An analytics agent for a neighbourhood store. It forecasts, spots problems and tells you what to do today, and you can just ask it. Type, tap a question, or use the mic.
Shelly tells you when a question is outside its skills (travel, email, writing, coding, investment advice, medical or legal questions, weather) and which kind of agent to connect in Settings. Nothing is sent to another agent unless you tap Send.
In-depth guide βAsk by typing, tapping a suggestion or speaking. Shelly checks whether the question fits its skills, answers from computed results, and can read answers aloud in your chosen region's voice.
In-depth guide βSix headline numbers for the last 7 days: sales, budget gap, gross margin, waste, delivery share and next week's forecast. Each is compared with last week, last year or a target.
In-depth guide βTap any number to ask Shelly about it.
Eleven report types (budget, 13-week forecast, weekly trading, category review, roster, stock, waste and four industry packs), each as an on-screen report, a PDF and an Excel workbook with dashboard, live formulas, PivotTables, lookups and raw data.
In-depth guide βEvery report opens on screen, saves as a PDF (white for printing, or the dark Shelly look for presenting) and downloads as an Excel workbook with a dashboard, live formulas, PivotTables, lookups and the raw data. Or just ask: "make a budget for the next 3 months".
Loading reportsβ¦
A ranked to-do list built from everything Shelly found. P1 = today, P2 = this week, P3 = next review. Each item gives the reason, the $ impact per week and an owner. Tick items off as you go.
In-depth guide βMajor judgements (big $, compliance, price, range, credit, markdowns) are proposals, not actions. Confirm or reject each one. Shelly records who decided what and when, and learns from your answers.
In-depth guide βMajor judgements (big $, compliance, price, range, credit) wait for you. Confirm or reject here, then Download and run python -m shelly decisions import shelly-decisions.json. Shelly learns from every answer.
The daily sales line with next week's forecast (dashed), a bridge that splits the week-on-week change into volume, mix and price, and the delivery channels for 14 days so outages stand out.
In-depth guide βVolume = units, mix = cheaper or dearer lines, price = promos and price changes.
A drop to zero is an outage, not a change in demand.
A profit and loss for each category (sales, margin, waste, contribution) plus the gap to budget. Tap a bar or row to ask Shelly about that category.
In-depth guide βUnusual events found by statistics, not guesswork: stock-outs, spikes, delivery outages, sustained declines (checked against last year for seasonality), waste blow-outs and SAP vs shelf count differences.
In-depth guide βJust-in-time reorder points with safety stock, capped at shelf life so fresh lines don't turn into waste. Recalled items are blocked. Roster hours come from forecast sales Γ· a productivity target.
In-depth guide βMove the sliders to see the weekly and yearly gross-margin effect of a price change (with elasticity), a waste reduction or fewer delivery outage days.
In-depth guide βThe same engine applied to consumer electronics (stock cover, aged stock, markdowns, price erosion, attach), wholesale (cost-to-serve, debtors, OTIF), warehousing (ABC-XYZ slotting, pick rates, capacity) and production (OEE, scrap, schedule adherence).
In-depth guide βThe same engine runs industry packs for consumer electronics, wholesale, warehousing and production. Demo data; full reports in docs/example.
Products grouped by how they behave (speed, margin, volatility, waste, delivery share, promo response) using K-means, shown on 2 PCA axes. Stability scores show which groups are real and which are noise.
In-depth guide βFive models compete on each category: Naive, Seasonal Naive, Drift, SARIMA-X and XGBoost. Each is scored on the last 4 weeks (rolling origin) and the winner is used. A simple model is allowed to win.
In-depth guide βHow Shelly works end to end (CRISP-DM) and how it checks the data first: duplicates, refunds keyed as sales, missing values, calendar gaps and unknown SKUs, summarised in a quality score.
In-depth guide βShelly follows CRISP-DM: business targets β data-quality checks β preparation β five competing forecast models, anomaly detection, K-means segmentation, JIT reorder and labour planning β rolling-origin evaluation (cross-checked in R) β this page, an Excel workbook, SQL, Power BI and Tableau outputs.
All data here is synthetic, generated to behave like a small Auckland convenience store. No employer data is used.