $ 145.513 y un bono de $ 100.000.

Algunos pueden llamar a ese mal pagados. Vicepresidente de ingeniería, Mike Schroepfer, de Facebook Inc tomó en $ 24,400,000 en premios de valores del año 2012 antes de la oferta pública inicial de la red social. También señaló a un sueldo de $ 270.833 y un bono de $ 140,344. Pero Facebook ese año registró ingresos de 3,71 mil millones dólares, 10 veces más que el Twitter de $ 317.000.000.

Grant dijo más de tres cuartas partes de los candidatos que tomaron VP de funciones de ingeniería a sus empresas clientes en los últimos dos años atrajo la compensación total en efectivo de más de $ 250.000. Muchas subvenciones de capital también recibieron un total de 1 a 2 por ciento de la compañía, agregó el reclutador.

CAPACIDAD DE 10X

La alta demanda para los ingenieros es impulsado en parte por un creciente número de empresas de nueva creación, los capitalistas de riesgo dicen. Unos 242 empresas Área de la Bahía recibieron financiación en fase inicial - conocida como una ronda semilla - en el primer semestre de este año, según la consultora CB Insights. Esto es más que el número para todo el año 2010.

Otro factor es la creciente complejidad de la tecnología. Muchos en Silicon Valley les gusta hablar sobre la historia del ingeniero de capacidad "10x", que es una persona tan talentosa que él o ella hace el trabajo de 10 ingenieros meramente competentes.

"Tener ingenieros 10x en la parte superior es la única manera de reclutar a otros ingenieros 10x ", dijo Aileen Lee, fundador de Empresas de vaquero, un fondo de riesgo en fase inicial.

Antiguos colegas dijeron Fry, quien se unió a Twitter a principios de este año, encaja a la perfección. El servicio de mensajería le escalfados del gigante del software Salesforce.com Inc, donde Fry había trabajado en varios puestos desde 2005, pasando de director de ingeniería del equipo de Servicios Web con el vicepresidente senior de desarrollo.

Quizás lo más atractivo para Twitter es el hecho de que Fry se unió a Salesforce cuando también era una empresa de 6 años de edad, con grandes ambiciones de asumir la creación de software. En ese momento, el desarrollo de productos de Salesforce necesitaba ayuda, Fry ha dicho en entrevistas anteriores. Él los sacó en forma, ayudar a construir la empresa en uno de los proveedores más populares de software para empresas de la industria actual.

Twitter ha tenido su cuota de problemas técnicos, como el famoso " fail whale ", que apareció regularmente en las pantallas durante las interrupciones. Eso hizo que la experiencia de Fry aún más valioso.

" Todo lo que necesitas es un par de malas incidentes en los que Twitter está caído, o que hay una brecha de seguridad. Ese podría ser el final de la compañía", dijo Chuck Ganapathi, un empresario que ya había trabajado con Fry en Salesforce, donde fue vice presidente de productos.

" Necesitas a alguien de este calibre para ejecutarlo. "

Ni Twitter ni Fry respondió a solicitudes de comentarios.

ESTUDIO DEL TAMBOR PERSONAL

Hoy en día, incluso los ingenieros de nivel básico pueden sacar jugosos sueldos en el Valle. Google Inc ofreció $ 150,000 en salarios anuales más $ 250.000 en opciones de acciones restringidas para enganchar un recién graduado de doctorado que había estado considerando un trabajo en Apple Inc, según una persona familiarizada con la situación.

El ingeniero promedio software manda un salario de 100.049 dólares en Silicon Valley, de acuerdo con los dados, un servicio de tecnología de la contratación. Eso está por debajo de $ 113.488 el año pasado, debido a un aumento en la contratación de ingenieros con menos experiencia, dijo un portavoz de los dados.

En comparación, el salario promedio para todas las profesiones en el área de San Francisco Bay es $ 66.070, según la Oficina de Estadísticas Laborales. Otros trabajos en el área pueden exigir salarios más altos - los médicos a tomar $ 133,530, un abogado acerca de $ 174,440 y un ingeniero civil hace 107.440 dólares -, pero la industria de la tecnología a menudo ofrecen acciones u opciones en la parte superior de los salarios restringido.

Incluso para los ingenieros de llano- vainilla, la competencia es intensa, dijo el CEO Mike Dados Durney, llevando a las empresas a ir muy lejos para atraer y retener a las personas adecuadas.
Servicio de hospedaje de investigación ApartmentList alquila un estudio de tambor en forma permanente para ayudar a retener un ingeniero clave, dijo el CEO John Kobs.

En uno de los ejemplos más conocidos, famoso Google permitió a los ingenieros dedicar el 20 por ciento de su tiempo a proyectos personales. Vale la pena, muchos reclutadores y ejecutivos de la industria.

Muchos de los ingenieros más talentosos llevar más de chuletas de programación, la promoción de la diversidad de la especie de carrera valorado en Silicon Valley.

Tome Fry, quien obtuvo un doctorado en ciencias cognitivas de la Universidad de California en San Diego en 1998. Él es un surfista, un marinero y un snowboarder, de acuerdo con su página web personal.

En un giro apropiado para Twitter, conocido por su mascota pájaro azul, Fry también tiene experiencia aviar. Su estancia posdoctoral en la Universidad de California, Berkeley, se centró en la corteza auditiva de los pinzones cebra.

( Editing by Edwin Chan, Tiffany Wu y Richard Chang )

Este artículo apareció originalmente en Reuters. Derechos de autor 2013.

Business Insider

Read more »

¿La oferta atrae a la demanda? (2/2)

Part II: How VCs Test Market Demand

ED ZIMMERMAN: My last WSJ Accelerators piece focused on how successful repeat founders tested their offerings/company prior to their launch. In this installment, we turn to how VCs test market demand for a product when conducting diligence prior to investing. While this is important for founders like those discussed in the last article, without the track record those founders had, diligence on market adoption becomes even more important.

So, what if you’re not David Chesky of HDTracks, Matt Keiser of LiveIntent, Amanda Hesser & Merrill Stubbs of Food52 or Craig Danuloff of Rewind.Me? What if, instead, you’re an unproven founder? How will VCs test the market demand for your company before investing?

To pull back that particular curtain, I called Ian Sigalow, co-founder and general partner at Greycroft Partners, an early stage tech VC. 2012 has been kind to Sigalow and his colleagues, as they’ve just closed on their third fund ($175M) and scored major exits when portfolio companies Buddy Media and Vizu were recently acquired. Greycroft initially invests $500K to $5M and typically increases that position over time. The fund also has a seed program which, Sigalow says, generally focuses on repeat founders re-entering a market in which they’ve proven themselves.

Sigalow would have sat out Rewind.Me, which raised before launching a product. When Greycroft eyes a consumer-focused startup “we need to see user traction — I can’t look at a product and say ‘this will be the next great thing’ because there are elegant products that never get adopted and crappy products that do.”

In fact, when Sigalow looks at a company likeHDTracks, which targets consumers, he fragments the market into two groups: companies charging consumers, and those making money based on large-scale user adoption (for instance, an advertising-based model). “If it’s monetized by advertising, then a few hundred users makes it hard for us to tell whether it will work (because the company will need millions of users). If, alternatively, (users are) paying a subscription each month, a few hundred users is probably adequate for us to determine what works.”

What about B2B companies (called “enterprise” businesses) like LiveIntent? While a fund like Greycroft may need hundreds of paying or 30,000+ free consumer users to feel like there’s sufficient traction, for enterprise companies, the fund may find sufficient traction with only half a dozen large customers using a product for free in beta.

Sigalow explains that for products sold as “freemiums” (the company provides a base product for free and charges for enhanced versions), Greycroft, not surprisingly, looks at whether customers want to pay for it.

Once they’ve decided they like a company enough to commence diligence, the Greycroft team compiles a list of dozens of people in their own network — experts as well as actual and potential customers in addition to references and customer contacts provided by the startup. Each call may take around half an hour to cover whether customers would pay for the service, to whom they’d recommend it and why, and who else they considered (or would consider) using.

Sigalow doesn’t just rely on the phone. Greycroft heavily uses Instant.ly, an online offering from Los Angeles-based uSamp. Instant.ly “has a 10 million person panel so we can target by age, geography, relationship status, job title, etc. and get thousands of survey results back in real time as if they were our own contacts.” You can specify how many survey results you want and, because Instant.ly usually has some 4,000+ users on its site, you can begin seeing results almost immediately. A few hundred responses from people Greycroft doesn’t know complement the data gleaned from calling their own network.

Sigalow underscored that “some funds like to risk more capital at an early stage. Seed funds, of course, are often doing pre-product.” Angels too invest pre-product, as I did when I invested in Rewind.Me and LiveIntent, but Sigalow’s detailed explanation is not far from how other Series A investors approach diligence prior to funding a promising startup.

Intrigued by Instant.ly, I reached out to uSamp founder Matt Dusig to see how the tester had done his own market testing before launching Instant.ly. “We built an entire business before building Instant.ly” he said, referring to uSamp, which “provides online market research panels and SaaS technology for global market research.” USamp receives inbound requests from corporations looking for market research. Those clients repeatedly told uSamp “we love working with the marketing agencies we use, but sometimes we can’t wait a month or longer for feedback, we want it instantly.” So it was less a question of Dusig testing the market and more of the market shouting at him, which, of course, is a great way to find a hole in the market. Much has been written about “founders in search of a problem” rather than those motivated by an existing problem.

Now that startups have become stylish, there’s a concern that people spend too much time seeking a problem rather than building useful experience and contacts which then would, almost in the ordinary course, reveal problems that desperately demand solving. USamp is definitely in the latter category. In fact, when asked, Dusig explained that prior to founding uSamp, he and his cofounder had built and sold another market research firm and then moved into another online business during the post-sale noncompete period (again, knowing what the market needed trumped testing).

This model of the repeat founder doubling down in a space he or she knows is precisely what VCs crave and, of course, that’s why Sigalow’s fund led the second venture round into USamp, which has now raised approximately $20M from Greycroft,DFJ Frontier and, more recently, Adam Marcus at OpenView Venture Partners.

So are you going to test the market or let the market shout at you? Please post a comment if you have questions or thoughts about the way investors diligence startups or if you have some great ideas and tools for testing the market for consumer or enterprise tech companies. We’ll do our best to answer!

WSJ

Read more »

¿La oferta atrae a la demanda? (1/2)

Part I: Build It and They Will Come?

ED ZIMMERMAN: “Passion. Instinct. That was it. Blind faith,” responded David Chesky, co-founder of HDTracks, when I asked him how he had tested the market before investing his time, reputation and cash to found his company and build his product. But let’s back up because I’ve been tweeting a lot about HDTracks — for instance, 9 months ago:

RT @EdGrapeNutZimm: don’t need discounts, I’m sold @HDtracks -the only place2hear it all (liberate me from compressed formats) #Hifi #Music

Why? Well, I’m a passionate music fan and an ignorant audiophile geek. That means I prefer tube amps over solid state, but I can’t explain the engineering underlying that preference. I also know that when I hear mp3 compressed files or the 1982 technology used in CDs, I feel cheated. When I hear uncompressed .flac or .aiff files (music stored in formats that let us hear all the info), it makes me happy and my kids and I are able to hear previously unheard details in familiar performances.

Here’s my test of any stereo equipment or new recording format: Does it make me want to listen to my favorite stuff all over again? HDTracks does precisely that. HDTracks has no outside investors; heck, I’m not even their counsel. I’m just a devoted customer. I’ve always loathed audiophile recordings because they’re songs you like by people you’ve never heard of, or people you’ve sort of heard of performing music you’ve never heard of (like a “worst of” compilation). Chesky solved that problem by having NYC-based HDTracks quietly strike licensing deals with the big music labels. He’s now actually making money for all involved and enabling fans to download an ever-increasing list of great albums in high def, uncompressed sound: Beach Boys, Clapton, Stones. But it isn’t just older stuff: Green Day, Tift Merritt andthe National have each released new albums on HDTracks.

Chesky has two advantages: (1) music is something about which many of us feel passionate and (2) in his words: “Here is the market research: …I am an audiophile who wants this and thinks it’s cool; therefore the rest of my kind will want it as much.” Of course, the question now is how big the market is – because he’s going to take it over, and if there are indeed a lot of “the rest of his kind,” we’ll all be sorry we haven’t invested!

Craig Danuloff, a serial founder, started Rewind.Me for the same reason: “I’m more of an ‘I think this is cool so I’m building it’ guy.” Rewind.Me enables us to use past check-ins, posts, tweets and other social media to take advantage of our own digital history – ‘have I been here before?’ (Disclosure: (a) I’m an investor and (b) Craig is an audiophile with a sick amount of .flac files…maybe this approach is just for us audio geeks?)

Are Craig and David crazy or lazy? Not according toJeff Richards, a partner at Silicon Valley and China-based GGV Capital, a leading growth stage fund that has backed Pandora, Buddy Media, Alibaba, SoundCloud and other great companies here and in China. (Disclosure: I’m an investor in GGV Capital).

Jeff says fewer people do testing “given how cheap it is to launch a product now – primarily due to the low cost and on demand availability of infrastructure (AWS [Amazon Web Services]) and ease of distribution (app store) – most entrepreneurs I meet are in a ‘launch and iterate’ mode. The flip side…it’s very hard to remain on top – competitors can launch and iterate quickly as well.” This doesn’t just apply to consumers – sure, we consumers are happy to be early adopters of iPhones or new cameras, but “perhaps even more interesting … I see enterprise customers willing to be part of the cycle – taking a chance on relatively new technology and in return getting to help shape the product. A decade ago no enterprise customer wanted to touch a product until it was ‘fully baked.’ ” This is an important change.

One of our amazing FirstGrowthVN companies, Food52, takes the “splash page approach” (which I see frequently used): “We put up a splash page before we launched to collect email addresses, and we let them join in stages. This was less about testing the specific concept and more about seeing if people had an appetite for a new kind of food site, specifically one coming from our perspective (or really [cofounder Amanda Hesser's], since she had already built such a strong reputation),” according to Merrill Stubbs, co-founder. Amanda Hesser garnered that reputation as food editor of The New York Times Magazine, author of several books, including “The Essential New York Times Cookbook,” and by playing herself in the film Julie & Julia(coincidentally, I’m available to play myself in a major motion picture…for the right price…just letting y’all know!). Here’s how Hesser leveraged that reputation for Food52: “We first tested our recipe contest system via email with family and friends. Then we bootstrapped by getting a book deal, and using the book advance to fund a year of testing.”

But what if you don’t have a product about which people are impassioned?

When Matt Keiser founded LiveIntent (in which I’m an investor), he marketed enterprise customers before he built product by using a PowerPoint deck as a proxy for a costly product (he saved the money up front that Danuloff and Chesky spent in building their actual product): “We made a bunch of mock-ups and a killer PowerPoint. We met with as many potential clients as possible. We refined the pitch until we got five very positive responses for every lukewarm response. We then built what we thought was the right product. It took a few tries. Year one is great, you believe your vision is correct. Year 2 sucks as you’ve learned that your baby has warts and you don’t know if you will be able to fix them. Year three rocks if you survive. Radical innovation is key if you want to be noticed.” He’s telling the truth – Keiser knows I hate to sit through PowerPoint decks but he’s made me do it repeatedly, and then I somehow find myself buying him dinner…talk about a great salesman! Still, Keiser had to rebuild that product (the wart-covered baby) and rebuild he did. It took some time, but boy has LiveIntent now found its groove!

Chesky, a 3x Grammy Award nominee, is deeply experienced and connected in his sector. What if you’re not? Keiser and Danuloff are serial founders who have successfully exited other startups. As for Hesser, well, she’s got plenty of street cred from her NYTimes gig and successful books. What if you don’t have these amazing credentials? How will VCs do their diligence and consider the market for your company’s products? Stay tuned for the next installment!

WSJ

Read more »

Como puede un emprendimiento evaluar el tamaño de mercado

How can a startup company best evaluate market size and find the market data?

Often times investors wants to know market and industry potential and stats of the startup product, how can startups best find and present these informatin?

In this blog post, my aim is to solve two common problems for startups.

1.      What is your total market potential?

2.      How much of that market can/will you capture?

Although it may be impossible to answer these questions accurately, you can use a data-driven, bottom up, and logical approach to help answer these questions in a way that will satisfy potential investors and bankers.
Calculating Market Potential – Example
I want to start by giving you a few tips and tools that you can use to help calculate your total market potential, and then I want to show you a real example of how I calculated the market potential for a product I created for this blog.

1.      Industry Reports – Depending on your market, it may be that a market research firm has already done this work for you.  Don’t make things hard on yourself, if you sell dog food try Googling “Size of Dog Food Industry.”  When I did this I found www.petfoodindustry.com which told me that by 2015 the global pet food industry should reach $56.4 billion.  I am certain that their extensive report will break down the industry by type of pet, so you would be able to find the market potential for the dog food industry.

2.      Competing Websites – This does not work for every industry, but for many industries I think this will be a good data point to use when determining the market potential.  Go to www.compete.com and search for your competitors websites.  If their websites are getting significant traffic you should be able to see an estimate of their traffic.  If you know that your competitor has 100,000 website visitors per month, at least you know that the market potential for your website is at least 100,000.

3.      Google Adwords Keyword Tool - Finally, I would suggest that you use the Google Adwords Keyword Tool to determine how many searches related to your market take place on Google each month.  The data that Google provides through this keyword tool gives marketers no excuse for building a product or service that no one wants, because Google users are telling you exactly what they want through search data.

Now let me show you exactly how I determined the market potential for a product that I released on my blog called the Executive Summary Toolkit.

First, there are no industry reports that show what total market is for business plan executive summary resources, and there are not any other websites that compete specifically for executive summary related keywords only.  There are a number of business plan websites that compete with me.  Based on the Compete.com data on Bplans.com they receive over 160,000 visits per month.  This gives some idea of the scale of the potential market.  Bplans is the leader in business plan software, templates, examples, and advice, so this number helps me know that my market is probably not 10 million visitors per month, but it is also more than a couple thousand visitors per month.

Determine Market Potential with Google Keyword Tool

Now armed with info from Compete.com on my major competitors, I use the Google Adwords Keyword Tool to find out how many searches occur every month related to executive summaries.

So I used the keyword tool and typed in “executive summary” this gave me the monthly search traffic for that keyword, along with search traffic for dozens of related keyword phrases.  I picked the top 14 relevant keyword phrases, and then added a 15th line item for “All Other Relevant Keyword Phrases”  I listed those keyword phrases and their corresponding monthly traffic estimates.

I added all of this traffic together and came up with a total of 724,620.  Now this is NOT your market potential yet.  There are a number of considerations to include first.

1.      This is only Google search data.  Google dominates the search engine market in many geographic locations, but there is still a healthy 33% or more that do not use Google as a search engine.  So I would simply multiply that Total Search Traffic Number by 1.33.  In my example, that gives us 963,744 as the new total search traffic number.

2.      Now I want to propose for this specific situation that at least one half of these searches are the same person searching more than once, and gradually refining their search terms.  So a user may start with a general search for “Executive Summary”, but what they really want is an “Executive Summary Example Template.”  So your market potential is actually smaller than that total number because each search does not equate to exactly one person searching.  Make sense?  SO… now I am going to cut that Total Monthly Searches number in half, which brings it down to 481,872.

3.      Lastly, I want to point out that there could be potential customers that are not searching for your product or service.  Some businesses are primarily social, so you might generate a majority of your website traffic through Twitter, Facebook, and Google+.  Check out your competitors, are they active in social media?  For me, I know that historically, 25% of my website visitors come from sources other than search engines.  So for the purpose of this example, I am going to multiply 481,872 by 1.25.  Now we end with a market potential of 602,340.

4.      The last consideration is how many of these potential visitors will actually pay for something?  Again this is going to vary by industry, pricing, product quality, etc.  Unless you have some historic data you are going to have to guess here.  I would suggest a LOW number. I know that approximately 1.25% of my website visitors will purchase an Executive Summary Toolkit.  If that is a characteristic of the market as a whole, then I can take 1.25% times 602,340 which gives me 7,529 potential buyers.  If I multiply that by the purchase price of the executive summary toolkit which is $4 and multiply by 12 months I end up with a total annual market potential of – $361,400.

I think we have come up with a data-driven market potential for my executive summary toolkit product, but unfortunately I can not expect to capture the entire market.  Some of you might be working in industries that have a total market potential of hundreds of millions or even billions of dollars, so now the question is how do you project the market share that you can capture?

How to Calculate Potential Market Share

I am going to continue with my example, and ultimately come up with, what I hope is a realistic financial projection for my Executive Summary Toolkit.  In order to do this I am going to take a few steps back and start with our total potential visitors which amounted to 602,340 per month.  We can break this big number down with click through rate data.  As you know, there are 10 search results displayed on the front page of Google, but every searcher does not click on all 10 search results.  Thankfully, Slingshot SEO put together a great report that outlined the click through rate for a search result in each of the top 10 positions on Google.  See the table on the right.

So based on your search position the potential market share that you can capture changes.  For most search phrases your website will not show up more than once in the top 1o results, so the maximum click through rate you should expect is 18.2% if you were in the top search position.

As a startup, it takes time to get your website to rank for the keywords that you want to rank well for, so in the first 6 to 12 months of business your traffic may come entirely from sources other than search engines.

Once you start to rank in the top 10 results for some of your keywords, you can better calculate your market potential by simply applying the click through rate percentages to each keyword phrase.  For example we can look at a few of the keyword phrases that I rank well for in Google:

·        Executive Summary Example – Rank 7 – Total Searches 49,500 – CTR for Rank 7 = 1.9%

·        Executive Summary Format – Rank 9 – Total Searches 6,600 – CTR for Rank 9 = 1.5%

·        Example of Executive Summary – Rank 8 – Total Searches 1,000 – CTR for Rank 8 = 1.7%

With some simple math you can determine how many visits you should receive from each keyword phrase.  Rather than bore you by doing this for every single one of my keyword phrases, I am going to simply utilize the data that Google Webmaster Tools provides me.  For the last month Google Webmaster Tools tells me that my website has showed up in search results 250,000 times, and I have received 12,000 clicks.

Now that I have an estimate for how many people will actually visit my website, I can apply my 1.25% sales conversion rate to my 12,000 visitors per month.  This gives me 150 purchasers per month times $4 sales price times 12 months and I end up with $7,200.  Finally this equates to 1.9% of the total market potential that we calculated above.

After going through this entire process you can now say confidently that your startup plans to capture X percentage of the total market.  Rather than simply pulling a number out of thin air, you can truly use a bottom-up, data-driven approach that will put you ahead of your peers when you present your startup to potential investors or bankers.


Quora

Read more »