Data Roles in Practice: How Data Engineers, Scientists, and Analysts Actually Grow SaaS and eCommerce Businesses

In my previous article, I explained the difference between data engineers, data scientists, and data analysts using a coffee subscription business as an example. That piece told you what each role is. This piece tells you what each role does, in granular detail, with specific examples for SaaS and eCommerce businesses.

For every use case, I will explain why it matters to your business, how it is actually done, what impact you can expect with realistic revenue and profit numbers, and I will rate the importance from 1 to 10 so you know where to focus first.

This is the practical playbook. No theory. Just tactics that make money.

Data Engineer: Building the Foundation That Makes Everything Else Possible

Before any clever analysis or prediction can happen, someone needs to make the data trustworthy and accessible. That is the data engineer. Here is what they actually build and why it matters.

1. Unified Customer View

Importance: 10/10

Why it matters: Your customer exists in fifteen different systems. Their email behaviour is in Klaviyo. Their purchases are in Shopify or Stripe. Their support tickets are in Zendesk. Their product usage is in your application database. Their ad clicks are in Meta and Google. Right now, these are fifteen different people as far as your data is concerned. You cannot answer basic questions like "which customers who clicked our Facebook ad last month are still active users today?"

How it is done: The engineer builds identity resolution, matching the same person across all systems using email, user ID, device fingerprints, and probabilistic matching. They pull data from each source via APIs or database connections, transform it into a consistent format, and load it into a central warehouse like BigQuery, Snowflake, or Redshift. They set up automated pipelines that refresh this data hourly or daily.

SaaS example: A B2B SaaS company with 5,000 customers discovers after building a unified view that 340 of their "churned" customers are actually still using the product under a different email domain because their company got acquired. That is £180,000 in annual revenue they thought was lost but is actually retained, and 340 customers they should be nurturing for expansion rather than ignoring.

eCommerce example: A fashion retailer links their online and in-store purchase data and discovers that customers who shop both channels have a lifetime value 3.2 times higher than online-only customers. They shift £50,000 of monthly marketing budget toward driving online customers into stores, increasing annual profit by £420,000.

Impact: Without a unified customer view, every analysis is partial and often misleading. This is literally the foundation.

2. Event Tracking Infrastructure

Importance: 9/10

Why it matters: If you want to know what users actually do in your product or on your website, you need to track events: button clicks, page views, feature usage, scroll depth, video plays, form submissions. Most companies either track nothing useful or track everything badly with inconsistent naming and missing properties.

How it is done: The engineer defines an event taxonomy, a structured list of every event worth tracking with consistent naming conventions and required properties. They implement tracking using tools like Segment, Rudderstack, or custom instrumentation. They build validation to catch tracking bugs before they corrupt your data. They ensure events flow reliably into your warehouse.

SaaS example: A project management SaaS adds proper event tracking and discovers that users who create their first project within 24 hours of signup have an 85% Day 30 retention rate, compared to 31% for users who take longer. This single insight reshapes their entire onboarding flow. They add a guided project creation wizard, Day 30 retention increases by 18 percentage points, and annual revenue increases by £890,000 from reduced churn.

eCommerce example: A home goods retailer implements scroll depth and product interaction tracking. They discover that customers who use the zoom feature on product images convert at 2.4 times the rate of those who do not. They make the zoom feature more prominent and add a prompt encouraging its use. Conversion rate increases by 0.3 percentage points, worth £720,000 in annual revenue on their traffic volume.

Impact: You cannot improve what you do not measure. Event tracking is how you measure.

3. Data Warehouse and Transformation Layer

Importance: 9/10

Why it matters: Raw data is messy. Dates are in different formats. Currencies need conversion. Session definitions vary. Product categories are inconsistent. Before anyone can analyse this data, it needs to be cleaned, transformed, and organised into business-friendly tables.

How it is done: The engineer builds transformation pipelines using tools like dbt, Airflow, or custom scripts. They create derived tables: a clean customers table with one row per customer, a clean orders table with consistent revenue figures, a sessions table with proper attribution. They document everything so analysts know what each field means.

SaaS example: A SaaS company's raw data shows 47,000 monthly active users. After proper deduplication and bot filtering in the transformation layer, the real number is 31,000. The previous inflated number was leading to dramatically wrong unit economics calculations. Correcting this reveals that their customer acquisition cost is actually 52% higher than they thought, which completely changes their growth strategy and prevents them from scaling a fundamentally unprofitable acquisition channel.

eCommerce example: A retailer's raw order data includes test orders, returns processed as negative orders, and duplicate records from a system migration. The transformation layer creates a clean revenue table showing true net revenue per customer. This reveals that their "best" customer by gross revenue is actually unprofitable after accounting for a 78% return rate. They stop sending this customer expensive retention offers, saving £34,000 annually.

Impact: Bad data leads to bad decisions. The transformation layer is where data becomes trustworthy.

Data Analyst: Turning Data Into Decisions

With clean, unified data in place, the analyst answers questions the business actually cares about. Here is what that looks like in practice.

4. True Customer Lifetime Value Calculation

Importance: 10/10

Why it matters: Most businesses calculate LTV wrong. They take total revenue divided by total customers and call it average LTV. This blends your best customers with your worst, your oldest cohorts with your newest, and tells you almost nothing useful. True LTV calculation segments customers properly and reveals where your actual profit comes from.

How it is done: The analyst calculates LTV by cohort (signup month), by acquisition channel, by initial product purchased, by customer segment. They account for gross margin, not just revenue. They model future value for customers who have not yet completed their lifecycle. They calculate the variance, not just the average, because a customer base with average LTV of £200 where everyone is between £180 and £220 is very different from one where half are at £50 and half are at £350.

SaaS example: A SaaS company calculates LTV by acquisition channel and discovers that customers from Google Ads have an LTV of £1,400, customers from content marketing have an LTV of £2,800, and customers from their affiliate programme have an LTV of £650. They were allocating budget by volume of signups, which favoured affiliates. Reallocating to weight by LTV increases annual profit by £340,000 with the same total ad spend.

eCommerce example: A beauty retailer segments LTV by first product category purchased. Customers who first buy skincare have a 3-year LTV of £380. Customers who first buy makeup have a 3-year LTV of £210. Customers who first buy fragrance have a 3-year LTV of £95. They redesign their acquisition funnels to emphasise skincare as an entry point, increasing average customer LTV by 23% and adding £1.2 million in annual profit.

Impact: LTV tells you how much you can afford to spend to acquire a customer. Getting it wrong by channel or segment means systematically overspending on bad customers and underspending on good ones.

5. Cohort Analysis and Retention Curves

Importance: 9/10

Why it matters: Looking at overall metrics like "monthly active users" or "repeat purchase rate" blends together customers acquired at different times under different conditions. Cohort analysis separates them, showing whether your January customers behave differently from your June customers, and whether your business is actually getting healthier or just growing the top while the bottom falls out.

How it is done: The analyst groups customers by signup month (or first purchase month) and tracks their behaviour over time: retention rate at month 1, 3, 6, 12; revenue per customer at each interval; engagement metrics. They compare cohorts to spot trends. They overlay changes like product launches, pricing changes, and marketing campaigns to understand what caused shifts.

SaaS example: A SaaS company's overall monthly churn is 4.2%, which seems stable. Cohort analysis reveals that recent cohorts are churning at 6.8% in their first six months, masked by excellent retention in older cohorts acquired before a pricing change. This early warning prevents disaster: they adjust the pricing packaging for new customers, bringing early-cohort churn back to 4.5% and avoiding what would have been a £1.8 million annual revenue loss as the newer cohorts matured.

eCommerce example: A subscription box company sees 55% of customers making a second purchase. Cohort analysis shows this has declined from 68% to 41% over eighteen months. They diagnose the cause: a supplier change reduced product quality eight months ago. Reverting to the original supplier and running a winback campaign for affected cohorts recovers £280,000 in annual revenue.

Impact: Cohort analysis is your early warning system. It shows problems before they become catastrophes.

6. Marketing Attribution Analysis

Importance: 9/10

Why it matters: Every marketing platform tells you it drove the sale. Facebook says the customer clicked their ad. Google says they searched first. Email says they opened the campaign. The affiliate says they used their link. They are all taking credit for the same £50 order. Without proper attribution, you have no idea which channels actually work and you are almost certainly wasting money.

How it is done: The analyst builds attribution models that assign credit to touchpoints based on their actual contribution. They compare last-click attribution (what most platforms show) with multi-touch models, incrementality testing, and media mix modelling. They reconcile platform-reported conversions with actual revenue in the data warehouse to find the truth.

SaaS example: A SaaS company spends £80,000 monthly on LinkedIn Ads. The LinkedIn dashboard shows 200 trial signups per month. The analyst matches these to the data warehouse and finds that only 127 of those trials actually signed up within 7 days of the click. Of those, 18 converted to paid. Of those paid conversions, 11 came from companies already in the sales pipeline from other sources. The true incremental paid conversions from LinkedIn are 7, making the actual cost per incremental customer £11,400, not £400 as reported. They reduce LinkedIn spend by 70% and reallocate to channels with proven incrementality, improving ROI by £380,000 annually.

eCommerce example: A retailer attributes 35% of revenue to their affiliate programme based on last-click. An incrementality analysis, where they turn off affiliates for a random subset of customers, reveals that 80% of affiliate-attributed orders would have happened anyway without the affiliate coupon. The programme is cannibalising organic sales. Restructuring the programme to target only new customers improves profit by £190,000 annually.

Impact: Attribution is how you stop wasting money on marketing that looks good but does nothing.

7. Funnel Analysis and Conversion Optimisation

Importance: 8/10

Why it matters: You know your overall conversion rate. But where exactly are people dropping off? Is it the landing page, the signup form, the payment step, the first-time user experience? Funnel analysis shows you precisely where the leaks are so you can fix the right things.

How it is done: The analyst defines the key funnels: visitor to signup, signup to activation, trial to paid, cart to checkout to purchase. They measure conversion at each step, segment by traffic source, device, geography, and customer type. They identify the highest-impact drop-off points and work with product and marketing to test improvements.

SaaS example: A SaaS company has a 2.1% visitor to trial signup rate. Funnel analysis reveals that mobile visitors convert at 0.4% versus 3.8% for desktop, and mobile is 45% of traffic. The mobile signup form is broken on certain browsers. Fixing it increases overall signups by 31%, worth £520,000 in annual revenue.

eCommerce example: A retailer's checkout funnel shows 68% of customers abandon at the shipping cost reveal. Testing free shipping over £50 (their average order value is £47) increases average order value by 12% and reduces abandonment by 23 percentage points. Net impact: £890,000 additional annual profit after absorbing the shipping costs.

Impact: Funnel analysis finds the specific problems. Without it, you are guessing where to optimise.

8. Product Usage Analysis

Importance: 8/10 (SaaS) / 6/10 (eCommerce)

Why it matters: In SaaS, customers who do not use the product cancel. Understanding which features drive retention, which are ignored, and which usage patterns predict expansion lets you focus product development and customer success efforts where they matter.

How it is done: The analyst measures feature adoption rates, usage frequency, depth of engagement, and correlates these with retention and expansion outcomes. They identify the "aha moments" that separate retained customers from churned ones. They build health scores combining multiple usage signals.

SaaS example: A project management tool measures usage of twenty-three features. The analyst finds that customers who use the Gantt chart feature have 89% annual retention versus 61% for those who do not. But only 23% of customers ever try Gantt charts. They create an onboarding flow that guides users to create a Gantt chart in week one. Feature adoption increases to 54%, retention increases by 12 percentage points, and annual revenue increases by £1.4 million from reduced churn.

eCommerce example: A furniture retailer analyses browse behaviour and finds that customers who use the "view in room" AR feature convert at 4.2% versus 1.8% for those who do not. They promote the feature more prominently in the product page hierarchy and add it to marketing emails. AR usage increases from 8% to 22% of sessions, driving £340,000 in additional annual revenue.

Impact: Usage analysis tells you what your product should emphasise and what you should guide customers toward.

Data Scientist: Prediction, Experimentation, and Optimisation

With clean data and clear analysis in place, the data scientist builds models that predict the future and tests interventions to change it.

9. Churn Prediction Model

Importance: 9/10

Why it matters: Reacting after a customer churns is expensive: you have already lost the revenue and you are starting from a position of broken trust. Predicting churn before it happens lets you intervene while there is still a relationship to save.

How it is done: The scientist builds a machine learning model trained on historical data of customers who churned versus those who stayed. Features include usage patterns, support interactions, billing issues, engagement trends, and customer attributes. The model scores every active customer daily or weekly with a probability of churning in the next 30/60/90 days. Customer success teams use these scores to prioritise outreach.

SaaS example: A SaaS company with £8 million ARR and 6% monthly churn builds a churn prediction model with 78% precision and 65% recall. They create an intervention programme for the top 200 highest-risk customers each month: proactive outreach, health check calls, training sessions. The programme reduces churn among contacted customers by 40%. Net impact after programme costs: £680,000 in retained annual revenue.

eCommerce example: A subscription retailer predicts which customers are likely to skip their next box or cancel entirely. They target high-risk customers with personalised offers and product previews. Intervention reduces cancellations by 22% among the contacted group, retaining £410,000 in annual subscription revenue.

Impact: Churn prediction turns reactive customer success into proactive retention. The revenue impact is immediate and measurable.

10. Customer Segmentation for Personalisation

Importance: 8/10

Why it matters: Not all customers are the same, but most businesses treat them as if they are. Data-driven segmentation groups customers by behaviour, value, and needs, enabling personalised experiences that convert better and retain longer.

How it is done: The scientist uses clustering algorithms (k-means, hierarchical clustering, or similar) on behavioural and transactional data to identify natural groupings. They profile each segment: who they are, what they buy, how they engage, what they value. Marketing, product, and sales teams then create segment-specific strategies.

SaaS example: A B2B SaaS segments customers into five groups based on usage patterns, company size, and feature adoption. Segment 3, "power users at small companies," has the highest expansion rate but the lowest proactive outreach. Focusing sales development on this segment increases expansion revenue by £290,000 annually with the same team.

eCommerce example: A home and garden retailer segments customers by purchase patterns and identifies a "seasonal decorator" segment who buys heavily in autumn and spring. Targeted campaigns to this segment four weeks before their typical buying window increase their annual spend by 34%, adding £520,000 in revenue.

Impact: Segmentation makes marketing more efficient and customer experience more relevant.

11. Price Optimisation and Elasticity Analysis

Importance: 8/10

Why it matters: Most businesses set prices once and forget them, or change them based on gut feeling. Price elasticity analysis reveals how demand responds to price changes at different levels and for different customer segments, enabling pricing that maximises revenue or profit.

How it is done: The scientist analyses historical data where prices varied (promotions, A/B tests, regional differences) to estimate elasticity: the percentage change in demand for each percentage change in price. They model optimal price points for different products, segments, and objectives. They design pricing experiments to test hypotheses.

SaaS example: A SaaS company tests three price points for their mid-tier plan: £79, £99, and £119 per month. Elasticity analysis reveals that demand drops 8% for each £20 price increase, but the revenue per customer increases more than enough to compensate. They raise the price to £119, volume drops 12%, but revenue increases 18% and profit increases 31% due to lower support load. Net impact: £460,000 additional annual profit.

eCommerce example: A consumer electronics retailer analyses price elasticity by product category. They discover that accessories (cables, cases, chargers) have very low elasticity; a 15% price increase reduces volume by only 4%. They raise accessory prices, increasing margin by £280,000 annually with minimal volume impact.

Impact: Price optimisation is one of the highest-leverage activities in any business. Small percentage changes in price flow directly to the bottom line.

12. Recommendation Engine

Importance: 7/10 (eCommerce) / 6/10 (SaaS)

Why it matters: Showing customers products or features they are likely to want increases conversion, average order value, and engagement. Manual merchandising cannot personalise at scale. Recommendation engines can.

How it is done: The scientist builds collaborative filtering models ("customers who bought X also bought Y"), content-based models (matching product attributes to customer preferences), or hybrid approaches. They train on historical purchase and browse data. They deploy recommendations across product pages, email, homepage, and checkout.

eCommerce example: A fashion retailer implements personalised recommendations on product pages, replacing "you might also like" selections made by merchandisers. The model-driven recommendations increase click-through by 45% and add-to-basket by 28%. Impact: £1.1 million additional annual revenue from increased average order value.

SaaS example: A B2B SaaS with a large feature set builds a recommendation system for features based on the customer's industry, size, and current usage patterns. They surface these recommendations in-app and via email. Feature adoption increases by 19%, which correlates with improved retention. Impact: £180,000 in retained annual revenue.

Impact: Recommendations personalise at scale. The ROI compounds as the model improves with more data.

13. Incrementality Testing and Experimentation

Importance: 9/10

Why it matters: Correlation is not causation. Just because customers who received an email bought more does not mean the email caused the purchase. Incrementality testing measures the true causal effect of a marketing or product intervention by comparing treatment groups with proper control groups.

How it is done: The scientist designs randomised controlled experiments. For marketing: hold out a random subset who do not receive the campaign and compare their behaviour to those who do. For product: A/B test new features with random assignment. They calculate incremental lift, not just pre/post comparison, and account for statistical significance.

SaaS example: A SaaS company believes their onboarding email sequence is essential. An incrementality test randomly withholds emails from 10% of new signups. The results show that users who receive emails have only 3% higher activation than those who do not, not the 25% lift they assumed from observational data. The sequence is mostly reaching people who would have activated anyway. They redesign the sequence to focus on truly at-risk users, improving efficiency and saving £45,000 in email platform costs while maintaining outcomes.

eCommerce example: A retailer runs a "win-back" campaign offering 20% off to lapsed customers. An incrementality test shows that 70% of the "won back" customers would have returned anyway without the discount. They are giving away margin unnecessarily. Restricting the offer to customers with lower predicted return probability increases net profit from the campaign by £120,000 annually.

Impact: Incrementality testing stops you from paying for behaviour that would have happened anyway.

14. Demand Forecasting

Importance: 7/10

Why it matters: Inventory costs money. Stockouts lose sales. Accurate demand forecasting balances these, ensuring you have enough product without having too much capital tied up in unsold inventory.

How it is done: The scientist builds time series forecasting models incorporating historical sales, seasonality, trends, promotions, external factors (weather, holidays, economic indicators). They produce probabilistic forecasts with confidence intervals, not just point estimates. They continuously update models as new data arrives.

SaaS example: While less directly applicable, a SaaS company uses demand forecasting to predict support ticket volume by day of week and time of day, enabling better staff scheduling. Optimised scheduling reduces average response time by 34% without increasing headcount, improving customer satisfaction and retention.

eCommerce example: A sporting goods retailer implements demand forecasting at the SKU level. Improved accuracy reduces overstock by 23% and stockouts by 41%. Impact: £380,000 annual savings from reduced inventory carrying costs plus £290,000 recovered revenue from fewer missed sales.

Impact: Forecasting is where data science directly touches operations and cash flow.

15. Lead Scoring for Sales Efficiency

Importance: 8/10 (B2B SaaS) / 5/10 (eCommerce)

Why it matters: Sales teams waste time on leads that will never convert while high-potential leads go cold. Lead scoring prioritises the sales queue based on actual likelihood to convert, focusing expensive human attention where it matters.

How it is done: The scientist builds a model predicting conversion probability based on lead attributes (company size, industry, job title), behaviour (pages viewed, content downloaded, email engagement), and fit (how similar they are to existing customers). Leads are scored and ranked. Sales works the list from the top.

SaaS example: A B2B SaaS company's sales team works leads in chronological order. Implementing lead scoring reveals that the top 20% of leads by score contain 68% of eventual conversions. Prioritising these leads reduces average sales cycle by 12 days and increases conversion rate by 28%. Impact: £720,000 additional annual revenue from the same sales team.

Impact: Lead scoring makes sales teams dramatically more efficient by focusing on the right prospects.

Where to Start: The Importance Rankings

If you are wondering where to focus first, here is the prioritised list by importance rating:

10/10: Do These First

• Unified Customer View (Data Engineer)
• True Customer Lifetime Value Calculation (Data Analyst)

9/10: Do These Next

• Event Tracking Infrastructure (Data Engineer)
• Data Warehouse and Transformation Layer (Data Engineer)
• Cohort Analysis and Retention Curves (Data Analyst)
• Marketing Attribution Analysis (Data Analyst)
• Churn Prediction Model (Data Scientist)
• Incrementality Testing (Data Scientist)

8/10: Do These When Foundations Are Solid

• Funnel Analysis (Data Analyst)
• Product Usage Analysis (Data Analyst)
• Customer Segmentation (Data Scientist)
• Price Optimisation (Data Scientist)
• Lead Scoring (Data Scientist, B2B)

7/10: Do These for Optimisation

• Recommendation Engine (Data Scientist)
• Demand Forecasting (Data Scientist)

The pattern is clear. Engineering foundations come first. Analysis comes second. Prediction and optimisation come third. Skipping steps does not save time; it wastes it.

The Cumulative Impact

Let me add up the examples from a hypothetical company that implements all of this over two to three years:

• Unified customer view: £420,000 (eCommerce example)
• Event tracking: £720,000 (eCommerce example)
• True LTV calculation: £340,000 (SaaS example)
• Cohort analysis: £280,000 (eCommerce example)
• Marketing attribution: £380,000 (SaaS example)
• Funnel optimisation: £520,000 (SaaS example)
• Churn prediction: £680,000 (SaaS example)
• Price optimisation: £460,000 (SaaS example)
• Recommendations: £1,100,000 (eCommerce example)
• Incrementality testing: £120,000 (eCommerce example)

Total: £5,020,000 in annual revenue and profit improvement.

This is not fantasy. These are realistic numbers from real implementations. The cumulative effect of getting data right across the entire business is transformational.

The question is not whether you can afford to build data capabilities. The question is whether you can afford not to.

Ready to start building these capabilities? I can help you assess where you are, prioritise what to tackle first, and either build it myself or help you hire the right people. Get in touch for a data strategy consultation.