Product Management – 3 Core Skills

Product management is a multifaceted discipline that sits at the intersection of creativity, technology, and business strategy.

Successful product managers (PMs) are visionaries who turn ideas into impactful solutions that drive business growth.

In the dynamic world of product management, mastering the triad of creativity, technical knowledge, and commercial acumen is crucial.

Each of these skills plays a vital role in transforming a product from a mere concept to a market success.

By honing your core competencies you can drive innovation, user satisfaction, and business growth.

Whether you’re an aspiring PM or a seasoned professional, investing in these three core skills will help you become a more effective and impactful product leader.

To excel PMs need to master three core skills: creative thinking, technical knowledge, and commercial acumen

1. Creative

Product management is a creative endeavour. Product managers are responsible for envisioning and designing products that solve real-world problems, delight users, and stand out in the market.

Creativity is about innovative problem-solving, empathy, and user-centric design.

Creative Skills Matters

Identifying User Needs

A PM must deeply understand the users’ pain points and needs. This requires creative thinking to go beyond surface-level solutions and discover what truly makes a difference for the end user.

Brainstorming Solutions

Creativity enables PMs to ideate multiple solutions and approach challenges from different angles. It’s about generating a wide range of possibilities and selecting the one that best aligns with the product vision.

Designing Differentiated Products

In a competitive market, products that capture attention are often those with unique, thoughtfully designed features. Creative product managers can craft compelling user experiences that differentiate their product from competitors.

Example: Airbnb

Consider how Airbnb revolutionised travel accommodation.

The creative insight to turn people’s homes into short-term rentals wasn’t just about filling a market gap; it was about creating an entirely new user experience that connected travellers with unique, local living spaces.

2. Technical

While product managers are not required to be engineers, having a solid technical foundation is crucial. 

Understanding the technical aspects of a product helps PMs make informed decisions, effectively communicate with development teams, and ensure that their product visions are feasible.

Technical Skills Matter

Facilitating Communication with Engineers

PMs need to bridge the gap between business goals and technical implementation. Knowing the language of developers helps in translating complex business requirements into actionable technical tasks.

Prioritising Features

A technically proficient PM can better evaluate the feasibility of various features, understand the constraints of the technology stack, and make informed trade-offs between different product enhancements.

Ensuring Product Quality

Understanding the technical underpinnings of a product allows PMs to assess risks, anticipate potential issues, and ensure a high-quality product launch.

Example: Mobile App

Think of a PM working on a mobile app. Understanding the implications of backend infrastructure, API integrations, and data security can help them collaborate effectively with developers to build a robust and scalable product.

3. Commercial

The third essential skill for product managers is commercial acumen; an understanding of the business side of products. PMs must be able to assess market opportunities, define product strategies, and align them with the company’s financial goals.

Commercial Skills Matter

Market Analysis and Positioning

PMs need to analyze market trends, competitive landscapes, and customer feedback to identify opportunities and threats. This knowledge helps in positioning the product effectively to maximise market penetration and user adoption

Revenue and Profitability

PMs are responsible for ensuring that the product contributes to the company’s bottom line. This involves setting pricing strategies, identifying monetization opportunities, and optimising the product to increase profitability.

Strategic Decision-Making

Commercial skills empower PMs to make data-driven decisions about product features, go-to-market strategies, and scaling efforts. This ensures that the product not only meets user needs but also aligns with broader business objectives.

Example : Freemium Launch

Consider the strategic decisions behind launching a freemium model. A PM with strong commercial insight might identify that offering a free version of a product can drive user acquisition, with the option to upgrade for premium features, thereby increasing revenue over time.

Balancing the Triad

The real challenge for product managers lies in balancing these three core skills. Creativity, technical knowledge, and commercial acumen are interdependent.

A product manager might excel in one area but must develop competencies across all three to deliver a successful product. For instance:A creative PM may excel in ideation but needs technical skills to assess feasibility and commercial acumen to ensure market viability.

A technical PM can understand the intricacies of product development but may need to enhance their creativity to design user-centric solutions and their business skills to optimise for revenue.

A commercially savvy PM may be great at identifying market opportunities and driving growth but needs creative and technical inputs to build a product that users love.

How Do Large Language Models (LLMs) Work?

Large Language Models (LLMs) have taken centre stage in artificial intelligence (AI) applications

They power a wide range of services, from conversational agents like ChatGPT to advanced text analysis tools. But how do these models work? 

We need to break down the complex mechanisms and processes involved in training, optimising, and deploying these models and their architectural principles.

1) The Basics of Language Modeling

A language model is a statistical model that predicts the probability of a word or sequence of words given the context of preceding words. The goal is to understand and predict language patterns, enabling the model to generate text, answer questions, or engage in dialogue.

Language models range from simple n-gram models, which predict the next word based on the last “n” words, to more sophisticated models like recurrent neural networks (RNNs) and transformers, which analyse longer and more complex sequences of text. 

The transition to LLMs like GPT (Generative Pre-trained Transformer) and BERT (Bidirectional Encoder Representations from Transformers) marks a significant leap due to their ability to capture and understand intricate language patterns.

LLMs are built using deep learning techniques, particularly neural networks—a computing system designed to recognize patterns. These models learn the structure of language by ingesting vast amounts of data, then using this information to generate human-like text.

2) Transformer Architecture: The Foundation of LLMs

The foundation of LLMs like GPT-4, BERT, and others lies in the Transformer architecture, which was introduced by Vaswani et al. in their 2017 paper titled “Attention is All You Need.” 

Before transformers, RNNs and LSTMs (Long Short-Term Memory networks) were popular for language tasks. They had limitations particularly in handling long-range dependencies and parallelization during training.

3) Key Concepts in Transformers

– Self-Attention Mechanism: This is a central innovation of the transformer architecture.

Instead of processing sequences in order like RNNs, transformers analyse relationships between words in a sentence at once, allowing them to weigh the importance of each word relative to others, regardless of their position. This mechanism is crucial for understanding context in long passages.

– Positional Encoding: While transformers process all tokens simultaneously, they still need to understand the order of words. Positional encoding introduces this order by adding unique, learned vectors to the word embeddings.

– Multi-Head Attention: To capture different types of linguistic relationships, transformers use multiple “attention heads” to focus on various aspects of the input sequence simultaneously.

– Feedforward Neural Networks: Once the self-attention layer is completed, transformers use standard feedforward layers to further refine and interpret the information.

– Layer Normalisation: After each operation, transformers apply normalisation to maintain stable training and prevent degradation of gradients.

This architecture revolutionised NLP (Natural Language Processing) because it allowed models to handle far more complex tasks than before. It’s the backbone of most state-of-the-art LLMs, including the GPT (Generative Pre-trained Transformer) series from OpenAI.

3) Training Process and Datasets

Training a large language model is computationally intensive and requires massive datasets. LLMs are typically pre-trained on extensive corpora of text data, which may include books, websites, academic papers, and more. The training process can take weeks or months on powerful hardware (usually GPUs or TPUs), and it involves several stages:

Pre-training

In the pre-training phase, the model is trained to predict missing words in a sequence (a task called masked language modelling in BERT) or to predict the next word in a sequence (as in causal language modelling in GPT). This stage is unsupervised, meaning the model doesn’t require labelled data; it simply learns by trying to predict text based on patterns observed in the data.

During pre-training, the model adjusts its weights—a set of parameters that represent what the model has learned—to minimise the error in its predictions. These weights are updated iteratively using techniques like backpropagation and gradient descent, optimising the model over time.

Fine-tuning

After pre-training, LLMs are often fine-tuned on smaller, task-specific datasets using supervised learning. A model might be fine-tuned on question-answering data, summarization tasks, or sentiment analysis to make it more effective for specific applications.

Fine-tuning allows the model to adapt its general knowledge to particular tasks, making it more precise in areas where high accuracy is required.

4) How LLMs Generate Text

Once trained, LLMs can generate text in a manner that mimics human-like responses.

Input Tokenization

When a user inputs text, the model first converts the input into tokens. Tokens are small units of text, which could be individual words or even sub-word fragments. Tokenization is essential because it helps the model work with fixed vocabulary sizes and ensures that even out-of-vocabulary words can be represented through sub-word components.

Contextual Understanding

Next, the model applies its deep learning layers to analyse the context of the tokens. The self-attention mechanism helps the model understand relationships between different parts of the input, identifying which tokens are most relevant in generating the next word.

Text Generation

When generating text, the model predicts one token at a time, using probability distributions over the possible tokens. These probabilities are generated based on the input and the model’s learned knowledge. The model may then select the most probable token, or employ techniques like temperature scaling (which controls the randomness of the predictions) or beam search (which considers multiple potential sequences before selecting the most appropriate one).

This step-by-step prediction is repeated until the model generates the full sequence of text, resulting in coherent and contextually relevant sentences.

5) Practical Applications of LLMs

LLMs have a wide range of practical applications, spanning various industries:

– Conversational AI: LLMs are used to power chatbots and virtual assistants that can engage in natural conversations, handle customer queries, and provide recommendations.

– Content Creation: They can generate creative writing, including articles, stories, and poetry. LLMs are increasingly used in marketing to draft emails, blog posts, and social media content.

– Language Translation: Models like GPT and BERT can be fine-tuned for translation tasks, helping bridge language barriers.

– Code Generation: Codex, a variant of GPT-3, can generate computer code from natural language descriptions, assisting software developers.

– Text Summarization: LLMs can condense long documents or articles into shorter summaries while preserving the key points.

– Sentiment Analysis: Companies use LLMs to analyse customer reviews, social media posts, and other textual data to gauge sentiment and make business decisions.

6) Limitations and Future Directions

Despite their impressive capabilities, LLMs have several limitations:

– Bias and Fairness: LLMs often reflect biases present in their training data, which can result in biased or harmful outputs. Researchers are actively working on methods to mitigate bias and ensure that AI systems are fair and unbiased.

– Context Length: While LLMs can handle long sequences of text, they still struggle with maintaining coherence in very long conversations or documents. Efforts are being made to extend context windows and improve the models’ ability to handle longer-term dependencies.

– Data and Energy Costs: Training large language models requires massive amounts of data and computational power. This has raised concerns about the environmental impact of such models, as well as the accessibility of AI technologies to smaller organisations that may not have the resources to train them.

– Factual Accuracy: LLMs sometimes generate plausible but incorrect or nonsensical answers. This is because the models are not truly “understanding” the world—they are pattern-recognition systems that rely on statistical correlations in the data rather than actual knowledge.

7) The Future of LLMs

We can expect LLMs to become more efficient, accurate, and accessible. Areas of active development include:

– Smaller, more efficient models that deliver performance comparable to large models but with significantly lower resource requirements.

– Multimodal models, which integrate not just text but also images, audio, and video, enabling more comprehensive understanding and interaction.

– Real-time adaptability, where models can update their knowledge dynamically rather than being fixed after training.

Large language models have revolutionised the way machines understand and generate human language. Their underlying architecture; built on the transformer model, combined with massive datasets and sophisticated training techniques, allows LLMs to perform a wide range of tasks that were previously thought to be the exclusive domain of humans.

Challenges like bias, high computational costs, and occasional inaccuracies remain. As the field evolves, LLMs will likely become even more powerful and adaptable, opening new possibilities for AI-driven innovation across industries.

AARRR Framework

AARRR is accepted as the 5 most important metrics for a startup to focus on. These metrics effectively measure your company’s growth while being both simple and actionable.

AARRR stands for the 5 phases in your customer life cycle. These are:

• Acquisition: Where are customers coming from?
• Activation: How to turn acquired customers into active customers
• Retention: How to make potential customers come back
• Referral: How to get customers to recommend your service
• Revenue: How to turn potential customers into paying customers

Who are your customers?

If you want to craft a great solution, you need to know your customers. Do not start by creating a product for everyone – instead focus on specific buyer personas.

To discover more about your target group, ask these powerful questions:
• What keeps my customers up at night?
• What motivates my customers to solve their problem?
• What’s the desired future state for my customers?

Acquisition

Activation occurs when a person becomes an active user. A user may download an app, but if they never use it again, they haven’t been activated.

Driving traffic is important, you should also conduct A/B tests with your landing page tests to determine what brings the best conversion results.

Track user behavior to determine what makes a user commit to your product.

Activation

Your users need an aha moment — the instant a user realises your product delivers meaningful value and makes them want to come back. If a significant percentage of users are not activating – create an onboarding strategy to educate and hand hold users through the initial phases of achieving their goals with your solution is one way to increase activation.

Retention

To reduce churn, you need a solid retention strategy. Beyond creating a valuable product, it’s also critical to keep in touch with your customers.

Automated emails are a simple retention strategy. Create email campaigns to automatically send at various intervals after sign-up. This helps remind your customers of your product and how to use it.

Grammarly sends weekly personalised emails to share users’ writing stats, eg their most common writing mistakes and level of accuracy

Referral

Referral is just one of the many acquisition channels. When someone likes your product enough to tell others, thats the best sign that you’ve created something of great value and people are ready to pay for it.

User recommendations is one of the most effective ways to acquire new customers. According to a Nielsen study, 83% of people trust recommendations from friends and family, and 66% of people trust consumer opinions posted online.

Revenue

One of the most important success metrics is customer lifetime value. The metric describes the total revenue generated by a customer over their lifetime.

To Calculate the customer lifetime:

Customer Lifetime Value = Average revenue per customer * (1/churn rate)
‍
If your average monthly revenue per customer is €100 and your churn rate is 5%, this means that your customer lifetime value: (100 * (1/0,05) = €2000.

Another approach to CLV…

Customer Lifetime Value = Average monthly revenue per customer X (# months) customer lifetime = ($) LTV

Customer lifetime value vs customer acquisition cost

A benchmark is a customer lifetime value to customer acquisition ratio of 3:1.

To find the channel with the lowest CAC and the highest return, apply the Bullseye Framework, which you apply during the acquisition phase.

A maturity model is a tool for evaluating how the processes, people, and systems that drive a product are performing.

A maturity model will give you tiered levels of achievement for objectively assessing the maturity in every aspect of your product and teams.

Models define effectiveness and sophistication levels and can pinpoint a person, team, project or company’s current position within the model.

Maturity models are an important strategic prioritisation tool because they provide flexible performance assessment and monitoring that can reveal valuable information about your company’s health and potential.

While a model won’t fix inefficiencies themselves, it can identify areas where you aren’t operating at standard and allow you to determine strategies that can improve your operations and processes.

3 Types Of Maturity Models

Business Process Maturity Model (BPMM)

The Business Process Maturity Model’s roots can be traced back to the Process Maturity Framework (PMF) created by Watts Humphrey at IBM in the 1980s.

The process maturity model explores the ways to introduce quality practices in software development. Humphrey and his colleagues introduced incremental stages to adopting best practices in software organisation.

The PMF served as the groundwork for the development of the Capability Maturity Model (CMM) for software in 1991. CMM then became the foremost standard for appraising the capability of software development organisations.

Initial

Inconsistent management practices or teams that react to crises rather than predict them.

Managed

Teams and businesses have a management foundation, but the individual teams within the business still work in silos with minimal collaboration or evidence of incorporating improvement strategies.

Standardised

The business is aware of its processes and is working toward consistency and uniform delivery.

Predictable

Organisations use their process infrastructure and asset capabilities to achieve reliable results by controlling the variations within their outputs.

Innovating

Your company is continuously improving and focused on innovation.

Agile ISO Maturity Model

By standardising the levels, agile ISO models set more clearly defined expectations determined by an international body.

Agile Maturity Model helps organisations understand their current practices and work toward improving them with the goal of increasing ability to respond to changing business conditions and better harnessing innovation.

Level 1: Documented Processes

Your processes are documented -in any form from a Word doc, to an Excel spreadsheet. Knowing what your processes are for different areas of your business is the first step to taking a process-led approach.

Level 2: Followed Processes

There’s little point in having documented processes if they’re not put into action. Processes have to be followed and used regularly by your team. This means that work gets done in the best way every time, and also means that the teams who follow them will be able to point out obvious flaws and inefficiencies.

Level 3: Managed Processes

Processes should shape and inform the way in which you conduct your operations. You need to structure things to allow processes to play a big role. This could include getting dedicated process management software, setting up a specialist process team, harvesting and analysing the data from the processes.

Level 4: Optimised Processes

By now you have the infrastructure in place to yield rich accurate data. This lets you begin to understand your organisation in a transparent manner– identifying every event as and when it occurs. This is an accurate model of your business and lets you know not just what happened this week, but what will happen next.

Level 5: Integrated Processes

The core business has been systemised in line with business needs, you can begin to develop and devise new processes to link different stakeholders. These could be processes between leadership and staff, they could be internal peer audit systems, or they could be processes which bring customers or business partners into the heart of the organisation.

Once the organisation has been systemised, the processes can be leveraged to improve the nature, culture, and awareness of the organisation. eg you could reduce your carbon footprint or make your supply chain more sustainable.

Capability Maturity Model (CMM)

The capability maturity model assesses the maturity of your organisation, or software development systems, by comparing it to best industry practices.

By measuring results and assigning maturity levels, companies and development teams can use their models to evaluate their awareness of business processes, effective management techniques and areas for improvement.

Continuous Delivery Maturity Model Example

Continuous delivery is all about reducing risk and making sure business get their return on investment as soon as possible. It is about removing focus from repetitive and risky tasks to devoting all you effort into delivering business value.

Continuous delivery has become increasingly popular. Implementing it is often considered to be difficult and risky

The road to continuous delivery is paved with many small goals, each of which deliver value by themselves. A maturity model will serve as a guide on this road.