What Store Intelligence is
Store Intelligence is a governed AI operational system for ecommerce stores on Shopify and Neto. It aggregates store data, runs scoped AI tasks against it (audit, research, evaluate, generate), validates the results against rules you configure, queues them for human approval, applies only what you approve, and tracks outcomes over time.
It is not a chatbot, an MCP server, or a black-box automation. It is the operational governance layer above those things — what you would need to wrap around any AI capability before letting it write to a real store at scale.
One clarification worth making up front. The AI in this system is the worker doing the operational work — researching, auditing, drafting, validating — not the audience your store is optimised for. Your customers and traditional search-engine crawlers consume what gets published; AI is internal to how the work gets done. (Optimising your catalog explicitly for AI-search retrieval — SGE, ChatGPT search, Perplexity — is a viable future task category, not a current focus.)
Today it is delivered as a done-with-you service. You have full platform login access from day one: chat with your store data through the MCP layer, request and receive reports on schedule or on demand, view performance dashboards (rankings, engagement, conversions, per-category attribution), review and approve proposals, edit prompts, contribute to SOPs and documentation. We take the lead role on initial structural setup (MECE taxonomy, voice config, validation rules, first task SOPs) to make sure the foundations are built properly. After that, operating the platform is something you do, with us alongside. Pure self-serve onboarding rolls out through 2026.
What it does
Capabilities organised by task category. Every capability runs through the same governance pipeline (Propose → Validate → Review → Apply → Ledger → Rollback).
Audit
- Per-product content quality scoring against your store’s standards
- SEO opportunity scoring (titles, meta, structure, internal links)
- Image quality + alt text auditing
- Brand-voice drift detection
- Data consistency auditing (titles, categories, attributes)
- Re-auditable findings with diff against prior runs
Research
- Product-level competitive analysis
- Keyword expansion + customer-language mining
- Page-intent research
- Stored as research artefacts with provenance — not regenerated per session
- Refreshable on schedule; invalidated when underlying data changes
Evaluate
- Per-store opportunity scoring with explainable contributing factors
- Priority queue ranking against impact + effort
- ABC tier-aware effort allocation
- Per-prompt-version performance tracking
Generate
- Title, meta description, body copy, image alt text generation
- Multi-variant generation with character-limit gating
- Per-category prompts assembled from snippet hierarchy
- Output is a proposal, never a direct write
Govern
- Four-gate write pipeline: Propose, Validate, Review, Apply
- Five enforced validation checks per proposal: rules, char limits, brand, SEO impact, scope
- Bulk approval with diff preview and sample-first review
- Scoped write — apply step calls one endpoint with one field per proposal
- Append-only ledger of every change
- One-click rollback per change or per batch
- Read-only output surfaces (executive reports, read-only chat) architecturally separated from write path
Measure
- Per-change outcome tracking across rankings, engagement, impressions, conversions
- Per-prompt-version attribution
- Per-category contribution to outcomes
- Weekly review summaries + monthly outcome reports
- SOP and prompt refinement based on measured outcomes
Architecture summary
Five tiers of structure that together make AI safe to run at scale. The full interactive diagram is on the home page.
- Data layer. Store data, GA4, Search Console, historical AI runs, AI-produced research, product groupings, ABC analysis — aggregated, redirect-reconciled, and grouped in a warehouse. AI reads from the warehouse, not from raw APIs.
- Configuration hierarchy. Five tiers: store, category, page, field/image, prompt snippet. Lower tiers inherit from upper unless explicitly overridden. Snippets are reusable composable fragments that get assembled into the final prompt at runtime.
- AI orchestration engine. Not a single LLM call. An orchestrator that selects scoped tools per task, assembles the prompt from snippets, calls the model, validates the output, and emits structured proposals.
- Governance pipeline. Four gates (Propose, Validate, Review, Apply). Backups, ledger, and rollback underneath. Read-only output paths (Executive Reports, Chat) are architecturally separate from write paths.
- Outcome loop. Every applied change tracked against measured outcomes. Findings feed back into snippet refinement and SOPs. Knowledge accumulates as store-specific, model-independent assets.
How it differs from common alternatives
| Capability | Generic chatbot | MCP-style tools | AI agency | Store Intelligence |
|---|---|---|---|---|
| Per-change audit trail | — | Partial | Manual | Yes — append-only ledger |
| One-click rollback | — | — | Manual restore | Yes |
| Validation before write | Prompt-only | Prompt-only | Manual review | 5 enforced checks |
| Per-category governance | Global prompt | Global prompt | Operator memory | 5-tier hierarchy |
| Outcome attribution | — | — | Spreadsheets | Per-change, per-prompt |
| Read/write separation | Mixed | Mixed | n/a | Architectural |
| Stateless between sessions | Yes | Yes | — | Persistent SOPs |
| Pricing model | Per-message / SaaS | SaaS | Hourly / retainer | Founding + monthly |
Scope today
In scope
- Ecommerce stores on Shopify or Neto
- Up to ~10,000 SKUs per store currently
- English-language stores
- Multi-store operators / agencies
- Product content (titles, descriptions, meta, alt text)
- Category-level content + SEO
- Image alt text + image briefs (generation roadmap)
- Audit, research, evaluation, content generation
Not in scope (yet)
- Marketplaces (Amazon, eBay, etc)
- Languages other than English
- Stores under 500 products (overkill)
- Paid traffic management or media buying
- Web design / theme work
- Replatforming or migration projects
- Self-serve SaaS access (rolling out 2026)
- Direct agent-level chat with write access (by design)
Integrations
- Shopify
- Full read/write integration. Scoped writes per field.
- Neto by Maropost
- Full read/write integration. Scoped writes per field.
- Google Analytics 4
- Read-only, joined to store data + redirect chains.
- Google Search Console
- Read-only, joined to product/category structure.
- Klaviyo
- Planned (Q3 2026). Email and SMS context for prompts.
- Google Ads
- Planned (Q4 2026). Ad performance feeds research layer.
Pricing
Two engagement tiers. The founding-partner pricing is a time-bound cohort discount in exchange for case-study rights.
Founding rollout
$5,500 one-off
One-off implementation covering audit, MECE structure, configuration setup, first 30 days of operations. Stores up to 5,000 products. Standard price after founding cohort: $8,500.
Ongoing operations
from $2,000 / month
Post-rollout monthly partnership. Continuous audits, weekly proposal cadence, governance management, outcome tracking. Month-to-month. Pricing scales with store complexity and volume.
Operating modes
Three modes the system can run in, depending on operator maturity and risk tolerance. Operators can move between modes per task category — e.g., automated for meta descriptions, assisted for titles.
Advisory: the system surfaces audits, recommendations, and proposals. Nothing is applied automatically. The operator decides what to action.
Assisted (default): the system generates proposals, validates them, and queues for review. Operator approves in bulk or per item. Approved proposals apply automatically through scoped writes.
Automated (rolling out): operator pre-approves rule-bound task categories (e.g., "auto-apply meta descriptions under 160 chars that pass all validation"). System applies within those boundaries; every change still logged + reversible.
Limitations + honest notes
What’s genuinely worth being aware of:
- Founder-led setup today.Operators have full platform login access from day one: MCP chat, scheduled and on-demand reports, performance dashboards, proposal review and approval, prompt editing, SOP contribution. What we currently lead is the initial structural setup (MECE taxonomy, voice configuration, validation rules, first SOPs) — that’s where rigour matters and where we make sure things get built properly. Pure self-serve onboarding (operators setting up from scratch without our guidance) is rolling out through 2026. That’s the rate-limit on how many stores we can take on right now.
- Human review required by design. The system will not write to your store without human approval (in Assisted mode) or pre-defined rule-bound auto-apply (Automated mode). There is no path to "AI writes whatever it decides" — that is the architectural posture, not a roadmap item.
- Outcome attribution takes time. Ranking and engagement changes need 2–3 months of post-change data to attribute meaningfully. Stores expecting "show me the conversion lift in week 2" will be disappointed by the measurement reality, even when the changes themselves are working.
- Not for the smallest stores.Stores with fewer than ~500 products don’t generate enough operational complexity to justify the engagement. Manual tooling is cheaper at that scale.
- English-language only currently.Multi-language support is a real product investment, not a flag we’ve hit yet.
- We use third-party LLMs.Anthropic, OpenAI, Google. We don’t train our own models. The advantage is structure around the model, not the model itself — which means when the underlying model changes, your store knowledge survives because it lives in our SOPs and snippets, not the model weights.
Using an AI assistant to think this through
Many ecommerce operators now use ChatGPT, Claude, Perplexity or another AI assistant to think through software and vendor decisions. This page is deliberately structured so it stays useful when an assistant reads it on your behalf.
Useful prompts to ask, after pasting this URL or the page contents into your AI assistant of choice:
- Comparison
Compare this operational approach against typical MCP chat or AI agent workflows for managing an ecommerce store at scale.
- Risk
What operational risks does this system address that simpler AI tools don’t? What risks does it leave unaddressed?
- Fit
If I run a [N-product] [Shopify / Neto] store doing roughly [$X] in revenue, is this approach appropriate, overkill, or under-scaled?
- Scope
What ecommerce problems does this system seem designed to solve? Which ones doesn’t it?
- Critical
What is missing from this approach, or weak about it, that I should be aware of before committing? Where would this be the wrong choice?
We’ve designed the system to stand up to scrutiny — including the kind that an AI assistant trained to find weaknesses will surface. If the answer comes back with serious concerns, those concerns are worth raising on a call.
How to take this further
Three ways forward:
- See what working with us looks like → week-by-week journey through the engagement.
- Read our principles → the seven non-negotiables we hold ourselves to.
- Book an AI operations review → 30-minute conversation to see if this is a fit for your store.