Companies usually need to improve data access and analytics before machine learning can deliver value in production.
2
A strong data warehouse helps teams answer questions faster, discover useful use cases, and build support for further data investment.
3
Mark's TRIBE framework is talk, requirements, iterate, build, and evangelize, with stakeholder involvement continuing after the product ships.
Summary
Mark Freeman explains why companies cannot jump directly from hiring data scientists to running machine learning in production. Data maturity develops through stages, beginning with domain knowledge and limited access to raw data, then moving through databases, data warehouses, analytics, and simple machine learning before production ML systems. For startups, infrastructure competes with product work, so Mark ties data projects to business outcomes such as reducing churn and expanding customer accounts. He describes a project at Humu that made data easier for customer success teams to use. That access helped those teams ask better questions and gave Mark more concrete ML opportunities to investigate. His TRIBE framework, talk, requirements, iterate, build, and evangelize, gives teams a way to involve stakeholders throughout the work. Mark also explains that senior data science work includes substantial stakeholder communication and strategic planning, rather than continuous coding.
LinkedIn works best when people choose the role they want to play
Mark says people do not need to post every day to use LinkedIn well. They can comment on useful posts, share their expertise, follow the right accounts, or create content themselves. He posts on weekdays because he enjoys it, but says forced content becomes obvious. His own posts act as a bridge between what he is learning and people at different stages of their careers. Teaching a concept also makes him process it more carefully and practise explaining technical work to other people.
After being laid off during the pandemic, Mark chose to publish content instead of relying only on applications. He describes the process with the sales-funnel terms awareness, interest, desire, and action. His posts made employers aware of his skills, conversations and phone screens created interest, interviews tested whether he could explain his experience, and an offer completed the process. He now sees the same approach helping Humu recruit, because people apply after seeing his LinkedIn content and job posts.
Data maturity develops through infrastructure and use
Mark describes a progression from executive gut feeling and domain knowledge, through raw data that only a few people can use, to databases and then data warehouses built for analytics. With a warehouse, teams can create SQL models, data marts, dashboards, and more frequent answers. Broader data access creates demand for more analysis. That foundation makes predictive work and early ML projects possible. Production ML comes later, after a company has learned why the infrastructure matters and can justify the cost of data and ML specialists.
Big technology companies are poor maturity models for most startups
Mark says companies such as Airbnb, Google, and Facebook operate at a scale that most organisations do not share. Their engineering practices can show what is emerging, but copying their end state is the wrong target. He points to the New York Times as a more useful example because it is a traditional company that has been scaling its technology work. Smaller organisations can provide examples that fit a startup's constraints more closely.
A data warehouse increases the organisation's ability to find useful problems
Mark calls a strong data warehouse and broad data access the foundation of data maturity. At Humu, data science had become a report-churning team because other groups could not answer their own questions. Mark interviewed teams by asking what data they used, what worked, and what did not. He worked with engineering to put missing data in the warehouse, created models and data marts, and exposed the results through dashboards. Customer-facing employees could then ask better questions and surface problems that ML might solve.
Business value gives infrastructure work a reason to continue
Mark gained support for data work by connecting it to customer success, a team responsible for preventing churn and expanding deals. Data could automate easier parts of the work while leaving customer-facing decisions with the team. Once customer success relied on the data, improvements to its timeliness and coverage became easier to justify. Mark says a team that drives revenue can provide a useful internal point of support for requesting more data infrastructure.
TRIBE keeps stakeholders involved before and after the build
Mark's TRIBE framework means talk, requirements, iterate, build, and evangelize. He starts by speaking with users about their pain points, then defines what needs to be built. He returns with an interpretation of the problem and iterates until stakeholders agree that the proposed solution addresses it. After building, he explains the pain point, the product, and how people can use it. At Humu, a Slack channel called curated data gave customer success one place to report successes and problems, creating a record for future prioritisation.
Senior data science includes more relationship work than coding
Mark says he once expected data science to mean analysing data and writing code all day. At a senior level, he spends about 70 percent of his time talking with people, although the balance changes by month. He works through priorities, relationships, and pain points before coding begins. He credits Stephanie Tigner with planting ideas with stakeholders that sometimes take a year or two to become requests. This work moves a data scientist beyond handling tickets toward helping the company decide where data can make a difference.
