How Marketing Mix Modeling Unlocks Scale & Incrementality -- Rohit Maheswaran // Lifesight

MarTech 播客是 I Hear Everything 播客网络的正式成员。

The MarTech Podcast is a proud member of the I Hear Everything Podcast Network.

想启动或扩大你的播客？

Looking to launch or scale your podcast?

I Hear Everything 提供播客制作、增长和变现方案，帮你把内容变成真金白银。

I Hear Everything delivers podcast production, growth, and monetization solutions that transform your words into profit.

准备好为你的品牌发声了吗？

Ready to give your brand a voice?

那就去 iheareverything.com 看看吧。

Then visit iheareverything.com.

从广告到 SaaS，再到数据。

Advertising to software as a service to data.

在我们所有的节目和客户中，打开率普遍在百分之五十五到六十五之间。

Across all of our programs and clients, we've seen a 55 to 65% open rate.

让品牌自然融入内容，效果比电视广告更好。

Getting brands authentically integrated into content performs better than TV advertising.

一篇报道的平均寿命大约是二十四到三十六小时。

Typical lifespan of an article is about twenty four to thirty six hours.

如果我们能联系到对的人，传递对的信息，再加上明确的行动号召，那剩下的就只是时机问题了。

If we're reaching out to the right person with the right message and a clear call to action, then it's just a matter

了。

of timing.

欢迎收听MarTech播客，由I Hear Everything制作。

Welcome to the MarTech Podcast, an I hear everything production.

在这档播客里，你会听到顶尖营销人如何运用科技推动业务成果、实现职业突破的真实故事。

In this podcast, you'll hear the stories of world class marketers that use technology to drive business results and achieve career success.

我们会深入挖掘营销与科技领域那些顶尖人才的实战经验，让你学到他们一路积累的工具、技巧和诀窍。

We'll unearth the real world experiences of some of the brightest minds in the marketing and technology space so you can learn the tools, tips, and tricks they've learned along the way.

接下来，有请MarTech播客主持人本杰明·夏皮罗。

Now here's the host of the MarTech Podcast, Benjamin Shapiro.

欢迎收听MarTech播客。

Welcome to the MarTech Podcast.

我是本杰明·夏皮罗，MarTech播客的执行制片人。

I'm Benjamin Shapiro, executive producer of the MarTech Podcast.

今天我们要带来一期特别节目，由Juan Mendoza主持，他正是《MarTech周刊》通讯的作者。

And today, we've got a special episode for you, which is going to be guest hosted by Juan Mendoza, the author of the MarTech weekly newsletter.

Juan曾经是MarTech顾问，现在转型做内容创作者，每周都写一篇关于MarTech行业的精彩通讯。

Juan is a recovering MarTech consultant turned creator who writes an amazing weekly newsletter about the MarTech industry.

我很高兴邀请他和几位朋友来分享他们的见解，献给我们的忠实听众们。

And I'm thrilled to invite him and some of his friends to take the mic and share their knowledge with you, our loyal MarTech Podcast listeners.

好了，接下来是这期由Juan Mendoza主持的MarTech播客特别节目。

Alright, here's a special episode of the MarTech Podcast guest hosted by Juan Mendoza, the author of the MarTech weekly newsletter.

大家好，各位营销人和MarTech爱好者。

Hello, marketeers and MarTechers.

我是Juan Mendoza，来自《MarTech周刊》通讯。

My name is Juan Mendoza from the MarTech weekly newsletter.

本周我们要聊的是：在在线隐私时代，如何做市场组合建模和现代媒体投放。

And this week, we are going to discuss market mix modeling and modern media buying in an age of online privacy.

今天和我一起的是罗希特·马赫斯瓦兰。

Joining me today is Rohit Maheswaran.

他是Lifesight公司的联合创始人兼首席产品官。

He is the co founder and chief product officer at Lifesight.

这家公司源于对日益严格的隐私法规和无Cookie时代挑战的深刻理解。

The company was born from a deep understanding of the challenges in the increasingly stringent privacy regulations and cookie less world.

Lifesight帮助营销人员打造有影响力的广告活动和个性化客户体验，弥补第三方数据缺失带来的空白，迎接以隐私为中心的无Cookie未来。

Lifesight empowers marketers to create impactful campaigns and personalized customer experiences, bridging gaps left by third party data for a privacy centric cookie less future.

今天，我和罗希特会聊聊营销组合模型——这个缩写词你可能经常听到，我们来谈谈它如何真正帮助扩大规模，优化媒体投放预算决策。

So today, Rohit and I, we are going to discuss how marketing mix modeling, we probably hear this acronym quite a bit floating around, how it actually unlocks scale and builds better budgeting decisioning when it comes to media buying and optimizing that media as well.

接下来是我们的第一场对话，嘉宾是Lifesight的罗希特·马赫斯瓦兰。

So here's our first conversation with Rohit Maheswaran from Lifesight.

他是联合创始人兼首席产品官。

He's the co founder and chief product officer.

罗希特，你最近怎么样？

Rohit, how are you doing?

我挺好的，老兄。

I'm good, mate.

你怎么样？

How are you?

我也不错。

I'm doing well.

谢谢。

Thanks.

我挺好的。

I'm doing well.

我们刚才没开录的时候还在聊，俩人都忙得不可开交，不过听上去Lifesight最近增长势头挺猛的。

We were just discussing before we hit record here about how busy we both are, but it sounds like Lifesight has been on a bit of a growth tear at the moment.

过去六个月怎么样？

How's the past six months been?

哦，特别棒。

Oh, it's been fantastic.

我觉得我们的新产品线已经开始起飞了。

I think our new product line is taking off.

过去我们主要在第三方营销效果衡量领域为人所知，但随着公司的发展，现在我们逐渐转向第一方数据的衡量方式，还引入了一些前沿的AI技术来优化营销效果分析。

We while we've been pretty much known in the third party marketing measurement space in the past, now we've kind of the evolution of the company has seen it move towards more first party based measurements, as well as introducing some of these more modern AI techniques for doing marketing measurements.

这太棒了。

That's awesome.

听到这个真让人高兴。

Well, it's great to hear.

现在在线营销的衡量领域变化特别多，很多都是因为无Cookie的未来越来越近，大家心里都有数，但我相信，等谷歌真正开始淘汰第三方Cookie之后，2024年会迎来一波大的变革。

I mean, right now there is a lot of movement in the measurement space in marketing online, and a lot of it is being forced by the cookie less future, which seems to be looming over us constantly, but I think a lot of change will happen in 2024 once Google starts deprecating third party cookies properly.

你能给我们简单解释一下吗？

But can you give us a bit of a definition?

什么是营销组合模型？

So what is marketing mix modeling?

你是怎么理解这个概念的？

How do you understand it?

那您觉得营销人员是怎么用它来实现更大规模的营销效果的呢？

And how do you see marketers using it to achieve that better scale in their marketing efforts?

是的。

Yeah.

营销组合模型其实已经存在很多年了，说真的，从六十年代人们开始决定广告预算该怎么花，它就一直在用。

So marketing mix modeling has been around for years, right, for decades, actually, since the sixties, ever since people were deciding where to spend their advertising dollar.

那时候，只有那些负担得起统计学家和数据科学家的大公司才用得上。

So ever since then, it was used, obviously, by some of the larger companies who could afford statisticians and data scientists and all of that.

但到了2010年代后期，基于追踪的测量方式兴起，彻底把营销组合模型给挤垮了——那时候真正的用户级追踪开始普及，归因分析进入了黄金时代。

But the advent of tracking based measurements completely killed it in the late two thousand tens, perhaps, I would say, when true tracking, individual level tracking, took off the golden era of essentially attribution.

从那以后，营销组合模型就慢慢失宠了。

That's when marketing mix modeling, I think, lost its popularity.

因为所有营销人员都能从点击或曝光归因工具里立刻看到效果。

Because everyone, all marketers could see the immediate result from a click based or view based attribution tool.

结果呢，这种即时反馈其实是一种虚假的错觉，让人误以为自己知道什么真正有效。

And then that ended up being like a fake delusion of sorts where this is what is working.

但实际上，营销组合模型才是真正能衡量线上线下任何媒体渠道影响力的方法，而不是只盯着数字媒体里的大平台，而是能准确评估这些渠道对关键绩效指标、线下销售的实际影响。

When in reality, marketing mix modeling is actually the true sense of being able to really measure the influence of any media channel online and offline, and not just gravitate towards the giants in the digital media space, and be able to then measure that influence on the KPI, right, or the offline or the sale.

这就是营销策略的回应。

So that's marketing responding.

当然，传统的分析方法现在已经升级了。

Obviously, the traditional methodologies have now since evolved.

如今有了基于云和人工智能的技术，能让营销组合模型的产出速度大大加快，即使是不懂技术、不懂数据分析的人，也能轻松生成同样的结果。

There are cloud based, AI based capabilities today that enable a lot faster output of these marketing mix models and is able to then give even a nontechnical or non marketing analytics person to deliver the same output.

正如你所说，营销组合模型已经存在很久了，但它一直在缓慢演进。

So as you say, marketing mix modeling has been around for a long time, but it's slowly evolved.

这有点像重回未来——对那些管理大量广告预算的营销人员来说，理解这些信号正变得越来越重要。

It's almost like we're going back to the future here a little bit because it's becoming increasingly important to marketers, particularly marketers that are managing a lot of media spend, to figure out those signals.

根据我的经验，它帮助我们不仅看清了渠道效果，比如资金来源和转化来源，还涵盖了天气、季节变化等整个营销组合中的其他因素。

From my experience, has played a role in understanding not just channel outcomes in terms of where the money's coming from, where the conversions are coming from, but also things like weather, seasonality, and other elements in the whole marketing mix.

它让你获得更全面的视角。

It gives you sort of a broader context.

但为什么是现在？

But why now?

为什么它成了营销工具中如此关键的一部分？

Why is becoming such an important part of the marketing toolkit?

确实如此。

Yeah.

本质上，这是因为自GDPR出台，以及iOS 14和各大浏览器（比如即将更新的Safari和Chrome）陆续限制了用户级追踪能力之后，一对一匹配变得越来越困难。

In essence, this is driven from the fact that ever since GDPR and technology changes such as iOS 14 and all the latest browser based limitations, right, Safari and Chrome that is coming up soon as well, has started limited the ability to do the one to one matching.

这不仅降低了数据信号的质量，很多时候甚至导致信号完全丢失。

And in essence, diminished the signal quality and also complete signal loss in many cases.

如果归因和基于追踪的方法无法准确执行，你的测量结果自然也就不可靠。

So if attribution can't be done or tracking based methodologies can't be done accurately, Your measurements outcome is just not accurate.

你看着Google Analytics里的所有报告，其实几乎全是错的。

So you're reading all your reports in Google Analytics, which is all almost completely wrong.

再加上各个平台自己的报告，给出的又是另一套说法。

And then you have the in platform reporting, which is telling you one thing.

然后你再看你的 Shopify 账户，又给出另一个说法。

And then you look at your Shopify account, it tells you another thing.

所以这方面完全是割裂的。

So it's completely disjointed in that sense.

这时候，营销组合模型就作为第二种方法派上用场了。

So that's where marketing mix modeling comes in as a secondary methodology.

在当今这个环境下，它能帮你补充决策能力，引导你通过多方验证，找出营销效果的真实情况。

And in this today's world, to supplement your decision making capabilities and try to guide you towards triangulating the actual truth of what's happening in the world in terms of your marketing outcomes.

就像你提到的，营销组合模型当然会用到各种数据。

So like you mentioned, marketing mix modeling, of course, takes in all kinds of data.

基本上，你可以把你的投放支出数据喂给它。

Essentially, can feed it your spend data.

还可以输入转化数据，以及其他各种相关因素。

You can feed it your conversions data, and then all other factors as well.

对吧？

Right?

就像你提到的，竞争对手有没有搞促销，有没有什么活动、体育赛事，或者像黑色星期五和网络星期一这样的节日大促，这些都会考虑进去。

So like you mentioned, whether your competitors have done some promotions, events that are happening, sports events or holiday sales events like BFCM and all of that.

它还会把所有这些因素，连同一些自然流量数据一起纳入分析。

It takes into all of that into account as well as some organic data.

对吧？

Right?

你的邮件通讯、TikTok上的视频播放量，不管是什么数据，都会全部输入模型，系统能据此综合分析，优化你的预算分配。

Your newsletters, your video views on TikTok, whatever it might be, and puts it all into the model and is able to then churn that out, take that all into account and allocate your budgets in the most optimized fashion.

所以它能捕捉到这些信号。

So has this ability to get that signal.

对吧？

Right?

那每个渠道到底贡献了多少？

So what is each channel contributing?

这其实挺有意思的，也能帮我们的听众更清楚地理解它到底是怎么运作的。

Which is kind of interesting because the way maybe you can help our audience understand how it actually works.

这是个黑箱吗？

Is it a black box?

你丢一堆数据进去，它就给你出结果。

You feed it a bunch of data and it gives you some results.

里面有加权吗？

Is there weightings?

用模型的时候，会不会用到不同的方法？

Is there different types of methodologies within using an model?

你能给我们具体讲讲吗？比如在Lifesight软件里，这个过程到底是怎么运作的？

Could you perhaps step us through, like, what does that look like on the ground perhaps with even the Lifesight software?

是的，当然可以。

Yeah, absolutely.

我们的模型就是一个预测模型。

So our model is a model, right?

它是个预测模型。

It's a prediction model.

这是一个统计模型。

It's a statistical model.

本质上，它的结构已经预先设定好了。

So in essence, it's the structure is already there.

它的基础是能理解所有不同的广告渠道。

The foundation is that it understands all the different ad channels.

它知道数据是如何影响具体效果的。

It understands how the data is impacting specific outcomes.

对吧？

Right?

它在广告领域抓住了两个核心概念：广告存续和饱和度。

It understands two core things in advertising, which is ad stock and saturation.

广告存续指的是广告对消费者的影响能持续多久，而饱和度则是指一个渠道在回报开始递减之前，最多能投入多少。

Ad stock being how long is the effect of an ad on a person, right, on a consumer, as well as then saturation as to how far can you go before the channel gives you diminishing returns.

所有这些因素都已经被整合到模型里了。

So all of these things are built into it.

本质上，你是在向系统输入过去所有营销活动的历史聚合数据。

And in essence, what you're doing is feeding it historical aggregated data of all your activities that you've done in the past.

这包括过去两年你在各个渠道的投放支出，以及对应那天或那周的销售数据。

This includes up to two years of your spend date across all the channels, as well as then the corresponding sales data for that specific day or for that specific week.

这也是一个很大的优势——我们根本不需要个人层面的数据。

So that's another great thing about is we don't need individual level data.

这种方式在隐私保护上完全合规，从任何角度来看都毫无风险。

This kind of completely privacy safe and privacy compliant in every angle that you can think about.

当你把数据输入进去后，所有外部数据因素——这些系统已经内置好了，对吧？

And then once you feed the data in all the external data factors, which is already built in, right?

这一点由Lifesight来实现，我们已经把所有这些外部数据因素都整合进去了。

So that's something that Lifesight provides where we already build in all these other external data factors into it.

然后你开始训练模型。

And then you train the model.

现在有了很多开源模型，这个过程变得简单多了。

So this is now made a lot easier with a lot of open source models.

最流行的是Meta推出的Robin。

The most popular one from Meta called Robin.

还有Google的轻量级方案。

You then have Google Lightweight.

Uber的Orbit，以及其他几个选项。

You have Orbit from Uber and few others.

总的来说，Lifesight把自己定位为一个数据平台。

And all of this essentially, Lifesight sees ourselves as a data platform.

对吧？

Right?

我们本身并不是一个机器学习产品，而是让你能在接入我们平台的数据上，自由使用任何机器学习模型。

We are not a machine learning product per se, but we are enabling you to apply any machine learning model on top of the data that you ingest into our platform.

这样一来，非技术背景的市场人员也能直接上手使用这些工具。

So in essence, making it super easy for nontechnical marketers to leverage some of these tools out of the box.

数据导入后，选好模型，一键生成就行。

Once you feed that in, you pick your model and then generate it.

当然，这里面有很多细节。

Obviously, there are a lot of nuances.

有经验的建模人员、统计学家、数据科学家或者营销分析专家都懂这些细节。

An experienced modeler or statistician or data scientist or marketing analytics person would know all the nuances.

他们知道怎么调整模型。

They know how to adjust the model.

这样你就能对模型进行校准。

So you can actually calibrate that model.

你可以修改一些叫超参数的东西。

You can change something called your hyperparameters.

你可以说：不对。

You can say, no.

其实我知道，Instagram 给我的广告展示时间更长，或者类似的情况。

Actually, I know that Instagram gives me a longer ad stop or something like that.

而且这些设置对每个品牌都是定制的。

And it's all custom for your brand as well.

所以核心就是为你品牌的特性、所在行业和国家量身打造一个专属方案。

So the whole idea is to build something custom to your brand and your industry and your country, for example.

然后只需要二十到三十分钟，模型就能完成训练。

And then from there, it is just trained in within twenty to thirty minutes.

训练用的是你提供的数据。

It's trained on.

我们用的是谷歌云平台。

We use the cloud, so Google Cloud.

整个过程非常快。

And that's done quite quickly.

接着你就可以查看训练结果了。

And then you get to look at the results.

这些结果会告诉你模型预测的准确率或置信度。

And the results will essentially tell you a prediction accuracy or a confidence level of what the model outputted.

根据这些结果，我们就能判断是否采用这个模型，它是否贴近你从归因数据中看到的真实情况，以及你过往的其他经验和认知。

And from there, we essentially are able to then decide whether you want to use that, whether that's close to your truth of what you're seeing in attribution as well, and perhaps all your other knowledge that you have and experience that you have.

从这里开始，你可以和我们的顾问合作，判断这个模型是否适合你。

And then from there, obviously, you can work with our consultants and be able to then decide whether this model is good for you.

有了这个基础，接下来就是构建各种场景，比如你的媒体投放方案。

Once you have that, then it's all about building up scenarios, your media plan scenarios, essentially.

你可以这样问：

And you can say, okay.

如果我把预算从上周的十万增加到十二万，能带来多少额外的销量？

If I increase my budget from a 100,000 last week to a 120,000, what incremental sales is that gonna give me?

正是这种在云端或软件环境中使用模型的灵活性，才赋予了你这样的能力。

That's where this flexibility of using this model in a cloud or software based environment gives you that power.

对吧？

Right?

你不用花一百万美元买尼尔森的一个模型，然后一整年都被绑死。

You're not paying Nielsen a million dollars for one model and then stuck with it for the whole year.

你可以灵活调整，快速做出改变。

You're kind of being able to adapt and move quite quickly.

我现在看到市场上，很多人正在放弃以往常见的最后点击归因或多点归因，比如谷歌分析那种方式，转而结合地理位置数据、对照组测试和增量分析来评估效果。

What I'm seeing in the market right now is that a lot of folks that are shifting away from say last click attribution or a multi touch attribution, which was pretty common and prevalent saying Google Analytics, etcetera, They're shifting away from that and they're kind of using a combination of and geolocation or holdout groups and incremental testing.

你在营销技术与分析工具栈里，看到的是什么样的情况？

How are you seeing sitting kind of in the MarTech analytics stack?

现在是只用一个工具吗？

Is it kind of one tool?

是不是正越来越成为权威的数据来源？

Is it becoming the source of truth increasingly?

你觉得目前大多数公司是怎么搭配使用这些方法的？

What's that mix look like, do you think, right now for a lot of companies?

我觉得任何有经验的市场科学家或市场分析师，都会把所有这些工具备齐。

I think any experienced marketing scientist or marketing analyst person would have all these tools in their tool bag.

我们通常第一步是先从客户出发。

We always start with in the sense that we take the customers.

第一次和客户沟通时，我们会说：咱们先重新分配一下你的预算。

When we speak to the customer for the first time, we say, let's reallocate your budgets first.

所以别搞什么大刀阔斧的改动。

So let's not do any big mega change.

别去追踪任何东西。

Let's not track anything.

就把你的数据给我就行。

Just give me your data.

我们先重新分配你的预算。

Let's reallocate your budgets.

我们第一步就这么做。

We start with that.

等他们有了这些，再开始跑广告。

Once they have that, then they run the campaigns.

当然，这背后要有扎实的数据流动做支撑。

Obviously, this is then backed with the proper movement of data.

我们还有一个数据激活工具，能把最优质的数据推送到所有广告平台，帮助他们更准确地追踪归因和转化效果。

We have a data activation tool as well, which essentially sends the highest quality data to all the ad platforms so that they track your attribution better and your conversions better.

第二步就是做归因和每日优化。

And the second step is then to do attributions and daily optimizations.

这一步离不开归因。

You need attribution for that.

你得知道该停掉哪些广告活动。

You need to be able to know which campaigns to switch off.

也要清楚该加大投入哪些活动。

You need to be able to see which ones to double down on.

归因就是用来做这种每日优化的工具。

So that's where attribution comes into play, which is daily optimization tool.

最后，你还需要验证这一切。

And then finally, you need to prove all of that.

我已经做了这么多广告投放。

Now I've done all this advertising.

现在我想用实验这个黄金标准，来证明和验证所有这些效果。

I wanna now prove and validate that all of those with the gold standard, which is experiments.

这时候你可以做受众分组测试或者地区测试，把整个群体放在一起，看看这些广告到底为你带来了多少实际增量效果。

That's when you can either do an audience split test or a geo test and do a whole lot group and see what exactly did those campaigns bring in terms of incrementality to you.

然后这就形成了一个反馈循环：用实验得出的提升数据来重新校准你的模型，因为这些数据更贴近真实情况。

So then it goes back into a feedback loop of then measuring, recalibrating your model with the lift data than you get from the experiments because that's closer to the ground truth of what's actually happening.

所以，没错，这些数据会进入反馈循环，接着你又重新分配预算了。

So, yeah, then that goes into the feedback loop, and then you're back again reallocating your budgets.

这更贴近真实情况，然后再回到归因分析，接着又回到实验环节。

That is closer to the ground truth, and then coming back to attribution and then going back to experiments again.

这就是我们看待这个组合的方式。

So that's how we see the mix.

这确实像是我们从新方法中走出来后，所谓的最佳实践。

That does seem to be the quote unquote best practice as we emerge out of new methodologies.

这几乎是把老方法重新翻出来，在如今这个在线隐私的新环境下赋予它们新的生命。

Well, going back to the future almost of finding the old methodologies and making them new again in this sort of new world of online privacy.

看起来它确实分成了两部分：一方面是预测，以及对媒体整体表现的宏观把握；另一方面则是具体测试，搞清楚归因到底在起什么作用。

It does seem to split between say forecasting and a sort of a sense of the holistic sense of your performance in terms of media, and then the, like the specific testing you do to figure out what's that attribution doing.

所以整体上是一个由宽到窄的过程。

So there's sort of a broad and then a narrowing in.

我觉得宏观层面是这样，而具体到缩小范围，就是地理隔离测试和各种增量测试方法。

I think is the broad picture, and then the narrowing is that geo holdout, those different types of incremental testing.

这些方法如何协同工作其实挺有意思的，但它们并不是同一种方法。

So it's kind of interesting how those work together, but they're not the same methodology.

它们在分析方式和所用数据上都截然不同。

They're extremely different in the analysis and even the data they use as well.

我想问问你，你在这个行业里待了挺久了吧？

But I wanted to ask you, I mean, you've seen this industry for a while.

你打造Lifesight也已经有不少年头了。

You've been building Lifesight for quite a time.

在使用营销组合模型时，你遇到过哪些主要挑战？

What have been some of the challenges around using marketing mix modeling?

在实际操作中，营销人员最容易在哪些地方出错？你是不是经常看到一些反复出现的误区？

What have been some of the things where marketers really struggle to get this right, or you kind of see an anti pattern come up time and time again?

是的，我觉得最大的挑战是让现代营销人接受这一点。

Yeah, I think the biggest thing is convincing the modern marketer.

我说的现代营销人，是指那些一辈子只用归因模型的人，他们并不在大企业里，没有一群营销数据科学家支持。

When I say modern marketer, the marketer has only been using attribution all their lives, And they're not part of, like, a large enterprise who has all of those marketing scientists and all of that.

对吧？

Right?

所以最大的难题是让他们明白，把这些方法结合起来用有多重要，还要让他们相信，这种概率模型才是正确的方向。

So the biggest challenge is educating them about the values of using it in combination and also convincing them that this probabilistic model of sorts is the way to go.

对吧？

Right?

因为大家都太熟悉那种确定性的追踪方法了，一对一匹配，清清楚楚。

Because everyone is so familiar with that certainty of deterministic tracking based methodology, one to one match.

对吧？

Right?

我点了广告，然后就看到成交了。

I clicked it, I saw the sale.

所以这确实是真实的。

Therefore, it is real.

这种思维模式得改一改。

That mindset needs to kinda change.

这一直是个大难题。

So that has been a big challenge.

现在的问题还包括怎么拿到数据。

Now the challenge is also getting access to the data.

对吧？

Right?

没几个人能把历史数据整理得井井有条，还随时能调用。

Not many people have historical data arranged in a nice way and be able to access it.

所以我们模型跑起来之前，得花大量时间做数据准备。

So we spend a lot of time on data prep before running the model.

当然，如果数据本身有问题，那就得请顾问出手了——得有专家懂怎么整理数据，或者调整模型，才能输出一个能用的结果。

And, of course, then you have bad data, and that's when you need a consultant to step in, an expert who knows how to arrange the data or tweak the model in order to be able to output a usable model.

所以这就是我们遇到的一些挑战。

So that's where we see some challenges.

但市场正在发生变化。

But the market is changing.

你可能也注意到了，现在各种工具层出不穷。

As you probably know, it's being flooded now with tools.

所以我们很高兴看到，这种教育正在大规模展开，营销人员也愿意尝试这些工具，而不再仅仅依赖谷歌分析四和平台自带的报告。

So, yeah, we are happy to see that that education is happening in a large scale and that marketers are open to testing out the tool beyond just relying on GA four and in platform reporting.

明白了。

Okay.

那我换个角度问你一下，假设你接了一个新客户。

So I'm gonna flip this a little bit, and I'm gonna ask you, okay, you've got a new customer.

他们第一次使用你们的服务。

They're getting onboarded with for the first time.

咱们给他们取个名字吧。

Let's give them a name.

好的。

Okay.

就叫他乔伊吧，随便取个名字。

The name is, I don't know, Joey.

乔伊来找你了。

Joey's come to you.

他之前在X和Y Zed这样的企业工作，现在想做混合媒体建模。

He's worked at x and y zed enterprise company, and they they wanna do mixed media modeling.

他们想摆脱GA4和最后一击归因的模式。

They wanna get off GA four and and last look aperture distribution.

好的。

Okay.

很好。

Great.

你走进他们的公司，心里想：行吧。

You go into their business and go, okay.

我们需要这些数据。

We need this data.

要有效运行模型，理想的数据显示应该是怎样的？

What would be the ideal data setup to effectively run modeling?

你会对乔伊说什么？

What would you say to Joey?

你需要什么？

What do you need?

通常我们至少需要一年以上的数据。

So we need typically at least a year's worth of data.

如果你的公司运营时间还没到一年，那可能还太年轻了。

So if you are a business that has been running for shorter than that, it's just you're probably too young.

起步阶段规模太小，企业得达到一定体量才行。

You start probably too small, so you have to be a certain size.

我建议每月的营销预算至少要有五万美元。

At least, I would say, 50,000 USD in marketing spend a month.

这是我们观察到的最理想的临界点，超过三个渠道时才真正值得做。

That's the sweet spot that we're seeing where it actually makes sense more than three channels.

在考虑使用营销组合模型之前，你必须先满足这些条件。

So you have to fulfill this criteria before you marketing mix modeling is suitable.

否则，根本没实际意义。

Otherwise, it's just not practical.

对吧？

Right?

你用现有的归因工具就已经足够了。

You can do just fine with your attribution tools.

然后，我们其实是在说，你得把数据接入我们的平台。

And then essentially, we are saying you've got to either connect to your data into the platform.

把所有我们支持对接的数字渠道都连上。

So connect all the digital channels that we have connectors for.

只要登录并授权，让我们能访问你的数据就行。

So just log in and authorize and give permission to access the data.

另外，如果你的数据存在线下系统里，也可以导出到谷歌表格这类工具，再把表格连到我们的平台上。

Alternatively, if you have data sitting in an offline store, you can then load that into a Google Sheet or something like that and then connect the Google Sheet to our platform as well.

然后数据就会开始自动流入。

Then the data starts flowing in.

我们会看到你所谓的数据结构，也就是模型实际接收的、整理得井井有条的数据格式。

We see what you would call an schema, what the model would be fed essentially in a nicely arranged format.

接着我们会把这个结构映射到模型上，之后就可以直接用了。

And then we map that schema to the model, and then you're good to go.

如果预测结果很高，准确率也很高，我们就会深入分析，看看它到底在说什么。

So if the prediction is high and accuracy is high, then we would really deep dive into it and see what it's saying.

而且所有结果都得合乎逻辑。

And it all has to make sense as well.

如果得出的结论完全违背常理，那我们就知道这个模型不对。

If it just says something completely non intuitive, then we know it's not the right model.

所以实际上，系统会生成很多模型，你可以从中挑出最适合你的那一个。

So, yeah, it can spit up quite a lot of models actually, and you can pick the right one for you in essence.

这就是你开始的方式。

So that's how you would get started.

然后它会生成一个你所说的预算表格。

And then it would output what you call essentially a budget worksheet.

它会明确告诉你，下周或下个月每个渠道该花多少钱。

So it would say exactly how much to spend on each channel for the following week or following month.

接着你就可以把这些数据直接导入Facebook、Google等平台。

And then you would then feed that data into Facebook directly or into Google and so on and so forth.

然后在接下来的几周里跟踪效果。

And then monitor the outcomes in the following weeks.

是的。

So yeah.

这就形成了一个反馈循环。

So then it's a feedback loop.

对吧？

Right?

你把数据反馈回来。

You come back with the data.

系统会自动更新，把这些数据重新输入模型，必要时还会结合提升数据进行调整和校准。

So that is automatically refreshed and feeds back into the model, and then the model is updated again and recalibrated with lift data if needed.

这真是太有意思了。

Well, that is fascinating.

那我们就说到这里。

We'll end it on that note.

Rohit 设想的这种模型架构，适用于规模化应用，能帮你更清楚地了解该把营销预算投到哪里。

Rohit's ideal setup for an model and implementing that for scale and to get more insight into where you should be putting your marketing budgets.

今天的 MarTech 播客节目就到这里。

That wraps up our episode today for the MarTech Podcast.

感谢 Lifesight 的 Rohit Maheswaran 做客我们的节目。

Thank you to Rohit Maheswaran from Lifesight for joining us.

在明天发布的第二部分中，Rohit 和我将更深入地探讨在隐私优先的时代如何优化媒体购买。

In part two of this interview, in which we'll publish tomorrow, Rohit and I are going to do a little bit more of a deep dive into optimizing media buying in the privacy first era.

如果你等不及下一期节目，想更多了解Rohit，可以在节目说明里找到他的LinkedIn链接。

So if you can't wait until our next episode, would like to learn more about Rohit, you can go find a link to his LinkedIn profile in our show notes.

你也可以在Twitter上联系他，他的用户名是Rohit Mah，或者访问他的公司网站lifesight.io。

You can also contact him on Twitter where his handle is at Rohit Mah, or visit his company website at lifesight.io.

Rohit，感谢你参与这次对话。

Rohit, thanks for joining me.

谢谢，Juan。

Thanks, Juan.

和你聊天很愉快。

Nice talking to you.

好的。

Okay.

这期《MarTech播客》就到这里了。

That wraps up this episode of the MarTech Podcast.

感谢我们的特邀主持人Juan Mendoza，他是《MarTech周刊》通讯的作者。

Thanks to our guest host, Juan Mendoza, the author of the MarTech weekly newsletter.

如果你想联系胡安，可以在我们的节目笔记里找到他的领英链接，也可以在推特上联系他。

If you'd like to get in touch with Juan, you could find a link to his LinkedIn profile in our show notes or you can contact him on Twitter.

他的用户名是Juan Mendoza，不过拼法有点特别，叫crazy pants。

His handle is Juan Mendoza, but it's spelled crazy pants.

拼法是j u four n m e n d zero z four，不过更简单的方式是直接访问他的公司网站，themartechweekly.com。

It's j u four n m e n d zero z four, or it's a little easier to just visit his company's website, which is themartechweekly.com.

我还想再给大家推荐一个我们节目笔记里的链接。

Just one more link in our show notes I'd like to tell you about.

如果你听播客的时候没来得及做笔记，可以去martechpod.com，那里有我们每期节目的摘要，还有嘉宾的联系方式。

If you didn't have a chance to take notes while you were listening to this podcast, head over to martechpod.com where we have summaries of all of our episodes and contact information for our guests.

你还可以订阅我们的周刊，甚至给我们投稿话题或提问营销问题，我们会选一些在节目里直接回答。

You can also subscribe to our weekly newsletters, and you can even send us your topic suggestions or your marketing questions, which we'll answer live on our show.

当然，你也可以通过社交媒体联系我们。

Of course, you can always reach out on social media.

我们的账号是martechpod，也就是m a r t e c h p o d，在领英、推特、Instagram和Facebook上都能找到，你也可以直接给我发消息。

Our handle is martechpod, m a r t e c h p o d, on LinkedIn, Twitter, Instagram, and Facebook, or you can contact me directly.

我的用户名是 benjshap，拼写是 b e n j s h a p。

My handle is benjshap, b e n j s h a p.

如果你还没订阅，又想每天在播客里获取营销和科技方面的干货，今年我们每天都会更新一集。

And if you haven't subscribed yet and you want a daily stream of marketing and technology knowledge in your podcast feed, we're gonna publish an episode every day this year.

赶紧在你的播客应用里点个订阅，明天早上我们再回来陪你。

So So hit the subscribe button in your podcast app, and we'll be back in your feed tomorrow morning.

好了。

Alright.

今天就到这里。

That's it for today.

不过在那之前，我的建议是，专心把客户的服务做好。

But until next time, my advice is to just focus on keeping your customers happy.

感谢收听《MarTech Podcast》，由 I hear everything 制作。

Thanks for listening to the MarTech Podcast, an I hear everything production.

想为你的品牌打造或扩大这样的播客吗？

Looking to launch or scale a podcast like this one for your brand?

那就去 iheareverything.com 看看吧。

Then visit iheareverything.com.