Why the Obvious Solutions Don't Work

In the last post, I explained why ecommerce companies struggle with data-driven decisions: fragmented data across platforms, limited analytical capacity, and the gap between having unified data and being able to use it.

Most companies respond in one of three ways. They hire analysts. They buy BI platforms. They add LLMs or automation tools.

All three can help. None of them, on their own, solves the real issue: how do I use my data to make better decisions consistently and at speed?

That is what operators actually care about. Not whether a dashboard exists. Not whether an analyst can eventually answer a query. Not whether an AI tool can generate a response. The real pressure is knowing what to focus on right now, what is actually driving performance, and what should change next.

Without clear answers to those questions, companies accumulate tools and reports, but decision velocity does not materially improve.

Hiring Analysts Creates Bottlenecks Nobody Can Route Around

Companies hire experienced analysts or data scientists, often at £80K to £150K or more for senior roles. Initially, this feels like progress. Questions get answered. Reports improve. Models get built.

Then demand increases. Marketing wants CAC by channel and blended. Operations wants inventory velocity by category. Finance wants contribution margin by SKU. Commercial wants LTV by acquisition cohort. In most mid-market organisations, these requests route through a small central team. A backlog forms. Priorities compete. Decisions wait.

The bottleneck shifts from access to data to access to analytical capacity.

A widely cited survey by CrowdFlower found that data scientists reported spending 60 to 80 percent of their time cleaning and organising data rather than generating insights (CrowdFlower / Figure Eight, 2016). While tooling has improved since then, fragmented cross-platform ecommerce data still consumes disproportionate effort in many organisations.

There is also domain complexity. Ecommerce is not generic analytics. Returns flows affect revenue recognition. Stacked discounts alter margin calculations. Attribution windows differ across platforms. Fulfilment costs fluctuate by channel and geography. Inventory ageing impacts working capital. These mechanics need to be understood properly before any metric is trusted.

Analysts hired as generalists require time to internalise that context. When they leave, much of that accumulated understanding leaves with them.

This is not a talent problem. It is structural dependency.

BI Platforms Show You What Happened, Not What To Do Next

The second response is to invest in a BI platform such as Tableau, Power BI, or Looker. These tools are powerful. They centralise reporting and improve visibility across the organisation.

They show what has happened. Revenue is down. Conversion has dropped. CAC has increased. Inventory turnover has slowed.

What they do not automatically provide is direction. A dashboard can highlight that performance has deteriorated, but it does not tell you which lever matters most right now, whether the issue is pricing, traffic quality, discounting, fulfilment, or product mix, or which intervention will have the greatest impact. Visibility is not the same as prioritisation.

Over time, ecommerce teams accumulate dashboards across marketing, operations, finance, and commercial functions. Each answers a specific reporting question. Few answer the broader operational question: what should we do next?

Industry surveys, including the BARC BI & Analytics Survey, show that many organisations struggle to achieve broad, sustained adoption of BI tools across business users (BARC, BI & Analytics Survey, latest edition). When dashboards multiply but do not translate into clearer prioritisation, engagement declines.

Teams rarely suffer from lack of visibility. More often, they suffer from too many metrics and no clear direction.

That is how organisations end up drowning in dashboards.

LLMs and Automation Tools Still Require Significant Build and Ongoing Maintenance

The newest response is to layer in LLMs or automation tools and ask questions in natural language rather than writing SQL or building dashboards. In demonstrations, this appears seamless. In practice, it requires substantial engineering and ongoing maintenance.

An LLM is a reasoning engine. It does not automatically connect to Shopify, Meta, GA4, Klaviyo, your ERP, or your warehouse in a structured, governed way. To use it reliably for ecommerce analytics, you still need to build and maintain API integrations, manage authentication and rate limits, handle schema changes, and create pipelines that normalise and join data across platforms.

You must also define how your business calculates its metrics. CAC, contribution margin, LTV, inventory velocity. You need to encode how returns are treated, how partial refunds affect revenue, how stacked discounts impact margin, how attribution windows differ across channels, and how fulfilment costs are allocated.

Without those definitions being explicit and governed, an LLM will infer logic from prompts and generic assumptions. Even with deterministic settings, outputs can diverge from how your business actually operates.

The work does not stop after initial setup. Ecommerce businesses evolve constantly. New discount structures are introduced. Channels are added. Attribution models change. Product categories expand. Pricing adjustments are made. Tax and currency rules shift. Each change requires updates to pipelines, metric definitions, validation rules, and monitoring systems.

Introducing LLM workflows also adds another operational layer. Orchestration frameworks, prompt management systems, model routing, cost controls, guardrails, logging, monitoring. You are not removing complexity. You are adding another layer to manage.

LLMs can accelerate analysis and improve accessibility. They do not remove the need to build and maintain the system that ensures metrics are consistent and decisions are grounded in reality.

Without something purpose-built for ecommerce decision-making, you are still assembling and operating that system yourself.

The Pattern Across All Three

Hiring analysts increases capacity but creates dependency. BI platforms improve visibility but do not guide action. LLMs improve accessibility but require significant build and maintenance.

All three approaches can contribute. None of them, alone, ensures that operators know what to do next.

For small to mid-market ecommerce companies, the constraint is not lack of tools. Executive research continues to show that becoming truly data-driven remains elusive for many organisations (NewVantage Partners, Big Data and AI Executive Survey, latest edition). The real gap is turning data into clear, prioritised direction fast enough to influence performance.

In the next post, I will outline what a system designed specifically to guide ecommerce decisions looks like and why it changes the equation.

Georges

Key Takeaways

• Hiring analysts increases analytical capacity but often creates throughput bottlenecks and dependency on scarce expertise.

• Data professionals frequently report spending 60 to 80 percent of their time on preparation rather than insight generation (CrowdFlower / Figure Eight, 2016).

• BI platforms show what has happened but typically do not prescribe what action to prioritise next.

• Many organisations struggle to achieve broad, sustained BI adoption across business users (BARC, BI & Analytics Survey).

• LLMs and automation tools require substantial build and ongoing maintenance, including integrations, metric definitions, validation, and monitoring.

• The real challenge is turning fragmented ecommerce data into clear, prioritised decisions at speed.

Frequently Asked Questions

Q: Why doesn’t hiring analysts solve ecommerce analytics challenges?

Analysts increase capacity but do not eliminate preparation overhead or dependency on scarce expertise. Research shows data professionals often spend the majority of their time on preparation rather than insight generation (CrowdFlower / Figure Eight, 2016). In mid-market organisations, analytical demand frequently outpaces small team capacity.

Q: Don’t BI tools like Tableau or Power BI solve the problem?

BI platforms improve visibility and reporting. They assume clean integration and consistent metric definitions already exist. Dashboards typically show what has happened, not what action should be prioritised next. Industry research indicates many organisations struggle with widespread BI adoption (BARC, BI & Analytics Survey).

Q: Can LLMs replace analysts or BI tools for ecommerce analytics?

LLMs can improve accessibility and accelerate analysis. However, they require structured pipelines, governed metric definitions, API integrations, validation frameworks, and ongoing maintenance. Without that foundation, outputs may be plausible but not consistently aligned with business logic.

Q: What is the real bottleneck for mid-market ecommerce companies?

The bottleneck is decision velocity. Turning fragmented data into clear, prioritised actions fast enough to influence performance.

Sources

CrowdFlower (Figure Eight). 2016 Data Scientist Report.

BARC. BI & Analytics Survey.

NewVantage Partners. Big Data and AI Executive Survey.