工具是用来解决问题的：Austin Hay 的 MarTech 实操指南

Hay defines marketing technology as a cross-functional discipline at the crossroads of product, growth, engineering and marketing. He compares it to a product manager whose focus is a system, covering both third-party tools and first-party solutions the company builds itself. The role has a people-and-process side and a system-and-platform side.

What the job looks like depends on the size of the company. At Airbnb, at roughly 800 people, there was a MarTech lead with product and engineering resources, while at a small startup like Runway there was no such function and a few people simply set up and used the tools. Many MarTech professionals started in growth or user acquisition roles because they had to use the tools to get the job done.

Hay says the critical mass is around 100 to 150 people. Up to about 30 or 40 people everyone pitches in on the customer data platform or analytics tool, but beyond that someone must know how data flows, what the schema is, and how contracts are managed. He compares this to IT: a village approach to single sign-on would be dangerous. The function can then land in product ops, IT, marketing or revenue operations, and in B2C it often sits within growth, whereas in B2B it often sits within rev ops.

Hay describes a common story: a growth team hires an engineer who stands up all the tools, and after two years the systems are too complex for one person. The MarTech work then includes setting up new tools and building on top of them, for example custom tooling behind Segment or Amplitude. Many people think MarTech is only third-party tools, but he says it is architecting and building on top of them.

His slogan is build and buy: purchase the tool to get 90% of the way, then build the valuable 10% yourself. That architecting decision typically falls to the MarTech person. He also loves the contract side, since early companies sign any deal to survive, but later they must think about liability exposure and what a cheap rate at 500 monthly tracked users costs at a million. He draws on his time at mParticle, a CDP vendor, where cost structures were designed so revenue grows with the customer.

Lenny summarizes that a sign you need a MarTech person is accumulating tools and sensing you could connect data better. Hay adds that pain drives hiring more than efficiency does: teams cannot change email or data tools because nobody understands how they were configured. That is especially true of email systems like Braze, which drive repeat usage.

Hay frames the org question along two axes: B2C versus B2B, and centralized versus decentralized. In B2C he thinks it is simple: marketing technology serves the growth team, so it can sit next to or under the head of growth, provided the person is a strong technical architect. A platform PM is a variation. He generally sees the decentralized model, with an ops person in every function, working poorly because more operational people create more systems.

B2B is messier, especially B2B2C businesses like Notion, which sell to both users and businesses. At Ramp the function has moved between rev ops and the CMO's organization and is now going back into revenue operations, depending on who the customer is and where resources are. He says there is no right answer and that the leader's profile matters: an ops-minded leader fits in ops, and a technical architect may belong in product or engineering.

Hay says every marketing technologist is at some level an individual contributor, because you must understand third-party tools by doing the work. More important than the number of people managed is how well a candidate manages upward, laterally and downward, since they depend on rev ops, product and data. He likens the role to a quarterback who calls plays across departments, and says it is a game of persuasion and salesmanship, often about slow changes such as retiring a duplicate CDP or attribution tool that might cost a million dollars.

Hay says half the job is administrative but high leverage at big-company scale: privacy requests, contracts, permissions and access. He warns that giving someone edit access in HubSpot can end with a test email sent to a million people. He aims to automate these approvals based on role and tenure. The other half is designing the vision for systems over one to two years, building financial models for contracts and costs, and persuading people to spend engineering time once a problem is big enough.

His advice for growth managers is to remember the company will outlast you, so ask what happens a year from now if nothing changes. As an example, paying about $2,000 and two days for SSO is a bargain compared with a security incident or hiring an IT person later. He acknowledges the trade-off with moving fast early on.

He distinguishes marketing technology, usually engineering-based, from marketing operations, which is semi-technical work such as campaigns, email sends and SQL queries, common in B2B. Hay himself is a former civil engineer who taught himself to program. At Ramp his job is to support the user acquisition team, for example by sending opportunity data back to ad networks as a synthetic event so campaigns optimize on real opportunities rather than button clicks.

On goals, his are tied to the teams he serves, such as growth and customer acquisition cost, plus a cost-efficiency goal he sets himself. The dream is for the business to grow while total tooling cost per person falls. Other goals are discrete projects, like moving off a small email tool to Braze or Marketo without losing money.

Hay says B2C stacks were largely settled between 2017 and 2020. Around 2016 to 2017, Segment and the rise of the customer data platform created a commoditized stack: integrate one SDK, collect users and events, and send data to ad networks, email tools and analytics. Then warehousing became cheap enough around 2020 that a company could model its data in something like Snowflake without a vast data team.

That led to reverse ETLs, which move data from the warehouse to tools, and to companies building their own CDP. Hay mentions a financial trading platform with plenty of warehouse data but old batch architecture that needed a reverse ETL to activate it. If you lack engineering resources and want simplicity, buy a CDP and connect your tools. If you have an advanced engineering culture with Snowflake and dbt, use a reverse ETL.

Reverse ETL is a capability found in CDPs such as Segment, mParticle and RudderStack, and in standalone products like Census and Hightouch. Hay is an investor in Hightouch and says Ramp uses it. The real risk is having several ways to move data, for example directly from Amplitude or from the warehouse, without a methodology for choosing, so he advises designing a clear rule for when to use each.

Hay recalls designing his first B2B system on a whiteboard in 2014 at Branch with COO Mike Molinet: an ingestion point for leads, Salesforce, an outbound tool, an enrichment tool and assorted other pieces. He says B2B still looks like that, with Salesforce at the center and more advanced tools such as Gong around it.

The big change of the last two or three years is B2B2C, which puts top-of-funnel user acquisition next to a CRM. B2C tools think in two objects, users and events, whereas B2B needs to tie users to companies. The hardest example is having both HubSpot and Salesforce, as he saw at Notion and Ramp, where neither is a CDP and mapping data between them determines how much pain you are in.

He says there is no good solution, only options: collect everything in Amplitude and merge into Salesforce, or route through Segment and HubSpot and then Salesforce. Each added step makes the system more complicated than one person can manage, so there is always a trade-off between complexity and resources.

Hay explains that multi-touch attribution (MTA) and mixed media modeling (MMM) are two ways to decide how to allocate campaign spend, and both depend on the user and event data you collect. Businesses usually start with first or last touch, then split credit 50/50, and then someone who feels short-changed demands MTA. He says there are six or seven things you can do to future-proof yourself from needing either choice.

For a website, he recommends capturing the landing URL, the referrer, every UTM and the parameters ad networks add, such as Google's click ID. These must be stored on the device as first and last values, with the last value overwritten at each visit, and written both as user attributes and on page view events. The warehouse team can then coalesce across events to recover first, middle and last touches.

Lenny suggests the key is a warehouse and a taxonomy. Hay agrees, noting that third-party B2C tools mostly allow only a user object and an event object, with Snowplow the one CDP he knows that lets you control the schema. His point is that the MarTech job is to look one to two years ahead and preserve future options in the least invasive way.

Asked about emerging channels, Hay points to Threads, where the question is whether Meta builds a separate ad API, and notes that Facebook has a conflict of interest in reporting results, which is why attribution vendors like Branch and AppsFlyer exist. He also notes that Reddit is opening up its conversions API and seeing more investment.

The deeper trend, he says, is that from 2010 to 2020 deterministic matching made it easy to know who installed an app, but today few users opt in to sharing their device ID. Ad networks are getting more sophisticated just as marketers lose precision, so much of his work is modeling about 30% of the population and extrapolating. Probabilistic matching and attribution is a skill he thinks more marketers should learn.

On the web, browsers strip URL parameters so more paid visits look organic, and cookie blockers stop tools like Segment from tying anonymous activity to a user once they log in. Many are turning to MMM without knowing when it is useful, and Hay says most businesses are not ready and need better probabilistic modeling. Tactics like geo-based billboard tests help, but he says there is no silver bullet.

The first thing Hay looks for is intellectual curiosity, because the tool landscape is far too large for anyone to be an expert in everything. The second is scrappy engineering ability: knowing JavaScript or Python and being able to read API documentation and make a request. He says people can teach themselves, as he did, and that around six months of study is enough for a marketing ops or user acquisition person to become dangerous.

His favorite question asks how the candidate prepared for the interview, an idea he credits to his wife and an a16z partner. A richer, more complex answer signals a richer, more complex candidate and a systems thinker. He also asks what they would do if they joined tomorrow and had to write a report on what to change by Friday, which reveals tool bias, since he prefers people who are tool-agnostic and start from the problem.

On red flags, Hay says that when he is the candidate, he asks for financials and distrusts companies that will not share them with a director-level hire. As a hiring manager he focuses more on false flags, such as resume gaps and school names, that can lead to bad shortcuts. He gives an example of a candidate with a two-year gap who used it to study philosophy or poetry and teach himself to program.

His first framework is that tools are just meant to solve problems, which he tells every hire. You do not have to buy a tool, or a specific one, to solve a problem. He pairs it with problem, people, system: when a request arrives, first ask what the problem is, then who is involved, and only then design the system. For example, a sales manager who wants easier onboarding may need approval from the CRO, so the solution may not be a new permission in a tool.

His third is build and buy rather than build versus buy, since the narrow choice creates a fight. He recounts a company considering building its own A/B testing tool and a similar debate at Ramp, where the engagement used a financial model to show that buying a third-party tool and building around it would make more money and move faster. He argues a vendor will accommodate a customer that has invested in building on its product.

Finally, Hay describes thinking gray, from Steven B. Sample's book The Contrarian's Guide to Leadership: delay decisions as long as you can before you have to make them. In systems work, he says, you can often let a problem sit rather than react to pressure. The same grace applies to judging people, since you have space to decide after meeting them once or twice.

In the lightning round Hay recommends The Contrarian's Guide to Leadership and Warren Bennis's The Art and Adventure of Leadership, and says his favorite interview question asks about the hardest thing a candidate has overcome in the last year. He also mentions the scheduling tool cal.com as a recent favorite. His life motto concerns the power of appreciation and understanding the challenges people have faced.

His promised golden stack for a B2C business is Amplitude as the CDP and product analytics, Customer.io for email with a possible upgrade to Braze, Snowflake as the warehouse, Hightouch for reverse ETL, and AppsFlyer for mobile attribution, or Branch. For B2B it is similar, with Branch preferred for web-only attribution, data connected to Salesforce, Hightouch, and Customer.io before moving to Braze rather than using HubSpot for as long as possible.