Marketero · Entrevistas
Ben Tiernan from SepulvedaPartners explains his Multi-Touch Attribution Suggestion Engine, I LOVE IT
25 ideas de marketing sacadas de «Ben Tiernan from SepulvedaPartners explains his Multi-Touch Attribution Suggestion Engine, I LOVE IT», de Professor…
- 25 ideas despiezadas
- Fuente Professor Charley T
- Dura 1 h 02 min
- Publicada en 2022
Ver la entrevista entera en YouTube
De qué habla
Las 25 ideas de esta entrevista se reparten así entre las etapas del marketing. El reparto es lo que distingue una entrevista de otra: dice en qué se detuvo de verdad quien hablaba.
Las 25 ideas, con su cita y su minuto
Cada una está sacada de un punto concreto de la entrevista. El enlace del titular lleva a
la página de esa idea, donde se ve quién más la dice por separado; el del pie salta al
segundo exacto del vídeo para que la compruebes tú.
Optimización del ecosistema global frente a la atribución individual de canal
Medición y experimentosOptimizar un canal de paid media como motor de generación de leads para nutrir un CRM potente suele generar mayor rentabilidad global para el negocio que forzar la compra directa con bajo ROAS, aunque a nivel de reporte individual el canal aparente peores resultados al alejarse de la conversión final.
there's a conflict and the business relationship me as an agency running your Facebook there's an extreme conflict for me to share and potentially look worse if I decide my real goal is Facebook is doing okay but your CRM is crushing your email team is excellent really maybe the best move for you is let's run Facebook more or less it's a lead generation to your CRM team and let them just Crush maybe the value of an email address is 50 and I can get you one for 12. maybe it's half as good as everything else so that's 12 is only worth 25 bucks you're just doubling your money or if I work really…
Escúchalo en el minuto 37:01
Medición y experimentosOptimizar la pauta publicitaria fijando como único objetivo un ROAS alto dentro de la plataforma limita el negocio; aceptar un ROAS menor (por ejemplo, 2x) con mayor volumen y escala suele aportar mayor crecimiento incremental real que sostener un ROAS elevado (por ejemplo, 5x) a bajo volumen.
um are taught to Value their work on kpis that don't directly access to the business maybe it's better for me to get a two x row as but deliver three times as many impressions as 5x row as because that number doesn't actually mean success your job as a as a performance marketer is rarely ever spend X money at X row that has nothing to do with the growth of the business like business how you can contribute to the incremental growth of the business
Escúchalo en el minuto 17:21
Tratar las plataformas publicitarias como palancas macro en lugar de microoptimizar audiencias
Canales de adquisiciónMicrogestionar decenas de campañas y audiencias en cada canal publicitario apenas ofrece mejoras marginales (1-2%). Resulta más eficiente y escalable estabilizar cada plataforma con pocas partes móviles para lograr predictibilidad y concentrar la optimización en la asignación presupuestaria a nivel macro entre plataformas, donde los cambios de inversión tienen un impacto mucho mayor.
uh On Any Given platform let's say as a for instance I've got five platforms everyone from Scott say five campaigns 25 total and each one of those let's say has five audience for the purpose of this conversation you've got 125 bears the best execution platform level for any one of those audiences means I might be able to at best make a one or two percent addressing which one of those five platforms to spend more on just to store that lever or pull it back I'd make it ten percent that I've taken that in is say my objective is not to maximize the value my objective is to stabilize the platform…
Escúchalo en el minuto 26:45
Medición y experimentosUn modelo analítico u econométrico solo puede optimizar dentro del universo de datos que ya conoce. Para descubrir si una escala mayor de gasto o un nuevo canal realmente funcionan, es indispensable realizar experimentos agresivos que superen los límites históricos y generen datos nuevos.
so that's that's exactly why we have the like we just need new data we need something to go outside the bounds no so we can say oh that one really or that really does work um otherwise a model will just optimize you within the world that of data that
Escúchalo en el minuto 45:33
En compra de medios, las apuestas estratégicas de alto riesgo mueven la aguja más que la microoptimización
Medición y experimentosEn paid media a gran escala, la filosofía de mejora continua marginal (estilo Kaizen o 1% diario) es insuficiente frente a realizar menos apuestas estratégicas pero de alto impacto y riesgo. Si los experimentos o anuncios ganadores se suceden con demasiada facilidad y frecuencia (por ejemplo, cada tres semanas), suele indicar que el equipo no está asumiendo riesgos suficientes para transformar el rendimiento del…
to get to like I had a boss once uh we were doing it's like a you know nine-figure SAS brand right it was a seven figure monthly profit Target on paid so and he told me that if I tested I was right I had to win more than every 60 to 90 days that wasn't working hard like I needed to like if I won every three weeks then I wasn't doing my job like I had to take a big enough risk that had moved the needle so much and I fewer strategic moves of higher confidence can move things far more meaningfully than a thousand little things that ultimately are one percent change we have this idea of like…
Escúchalo en el minuto 43:17
Sesgo de atribución en plataformas publicitarias por falta de visión omnicanal
Medición y experimentosLas plataformas publicitarias individuales (como Meta o Google) evalúan su rendimiento en silos y se atribuyen conversiones impulsadas por otros medios (como TV u otros canales). Al carecer de visibilidad sobre el mix global de marketing, cuando se apagan campañas externas, los modelos de la plataforma sufren caídas de rendimiento que no pueden explicar por sí mismos.
they can model against they don't actually know how they're contributing very much compared to all the other data sets which are like if they're running TV Facebook's running around the thing is doing great because it's it is judging its performance versus have a TV campaign going sure but any trim the campaign off and Facebook's trying to model that they're like why'd we stop working sure right absolutely yeah and what's going to be hard for these platforms they away from cookies and um because they don't have the
Escúchalo en el minuto 34:17
Transición del ROAS por plataforma a la eficiencia de marketing global
Medición y experimentosEl análisis publicitario moderno se aleja de la métrica de ROAS aislada por plataforma (donde cada canal compite por atribuirse la venta) y migra hacia la eficiencia global de marketing (MER), evaluando cómo el gasto total impulsa los resultados en todo el ecosistema del negocio.
because it's starting to understand that some of that worked and I've noticed the conversation really switched from platform level row as to marketing efficiency how much of my overall spend do I adjust but even with that it's still this retroactive self-preservation
Escúchalo en el minuto 38:15
Medición y experimentosLa democratización del big data y el procesamiento avanzado permite integrar datos minuto a minuto de impresiones, clics, ventas entre plataformas y variables externas (clima, economía) para modelar la contribución real a nivel de canal y campaña, optimizando la asignación presupuestaria entre medios tradicionales y digitales.
democratization of Big Data uh Computing and things like that yeah give it all give us your stuff by the minute give us all your impressions all your clicks every all your sales data across all the platforms that was all the economic data give us all the environmental data capability of kind of building models real decisions about how we allocate our fund better predictions element of it you can look down to the campaign level across the board and you can look across what we do is like you can bring in linear data and you can simply have a Dr yeah but you can also get it down to the channel…
Escúchalo en el minuto 10:10
Escalar solo el anuncio con mejor CPA final no escala la adquisición futura
Medición y experimentosLa atribución retrospectiva muestra qué funcionó en un contexto pasado aislado, pero no predice el comportamiento futuro; sobreinvertir únicamente en el anuncio o punto de contacto final con mejor CPA no amplía la captación de nuevos clientes si no se alimenta el resto de la demanda.
that doesn't tell me anything about what future people are doing that also sort of assumes that the Final Touch point is an actionable thing like if I use the solution isn't to hire 10 000 car salesmen that's not going to sell more right so finding the ad that had the best CPA today and then spending a bunch more on that ad isn't the way that I'm going to keep getting new customers in so the problem I have with that is retroactively looking at some period of time to say how did each ad do in isolation doesn't really tell me how to manage my media
Escúchalo en el minuto 3:57
ConversiónCaptar demasiados tipos de audiencias hipersegmentadas en paid media genera fricción en las etapas posteriores del negocio: obliga a adaptar landings, secuencias de email y procesos de venta para resolver demasiados problemas distintos a la vez, lo que diluye la eficacia global del embudo.
if I'm addressing 125 different parts how many different types of customers am I bringing into my how many different people am I trying to solve for like my sales people if it was different types of people to my how am I going to optimize my landing page I'm gonna figure out my email flow or like figure out on TV if I have to solve 120 different problems I might get a 10 game someplace else and that impact is minus three because I've heard 20 other things
Escúchalo en el minuto 28:03
Medición y experimentosUsar datos históricos de atribución de forma aislada para decidir incrementos de presupuesto en campañas o canales específicos suele fallar: tan pronto como se modifican los niveles de inversión, las correlaciones y la efectividad del mix de medios cambian drásticamente respecto al modelo previo.
but it only holds true as long as you don't move any barrier as soon as you do it changes dramatically and generally speaking most of these people are using that as information to say great spend up more on this ad set or on this campaign and then that effectively immediately destroys the entire medium mix so you are looking at out of context data historically to plan future action under the assumption that that mix of credit kind of holds true
Escúchalo en el minuto 5:11
Umbral de inversión publicitaria y volumen de impresiones para modelado de atribución con machine learning
Medición y experimentosPara generar suficiente volumen de impresiones y conversiones que permitan entrenar modelos de atribución robustos mediante machine learning, se requiere un gasto anual aproximado de 5 millones de dólares con un CPM promedio de 10 dólares (o una inversión mayor si los CPMs son más elevados).
exactly the thing so there's a certain uh it it said if you're running about five million a year at about an average ten dollars CPM um that will give you enough Impressions to be able to modeling and if they're running at like 15 20 30 it just means they need a greater in cpms just you need a larger level but you're netting out of that the impression volume data yeah so and enough conversion information to get an understanding of yeah machine learning so the
Escúchalo en el minuto 59:33
Medición y experimentosEn negocios donde factores estructurales como la ubicación física o el tráfico orgánico determinan las ventas base, la inversión publicitaria puede no correlacionar directamente con los picos de ventas totales, sino con la ganancia y fluctuación de cuota de mercado.
hey turns out your marketing has no relationship to your sale or not no but like the correlation is far less than there's all these other things that are driving cells you know think of it as kind of uh location you know foot traffic there's all this other stuff but what your marketing does is driving market share uh so like your your market share goes up and down a very close correlation we're able to kind of snap
Escúchalo en el minuto 48:25
Reducción del 80% en la variabilidad del CPA diario mediante visibilidad analítica
Medición y experimentosMejorar la visibilidad sobre la efectividad de las campañas puede reducir la varianza del costo por adquisición (CPA) diario hasta en un 80%, estabilizando la curva de rendimiento y aumentando la eficiencia de la inversión.
better visibility into what was working but a a company has got a variance in their daily CPA reduced by basically 80 uh so the state the bell curve yeah instead of being like this yeah the bell curve was like this and um the average better so yeah it was more efficient more and more efficient
Escúchalo en el minuto 50:25
Alinear los modelos de medición con los objetivos del negocio y no con métricas aisladas de canal
Medición y experimentosLa analítica y el modelado de atribución deben responder a las necesidades financieras y estratégicas de la empresa (la marca y su rentabilidad) en lugar de orientarse a satisfacer las métricas tácticas o de vanidad de ejecutores y agencias que solo buscan justificar su canal específico.
a lot of people are trying to meet the needs of the marketer yeah instead of meeting the need of the business right and I think you guys are solving for the needs of the business right and that gives direction to the marketer instead of feeding the need of the marketer because your client isn't the agency trying to keep brand right your client is the brand trying to find the right
Escúchalo en el minuto 33:04
Medición y experimentosPara planificar la inversión publicitaria de forma dinámica, los modelos predictivos deben incorporar datos contextuales externos (fuerzas de mercado, estacionalidad y acciones de competidores principales) en lugar de depender exclusivamente de reportes retroactivos de rendimiento interno.
instead of being a retroactive report suggestive machine that uh dynamically changed and accounted for that different weighting and the changes in the marketplace just deliver a specific um uh then we also bring in environmental data so that we can kind of anticipate other Trends yeah yeah like Market forces and all that other most marketers aren't accounting for that right right and so then you go back where it's completely irrelevant to look at last year's data to understand future competitive your major competitor always comes out with a major push during this
Escúchalo en el minuto 6:39
El volumen de impresiones no equivale al poder de persuasión en la atribución
Medición y experimentosAsignar valor o crédito de conversión a un canal simplemente porque acumuló más impresiones en el recorrido del cliente es engañoso; mayor volumen de impactos no implica necesariamente mayor capacidad persuasiva para generar la venta.
um but saying it's a 100 kind of uh value on the conversion saying this you know uh social media or Facebook made you twenty dollars and search maybe fifty dollars and display may do thirty dollars of kind of transaction not sure it's exactly true I think just because you something got the most Impressions maybe got 10 times more questions doesn't mean that it was the most persuasive right sure or should be given equal amount credit for persuasion as it is for
Escúchalo en el minuto 1:50
Medición y experimentosFijar el presupuesto de marketing mediante ratios fijos sobre la facturación proyectada (por ejemplo, asignar 1M para generar 2M) genera una falsa sensación de eficiencia orientada a datos, cuando en realidad es una regla contable arbitraria carente de modelado estratégico.
you take in a lot of organizations uh there are people CFOs uh you know heads of strategy like setting the media but or setting the entire marketing uh budget based off of um and there's a GMO showing me all this data-driven stuff but it's not real strategy you know it's not it's not yeah it's not the goods they might say oh you're gonna do two million so you have one million to spend so we have a marketing efficiency against Revenue yes
Escúchalo en el minuto 14:27
Medición y experimentosLa atribución multi-touch (MTA) basada en etiquetado exhaustivo de eventos y pixeles impone una carga operativa excesiva y meses de desarrollo sobre el cliente, además de presentar un alto margen de error en comparación con enfoques de modelado estadístico.
but it's nothing is just difficult as doing all that gagging but yeah uh like all that tagging that's within MTA is just a really really time consuming and a lot of room for you also puts a lot of burden on the client yeah right absolutely like for us you spend six months on your Dev team a lighter weight and either I think that
Escúchalo en el minuto 57:31
Agrupar creatividades para evitar la dispersión de datos y evaluar métricas clave
Medición y experimentosTener demasiadas creatividades activas fragmenta el volumen de datos por anuncio impidiendo un análisis estadístico fiable; agrupar las creatividades por temáticas o categorías permite consolidar la muestra para medir métricas como el LTV o la tasa de conversión según el objetivo del negocio (flujo de caja, margen inmediato o adquisición de clientes).
and you're not running like a gazillion uh you're not running a gazillion different uh creatives we could usually how much spend yeah and it's like what's the what's the data set per asset we have a million assets yeah then you've got a very small data pool for one yeah maybe even if you have a lot of assets the nature could be summing it up in the group a yeah Group B Group C yeah and one of these has LTB of x one of these has conversion rate of Y are you trying to buy cash flow are you trying to buy uh profit margin today you're trying to do customer those are all different things that…
Escúchalo en el minuto 1:00:33
Medición y experimentosOptimizar campañas para que todas contribuyan de forma equilibrada a un CPA objetivo encierra la estrategia dentro de los límites de los datos conocidos. Este enfoque algorítmico perfecciona la eficiencia en el rango histórico probado, pero ciega a la marca ante saltos de escala o nuevos canales fuera del marco preexistente.
you have everything optimized perfectly it's everything to all the different campaigns are contributing same degree and everything's fully optimized but the issue with that is and that's a that's where the model or a model will make you it is like a perfectly a way to talk about that is a five dollar a CPA right like so we'll just optimize for the five dollar CPA and everything's contributing perfectly but you're only working within the data that you know right and if you're working and if you've always run your channel we've run these four channels a range of this
Escúchalo en el minuto 44:24
Medición y experimentosOptimizar campañas para que todas contribuyan de forma equilibrada a un CPA objetivo encierra la estrategia dentro de los límites de los datos conocidos. Este enfoque algorítmico perfecciona la eficiencia en el rango histórico probado, pero ciega a la marca ante saltos de escala o nuevos canales fuera del marco preexistente.
you have everything optimized perfectly it's everything to all the different campaigns are contributing same degree and everything's fully optimized but the issue with that is and that's a that's where the model or a model will make you it is like a perfectly a way to talk about that is a five dollar a CPA right like so we'll just optimize for the five dollar CPA and everything's contributing perfectly but you're only working within the data that you know right and if you're working and if you've always run your channel we've run these four channels a range of this
Escúchalo en el minuto 44:24
Planificación macro de inversión por canales frente a la micro-optimización táctica
Canales de adquisiciónAl buscar objetivos de crecimiento acelerado (como replicar la facturación del año anterior en seis meses), la asignación estratégica y periódica del presupuesto entre canales clave aporta mucho más impacto que perderse en la micro-optimización de audiencias o estilos de creativos.
happen might say what I need to do is every six months replicate what I did in 12 months before prior to that these are the four or five if that the decision tree that's gonna get me in a way that I've got the most confidence to so the ability to go into that machine and make that broad like not week by week planning but almost like epic by epic right like H1 H2 type planning and H1 we're going to replicate all of last year in that six months that's our goal get out of the weeds of audience or creative style like that stuff is way less meaningful when it's time to move
Escúchalo en el minuto 22:40
Medición y experimentosAceptar recomendaciones de presupuesto más agresivas introduce la varianza necesaria en los datos para que los modelos analíticos identifiquen rápidamente qué combinaciones de canales funcionan y cuáles no.
and if you're always taking the aggressive recommendation Mike you're taking more risk might be but uh you're also changing what's a model aggressively variation gives it more stuff to say oh that doesn't work right or that does need to take some of those risks you know they're they're you're getting confidence level recommendations exactly essentially just make a bunch of wrong decisions make a bunch of right decisions hopefully you're making more right decisions and wrong decisions and
Escúchalo en el minuto 20:46
La desconexión entre métricas de marketing y finanzas fomenta modelos de conveniencia
Medición y experimentosCuando los modelos de atribución y reporte se configuran principalmente para validar departamentos o justificar presupuestos frente a la dirección, se pierde el foco en el crecimiento real y se genera desconfianza ejecutiva hacia la disciplina de marketing.
whatever modeling makes sure that whoever runs every Department that's on the report look good that's the model choice that the companies the primary job of any marketers to make I think is to make the person hired you look good so you don't get the goal of a business is to grow and to understand Market opportunity and all I cynical business uh say that marketing um and they'll say that because there's just not enough Clarity on marketing and marketers themselves are that are useful for making this in um but aren't tied to kind of the major Financial
Escúchalo en el minuto 6:04
Seguir leyendo
Esta es una de las 1872 entrevistas despiezadas. Puedes
ver todas, leer las ideas por tema o
empezar por la ruta guiada si no sabes por dónde.