There’s a growing belief that plugging MCPs into ChatGPT or Claude is the future of AI-enabled commerce.

Connect your CRM.
Connect your helpdesk.
Connect your analytics.
Ask smarter questions.

Problem solved.

Except it isn’t.

Because AI only becomes truly powerful when it has full business context.

And most MCP-driven implementations connect a fraction of what actually drives performance.

If you’re feeding AI 1% of your business data, you shouldn’t expect 100% intelligent decisions.

You’ll get answers.

But you won’t get outcomes.

The Context Problem

AI models are exceptional at reasoning.

But they can only reason over the data they see.

When you connect a single source — like your CRM — the model can analyse that dataset well. It can identify patterns, rank flows, summarise performance, and answer direct questions.

What it cannot do is:

• Understand margin versus revenue

• See inventory constraints

• Detect paid channel cannibalisation

• Adjust for seasonality

• Identify incrementality

• Connect behavioural timing signals

• Apply compliance constraints

• Coordinate across lifecycle stages

Without full context, AI produces partial intelligence.

And partial intelligence is often more dangerous than no intelligence at all.

Example 1: “Which of My Flows Generate the Most Revenue?”

You connect your CRM to Claude.

You ask: “Which of my flows generate the most revenue?”

You get a clean, ranked answer.

Great.

Now what?

You know which flow generates the most revenue — but you don’t know:

• Which products inside that flow are actually driving retention

• Whether the revenue is coming from discounted, low-margin items

• Whether that revenue would have happened anyway

• Whether the timing of the flow is suboptimal

• Whether paid retargeting is claiming the same conversions

• Whether the flow is over-saturating high-value customers
  
  

You have a metric.

But you don’t have a strategy.

Now compare that to asking a fully connected system:

“Which of my flows generate the most revenue?”

A connected system responds with:

• Revenue and margin impact

• Incrementality analysis

• Product-level contribution

• Customer cohort performance

• Recommended timing adjustments

• Risk of fatigue or overlap

• Suggested optimisations
  
  

That’s the difference between querying data and orchestrating performance.

Example 2: “Should I Increase My Discount?”

Let’s say revenue has slowed.

You connect your ESP and ask: “Should I increase my discount from 10% to 20% in my win-back flow?”

The model may analyse past conversion rates and suggest: “Yes, historically higher discounts increase conversion.”

But without margin data, contribution data, or lifetime value context, that advice may destroy profitability.

A fully contextual system would instead evaluate:

• Margin thresholds by product

• Customer lifetime value by cohort

• Inventory velocity

• Purchase intent signals

• Cannibalisation risk

• Historical discount sensitivity
  
  

It might conclude: “Don’t increase discount. Instead, shift the timing earlier by 4 days and prioritise high-margin SKUs.”

That’s intelligence grounded in commercial reality.

ALSO READ: The Future of AI and Shopify: The Transaction Layer Is the New Battleground

Example 3: “What Should I Promote This Month?”

You connect Shopify and ask: “What should I promote this month?”

The AI may surface:

• Top-selling products

• Recently trending SKUs

• Best historical performers
  
  

But it won’t know:

• Which SKUs are margin-protected

• Which products drive second purchase

• Which products increase LTV

• Which SKUs are overstocked

• Which are supply constrained

• Which drive subscription attachment
  
  

A fully connected commerce AI would instead say: “Promote Product X to first-time buyers, Product Y to high-LTV cohort, and suppress Product Z due to margin compression.”

That’s not just analysis.

That’s orchestration.

Example 4: Attribution Blind Spots

You connect your email data and ask: “Why is my welcome flow underperforming?”

The model sees:

• Lower open rates

• Lower conversion

• Lower attributed revenue
  
  

So it suggests:

• Improve subject lines

• Adjust creative

• Increase urgency
  
  

But what if:

• Paid traffic quality has dropped

• Acquisition campaigns are discount-heavy

• Customers are arriving already converted

• Deliverability has shifted

• iOS privacy changes are masking performance
  
  

Without cross-channel visibility, AI will optimise the wrong lever.

With full context, it can identify whether the issue is acquisition quality, timing misalignment, or lifecycle saturation.

MCPs vs. Contextual Systems

MCPs are connectors.

They provide access to isolated datasets.

They are pipes.

But pipes do not create strategy.

True AI leverage in commerce requires:

• Unified data models

• Behavioural event streams

• Margin and product intelligence

• Compliance logic

• Cross-channel attribution

• Lifecycle state awareness

• Real-time orchestration
  
  

Without this, you’re effectively asking a very intelligent assistant to make decisions while blindfolded.

The Illusion of Smart Answers

There’s a subtle risk here.

When AI provides articulate answers based on limited data, it feels intelligent.

But clarity is not the same as correctness.

The more confident the answer sounds, the easier it is to trust.

And that’s where brands go wrong.

Because scaling a brand isn’t about isolated answers.

It’s about coordinated decision-making across:

• Revenue

• Margin

• Retention

• Inventory

• Acquisition

• Compliance

• Timing

• Customer psychology
  
  

That requires an agentic system — not a chatbot attached to a single dataset.

What Agentic Commerce Actually Means

Agentic commerce isn’t about asking better questions.

It’s about having a coordinated workforce of specialised AI agents that:

• Validate compliance before triggering flows

• Optimise product feeds based on margin rules

• Detect behavioural timing shifts

• Identify incrementality versus overlap

• Protect contribution margins

• Coordinate with acquisition activity

• Adapt over time as data compounds
  
  

As the system matures — typically over 60–90 days — performance becomes more precise.

Signals strengthen.

Recommendations sharpen.

Incremental lift compounds.

That doesn’t happen with isolated connectors.

It happens with connected intelligence.

AI Without Context Is Interesting

AI With Context Is Transformative

If you want AI to summarise your CRM, an MCP will do the job.

If you want AI to scale your brand profitably, it needs the full picture.

Because scaling isn’t about answers.

It’s about decisions.

And decisions require context.