# Collaboration and Strategy

Vin Vashishta, V Squared | MLOps Podcast | Episode 176 | 51:54
Hosted by Demetrios Brinkmann

Source: https://www.youtube.com/watch?v=GnzFO4Z1u24
Channel: MLOps Community, now AAIF Live (https://www.youtube.com/@AAIFLive-x1r). Summarised by MLOps Talks.
Page: https://mlopstalks.com/talks/collaboration-and-strategy
Published: 2023-09-19
Tags: engineering-culture, product-strategy, team-adoption

## TL;DR
- Generative AI makes many data tasks faster, but it does not remove the need for data, models, causal reasoning, or data science teams.
- Data and AI products need business strategy, product management, and domain knowledge because their economics and customer use differ from digital software.
- Technical professionals increase their impact by becoming multipliers who teach, remove roadblocks, and help their teams deliver more value together.

## Summary
Vin Vashishta argues that data and AI work only creates business value when teams understand the strategy behind a project. Generative AI has sped up parts of data work, but it has not made data collection, cleaning, modeling, causal analysis, or deployment easy. Businesses still need data professionals, while also adding people who understand strategy and data and AI product management. Vin explains that data and AI assets can support many use cases, so teams should choose opportunities based on total monetization potential rather than the first obvious feature. Large language model features also need a competitive analysis because an unpriced feature that competitors can copy may have no moat. He recommends roles that connect business goals with technical execution. Data and AI strategists can help leaders plan, while product managers can guide adoption and returns. Technical professionals can grow by becoming multipliers through teaching, mentoring, documentation, and better team practices.

## Key ideas
### Generative AI speeds up data work without making the hard parts disappear
[05:02](https://www.youtube.com/watch?v=GnzFO4Z1u24&t=302s)
Vin says generative AI lets him do the same work faster, but it does not make the work easier. Data science still needs data, and teams still need to build models. He says generative models do not work out of the box for every task. He expects continued work on causal machine learning, causal discovery, robotics, autonomous systems, and models that run on lower-resource devices. He argues that businesses would hobble themselves by getting rid of their data science teams. Generative AI may reduce the need for highly specialized researchers in many companies, while foundational model research remains concentrated in AI-first companies.

### Projects should be selected when their returns can grow faster than their costs
[10:13](https://www.youtube.com/watch?v=GnzFO4Z1u24&t=613s)
Vin's business test is whether returns scale faster than costs. As models become more advanced, their costs can grow faster than their returns for many use cases. Companies such as Google, Amazon, Meta, Walmart, Airbus, and shipping firms can justify highly reliable models because they have problems and transactions at scale. Smaller businesses cannot assume that a tiny accuracy improvement will produce enough value. Vin says data professionals need business acumen, while leaders need to know how to involve other parts of the business and lead strategy. He separates the need for strategy from the need for more PowerPoint presentations.

### Strategy connects technical execution to business value
[14:32](https://www.youtube.com/watch?v=GnzFO4Z1u24&t=872s)
Vin acknowledges that strategists have earned a poor reputation through vague recommendations, hand-waving, and decisions such as CNN Plus. He calls the accumulated effect of these weak decisions strategic debt. He explains that strategy is a framework for creating several possible plans and for deciding which goal to pursue, rather than a fixed list of instructions. Business conditions change, so advice that works for one company may fail at another. Data science can support strategy by modeling complex business systems, helping teams understand customers and markets, and measuring feedback loops when expected outcomes do not appear.

### Technical professionals grow by becoming multipliers
[20:18](https://www.youtube.com/watch?v=GnzFO4Z1u24&t=1218s)
Vin asks engineers, analysts, and data scientists to consider how they can become multipliers. Technical skill and personal output eventually hit a ceiling because one person can work only so many hours. Staff-level contributors increase their impact by teaching, mentoring, and making the whole team more capable. Leaders create opportunities, remove roadblocks, and focus their teams on higher-value work. Demetrios Brinkmann connects this idea to documentation, pairing, code reviews, and model reviews. Vin agrees that the multiplier is the improvement other people can make after learning from you, rather than simply using a tool to produce more individual output.

### Data and AI need product managers who understand their distinct economics
[29:29](https://www.youtube.com/watch?v=GnzFO4Z1u24&t=1769s)
Vin says data and AI product managers are becoming more common because these products are built, used, and monetized differently from digital software. A data set can support reports, models, and later use cases without being depleted. Software usually requires substantial work to adapt from one use case to another. Products can move from digital features to data features, analytics, and machine learning over time. Vin gives Microsoft's use of Copilot and GPT across products as an example of repeated monetization. He argues that putting data and AI into a digital product framework causes businesses to miss how customers interact with these assets.

### A generative AI feature needs a moat as well as a business case
[38:44](https://www.youtube.com/watch?v=GnzFO4Z1u24&t=2324s)
Vin agrees that a feature can be worthwhile even when customers do not pay for it directly, if it brings in enough new customers or improves another business metric. He says that is only the start of the analysis because competitors may copy an API-based feature after the market has been proven. His advice is to ask who can copy the feature, how long copying will take, and what protects the investment. He sees data, domain expertise, business understanding, data quality, and the combination of prompting and fine-tuning as possible barriers. The best opportunities are those where the company's own knowledge makes imitation difficult.

### Organizations need roles that exchange business and technical knowledge
[45:12](https://www.youtube.com/watch?v=GnzFO4Z1u24&t=2712s)
Vin recommends that data professionals advocate for roles that are currently missing, including data product managers, AI product managers, and technical or AI strategists. Product managers can sit in a product organization when it collaborates openly with the data organization. Strategists should work with the strategy or product organization when one exists, or from the data organization when necessary. A CDO or senior data leader should have access to the C-suite and connect technical concerns to strategy planning. These roles teach the business about data and AI while bringing business literacy into the data team. They also let data professionals return to engineering or science work if they do not want strategy responsibilities.

## Notable quotes
- Vin Vashishta: "Your initiatives returns must scale faster than costs." (10:13)
- Vin Vashishta: "Strategy isn't a plan, strategy's a framework that allows you to come up with multiple plans that lead you to success to that goal that you've decided you want to pursue." (17:24)
- Vin Vashishta: "How can I become a multiplier? That's the biggest question that Engineers, data scientists, analysts, data Engineers, doesn't matter if you're technical, you need to be asking yourself." (20:38)
- Vin Vashishta: "If you're introducing the feature and no one's willing to pay for it, why are you doing it?" (41:42)
- Vin Vashishta: "You can't have an AI product without someone that understands AI products." (43:39)

## Tools & references mentioned
- V Squared
- From Data To Profit
- ChatGPT
- Google
- Amazon
- Meta
- Walmart
- Airbus
- FedEx
- Microsoft
- Azure
- OpenAI
- Bing
- Office 365
- VS Code
- Mastercard
- PowerPoint
- Transformers
- causal machine learning
- causal inference
- robotics
- autonomous systems
- IoT
- Quantum Computing

## Who should watch
- You are building a data or AI feature and need a way to connect its technical cost to business returns.
- You lead a data team that is being asked to own product decisions or business strategy without clear roles or support.
- You are a technical contributor deciding how to increase your impact beyond writing code and delivering individual projects.

## Related talks

- [Product Thinking in Data & AI](https://mlopstalks.com/talks/product-thinking-in-data-ai) (Stuart Winter-Tear, 10:27)
- [Vision and Strategies for Attracting & Driving AI Talents in High Growth](https://mlopstalks.com/talks/vision-and-strategies-for-attracting-driving-ai-talents-in-high-growth) (Ashley Antonides, Two Six Technologies & Olga Beregovaya, Smartling & Shailvi Wakhlu, Shailvi Ventures LLC, 30:34)
- [AI Operations Without Fundamental Engineering Discipline](https://mlopstalks.com/talks/ai-operations-without-fundamental-engineering-discipline) (Nikhil Suresh, Hermit Tech, 49:28)
- [What Business Stakeholders Want to See from the ML Teams](https://mlopstalks.com/talks/what-business-stakeholders-want-to-see-from-the-ml-teams) (Peter Guagenti, Tabnine, 1:21:28)
- [Extending AI: From Industry to Innovation](https://mlopstalks.com/talks/extending-ai-from-industry-to-innovation) (Sophia Rowland & David Weik, SAS, 1:01:37)