Skip to main content
ftes.aiftes.ai
E-commerce · the commerce engine

An AI analyston your ownstore data.

The commerce engine is an AI business analyst that runs on a brand's own commerce data. It connects to the store read-only, copies every order, product, variant, customer, inventory level, discount, refund and tax line into its own database every fifteen minutes, computes every business metric itself in a deterministic analytics engine, and then uses AI to explain what the numbers mean, draft the messages a manager would write, and propose the decisions worth making today. Around that core sit the systems a real retailer runs on — support tickets, returns, warehouse packing, in-store footfall, email marketing, search, social — each connected so its data can be joined to sales. The result is one place where questions that used to need six reports have one answer, priced in dollars, with the evidence attached.

  • It is not a customer-support chatbot.
  • It is not a thin AI layer on top of a store admin.
  • It never changes commerce data.
Get startedSee the e-commerce workforce
The three laws

Trust is the architecture, not a line on this page.

These are the promises the whole engine is built on. They are what makes AI usable for money decisions.

  1. 01

    The backend calculates.

    Revenue, margin, cover, return rate, conversion, tax, landed cost — every figure is computed by a deterministic, unit-tested analytics engine using set-based SQL against the platform's own database. The same number appears on every page.

  2. 02

    The AI interprets.

    AI agents receive already-computed figures and turn them into verdicts, narratives, drafts and recommendations. No AI in the product can produce a number of its own. If a number appears in an AI answer, a backend tool produced it.

  3. 03

    A human approves.

    Every action that leaves the platform — an email to a manager, a list pushed to a marketing tool, a post published, a blog article, ad spend — is shown in a Review & Send screen first. Unattended sending is an explicit switch, off by default. Every send is logged and later scored.

  4. 04

    Honest data.

    A fourth, closely related promise: honest data. When a figure cannot be computed, the page says why — “No cost data”, “Not reported”, a coverage figure — rather than showing a zero. Test orders and catalogue errors are excluded and disclosed.

01 · CALCULATES02 · INTERPRETS03 · APPROVES
[ E.01.A ] Three cells, three laws. The engine is the tallest because every other cell stands on it; the signature is a person on paper.
How it works

Connect, sync, compute, interpret, decide — and then measure.

No page ever calls the store live. The engine keeps its own copy of the data and does its own arithmetic, which is why it is fast and why every number agrees with every other.

  1. Step 1

    Connect

    The commerce store connects read-only. Other systems — helpdesk, returns, warehouse and packing, in-store people counting, email marketing, web behaviour analytics, search console, social — are added on an Integrations page as needed, with write-only secret fields or OAuth.

  2. Step 2

    Sync

    A scheduler pulls new and changed records every fifteen minutes into the platform's own database. Heavy sources run on their own nightly schedules. Inventory updates in real time by webhook. Every run is logged; a run ledger shows what ran, when it last succeeded and what it produced.

  3. Step 3

    Compute

    The analytics engine calculates every metric with set-based queries — dashboard KPIs, true profit per SKU after every deduction, days of cover, conversion per trading hour, tax per jurisdiction, landed cost per line. No page ever calls the store live, which is why pages load in milliseconds against millions of rows.

  4. Step 4

    Interpret

    AI agents read the computed figures and write the morning brief, the executive report, page commentary, chat answers, draft manager emails, long-form articles, social posts and creative.

  5. Step 5

    Decide and measure

    The owner approves, sends, publishes or skips. A Results ledger then measures what each action actually changed — a before/after verdict per send, rolled up into a value ledger. Separate surfaces show how the AI itself performed and whether every scheduled job actually ran.

What it does

Six areas. Every named surface exists and is running.

Organise the product around these areas. Each surface below is a page in the engine, and each area is staffed by roles from the e-commerce org chart.

01 · Where every day starts.

Command Center

Morning Brief
The daily action page: what needs attention, today's decisions with their evidence and dollar value, what happened overnight, what the platform is watching, and integration health. Each decision card has a Review & Send flow.
Dashboard
The numbers page for any date range: a money-at-stake strip (cash sitting still, cash leaked in the last thirty days, cash at risk if nothing changes), KPI tiles with comparison deltas, a net-sales bridge from gross through discounts and returns to net, a what-changed-and-why decomposition, dollar-priced alerts, channel split, per-location cards and top products.
Chat with the analyst
Ask anything in plain English. Answers stream live, lead with the conclusion, and come with source chips, a trust footer showing exactly which tools were called and how fresh each system's data is, an optional chart derived from the tool output, and one-click follow-up actions. A full workspace and a side panel on every page; Explain this on every KPI card; select any text and ask; slash shortcuts; projects and pinned answers; a watchlist; and voice — a live spoken conversation with the same analyst.
Executive Report
A written daily, weekly or monthly report, store-wide or scoped to a brand or location, generated on a schedule or on demand, with every past report kept and a PDF export.

Staffed by · Chief of Staff FTE · Data Trust FTE

02 · What to sell, stock, buy and clear.

Merchandising

Product analyst
Winners, losers and money leaks: which products and brands make money, which quietly lose it, true profit per item after discounts, returns and cost, brand-level verdicts, and a landed-cost engine — record a shipment from its invoices plus a CSV and every margin figure switches to real landed cost for that sale date, cent-exact.
Inventory analyst
What to reorder, what to move between locations, what is stuck, what is bleeding cash. Every list ranked by money, with the send-to-manager workflow built in.
Foresight
The forward-looking view: true profit after every deduction, the season that is coming, pace to goal, size intelligence, stockout predictions and new-drop winners. Every forecast is scored against reality daily.

Staffed by · Controller FTE · Cash Officer FTE · Buyer FTE · Allocator FTE · Demand Forecaster FTE

03 · The transaction record turned into worklists.

Operations

Orders
Every order from placement to fulfilment to refund; searchable, filterable, exportable; the drill-down target for every order alert. Years of history backfilled.
Warehouse
Packing performance per packer and per order, plus the live unfulfilled backlog and whether orders are getting stuck.
Returns analytics
Why products come back and what it costs: return rate, profit lost, sizing defects, category breakdown, and the specific page fix each product needs.
Discount monitoring and loss prevention
Company to location to employee to transaction to item drill-down of every discount, split into automatic and staff-entered manual, with policy alerts (oversized, over the limit, stacked, employee outlier), and per-code before-versus-during verdicts showing which codes gave margin away to people who were buying anyway.
Sales tax reporting
Per-period reports reconciled to the cent, exported to a spreadsheet in the accountant's own column layout, with a filing view by jurisdiction and a nexus watch by state.

Staffed by · Logistics Officer FTE · Warehouse Officer FTE · Returns Officer FTE · Pricing Officer FTE · Landed Cost & Compliance FTE

04 · Who stopped buying, who is asking, who is walking in.

Customers and Support

Customers and money finder
Who stopped buying, what they are worth, when they usually come back, and a vetted win-back list that a human reviews before it is pushed to the email tool. Plus a customer health view.
Team performance
Customer support measured: volume, speed, resolution, CSAT and backlog per agent, every conversation searchable, revenue at risk, and what customers keep asking about.
Store traffic
In-store footfall paired with sales: visitors, conversion, which store wins, lost sales in busy hours, peak hours, a cost-neutral staffing plan; plus website friction from behaviour analytics.

Staffed by · Customer Intelligence FTE · Lifecycle Officer FTE · Voice of Customer FTE · Store Officer FTE

05 · The digital employees that earn attention.

Growth

Brand visibility engine
A cited Brand Brain knowledge base (upload documents, files, URLs; the platform extracts facts, personas, voice samples and word rules, each human-approved, each re-checked on a schedule), an inbound topic engine built from real search data, long-form article generation grounded in that knowledge base, human approval, publishing to the storefront blog, and measured reach from search console.
Organic social manager
Turns the catalogue into posts, reels and carousels in the brand voice, plans the week, routes everything through a human, publishes to the connected social platforms, and measures what it earned. Includes AI image, carousel and video generation.
Performance advertising
The media buyer's control room: profit-aware campaign proposals grounded in real commerce data, human-gated spend, a monthly cap, a real kill switch and an audit ledger. Money-safety first: no real spend can move without explicit human approval.

Staffed by · Visibility Officer FTE · Content Officer FTE · Social Officer FTE · Acquisition Officer FTE

06 · Did it work? Did it run?

Outcomes and Platform

Results
Every message the platform sent on the owner's behalf and whether it worked: a before/after verdict per send, rolled into a value ledger, plus manager accountability.
AI performance
What the AI agents did, how reliably, how fast, and what it cost. Every surface metered from the agents' own traces, with a budget guard.
System
Did the platform actually run? Every scheduled job, its health judged against its own cadence, its last success, what it produced, and its full run history.
Integrations and Settings
Every connected system in one place, with write-only secrets, OAuth connections and swappable AI provider cards (text, image and video vendors are configurable data, not hard-coded). Profile, connection status, report schedule, unattended-sending switches, and the manager each location's alerts go to.
In-app documentation
One page per feature, written so that anyone who reads it before opening the software already knows all of it. A test fails the build if any part of the app ships without its documentation.

Staffed by · Chief of Staff FTE · Data Trust FTE

The roles, by department, in the org chart

Connected systems

Every system a retailer runs on, joined to sales.

Integrations are described by capability. The commerce platform is read-only and feeds everything; the rest are added one at a time on the Integrations page.

CapabilityWhat it contributes
Commerce platform (read-only)Orders, products, variants and costs, inventory, customers, discounts, refunds, locations, tax lines — every fifteen minutes. Feeds everything.
HelpdeskSupport tickets, messages, agents, CSAT.
Returns platformReturn records, reasons, sizes, exchanges, processing times.
Warehouse and packingPacking performance per packer and order.
People countingIn-store footfall per store per hour.
Email marketingReceives the human-reviewed win-back list.
Web behaviour analyticsPage, source and device behaviour; deep-links to recordings rather than rebuilding them.
Search consoleImpressions, clicks and position per published article.
Social platformsPublishes approved posts; returns insights.
Email (SMTP)Sends every manager and supervision email.
AI model providersMultiple vendors configurable as cards — one default for text, one for images, one for video. Vendor-independent by design.
Who it is for

Brands whose tooling computes revenue, not profit.

Brands roughly one to fifty million in revenue, typically apparel or lifestyle, usually running both an online store and physical shops. The segment's defining problem: their tooling computes revenue, not profit. Enterprise systems that do compute true margin start far above this price point.

  • The owner or founder

    Wants four decisions with the money attached before the shops open, instead of four tools to open.

  • The e-commerce manager

    Wants sales, products, inventory, customers, discounts and refunds watched in one place.

  • The operations manager

    Wants stock, fulfilment and availability tracked against what is actually selling.

  • The buyer or merchandiser

    Wants to decide what to reorder, transfer, mark down or drop, with the arithmetic shown.

How it is different

It comes back to say whether the advice worked.

When you ask how this is different from what you have, these are the answers — with the ones still being built marked as roadmap.

  1. 01

    It measures its own advice.

    Every tool in this market gives advice; almost none comes back to say whether the advice worked. The Results ledger records a before/after verdict per send today; the roadmap extends it to a declared prediction, its measurement and a published accuracy — including the bad months.

  2. 02

    Fully-loaded contribution margin at this price point.

    Cost, freight, duty, discount, returns, shipping, fees and carrying cost — not revenue minus COGS.

  3. 03· Roadmap

    A demand calendar that knows when this business's peaks actually fall.

    Moving festivals, school terms, wedding seasons, local weather — instead of a generic calendar that says mark down exactly when demand is about to arrive.

  4. 04· Roadmap

    Fit intelligence built from the brand's own returns and exchange evidence.

    “This label runs one size small across these styles” — evidence a competitor without the returns data cannot reproduce.

  5. 05

    Realised discount and stacking measurement.

    The actual blended discount given, stacking included, against the headline rate on the code — an essentially unowned category.

  6. 06

    Physical-shop depth.

    Contribution per trading hour, footfall-to-sale conversion, clienteling, online-to-store attribution — what online-only tools structurally cannot match.

  7. 07

    Everything priced in dollars.

    Wherever a list could be sorted by count, it is sorted by money instead, so the top row is always the biggest lever.

  8. 08

    Trust as architecture, not as a marketing line.

    Deterministic calculation, human approval, full audit, honest gaps, and a data-trust gate that makes advice go quiet and say why when its foundation is shaky.

Roadmap · the direction of travel

A digital workforce, measured on what it earned.

The product's organising idea is a digital workforce: each capability becomes a named employee with a written job description, a human manager, an authority level, and one number it is judged on, plus a monthly scorecard. “Enable the inventory module” is a software evaluation; “hire a merchandiser” is a purchase decision compared against a salary. The roles are catalogued today in the e-commerce org chart; the mechanics below are the direction the engine is being built in.

  1. Roadmap · 01

    A shift report as the daily front door

    What was done overnight, what needs a decision ranked by money, what is being watched, how last month's decisions were measured, what needs attention. Conflicts between employees — one wants to reorder what another wants to mark down — resolved before the owner sees them.

  2. Roadmap · 02

    An authority and approval framework

    Written per-employee limits, an approval queue, a permanent log, one-click undo, and a kill switch. Owners refuse software that acts because they cannot see what it will do, cannot bound it, and cannot undo it; solve those three and automation becomes something people pay for.

  3. Roadmap · 03

    Employee memory

    It remembers recommendation, decision, reason and outcome, learns thresholds, and stops re-proposing things you declined. A competitor can copy a feature in a quarter; they cannot copy years of measured decisions on your data.

  4. Roadmap · 04

    Prediction capture and outcome measurement

    No recommendation can be shown unless it first declares what it expects, by when, measured how — and the system then runs its own declared measurement, records the result, computes the error, and publishes its own accuracy whether or not the news is good.

  5. Roadmap · 05

    A value ledger

    An auditable tally of money identified, released, at risk and forgone, drillable to the individual decision. Outcome-based pricing and a found-money guarantee become possible precisely because outcomes are measured.

Major planned capability groups, in rough priority order: real margin · cash locked in stock · fit, returns and quality · buying and the real demand calendar · discount leakage · customer lifecycle · AI-shopper visibility · shipping, duty and disputes · physical shops and omnichannel · suppliers and product development · the workforce platform itself · proving what it earned · multi-tenancy · go-to-market.

The roles that exist today, in the org chart

Questions

The questions visitors ask, answered honestly.

What is it, in one line?

An AI business analyst that reads your store's data, works out what is actually making and losing money, and tells you what to do about it each morning — with every number computed, not guessed.

Is it a chatbot?

There is a chat analyst in it, but that is one surface. The product is an analytics engine plus a set of AI workers, and the chat is how you interrogate them.

Can AI be trusted with my numbers?

The AI never calculates. Every figure comes from a deterministic engine that is unit-tested and produces the same number on every page. The AI's job is to read those figures and explain them. And every answer shows you which tools produced which number and how fresh each source is.

Will it change anything in my store?

Not your commerce data — the connection is read-only. The only things that leave the platform are content and messages a human explicitly approved, and there is a kill switch and an audit log for all of it.

How long does setup take?

Connecting the store is the whole first step; historical data backfills and then syncs every fifteen minutes. Other systems are added when you want them, one at a time. For an exact figure for your setup, ask the team rather than take a guess.

How much does it cost?

Pricing depends on setup and which capabilities are switched on. The roles it runs are hired on ftes.ai's metered plans; the engine's own terms are confirmed with the team.

Does it work with a platform other than Shopify?

The platform is built today on Shopify, and the architecture separates every connected system behind an adapter, so other platforms are a build, not a rewrite. The team confirms timing.

Do you support multiple stores or brands?

Every record is scoped by store from the ground up, so multi-store is architecture that already exists. Current availability is confirmed with the team.

We already have an analytics tool.

Ask what question it cannot answer. The usual gap is profit rather than revenue, cash locked in stock, realised discount, and nobody coming back to check whether last month's advice worked.

Glossary

The words the engine uses, defined once.

Landed cost
The true cost of a unit on the shelf: supplier price plus freight, duty, clearing, handling and shrinkage, dated so past periods do not move.
Contribution margin
What is left per order line after discount, returns, shipping, fees, acquisition cost and carrying cost. The one honest number.
Days of cover
How long current stock lasts at the current rate of sale.
Dead stock
Units with no sale in a defined window, carrying a computed cash and holding cost.
Net-sales bridge
Gross sales, then discounts, then returns, then net sales, shown as a waterfall.
Realised discount
The actual blended discount given, including stacking, as opposed to the headline rate on a code.
Review & Send
The screen that shows exactly what will leave the platform, to whom, before it goes.
Value ledger
The auditable running tally of money identified, released, at risk and forgone.
Digital employee / FTE
A named capability with a job description, a human manager, an authority level and one number it is judged on.
Run ledger
The record of every scheduled job: what ran, when it last succeeded, and what it produced.

Connect the store. Read the first brief.

The store connects read-only, the history backfills, and the roles from the e-commerce org chart start proposing — with the money attached and your signature required.

Get startedTalk to us