Why Prompt Engineering Decides Whether Sidekick Forecasts Anything Useful
Ask Sidekick "how is my inventory doing?" and you get a polite paragraph. Ask it the right way — with a time window, a lead time, a safety buffer, and a sort order — and you get a ranked purchasing decision you can act on before lunch. That gap between "polite paragraph" and "executable answer" is the entire game of using Shopify Sidekick prompts for inventory forecasting.
Most merchants writing about Sidekick focus on what it can do. This guide focuses on exactly what to type. The 18 prompts below are copy-paste templates organized across six forecasting jobs: reorder timing, seasonal forecasting, dead-stock identification, contribution-margin ranking, supplier lead-time planning, and vendor diversification. Each prompt includes the variables to swap, a realistic example output, and a "tweak" line that expands the prompt into adjacent questions you will want to ask next.
If you need the capabilities walkthrough first — daily routines, morning check-ins, how Sidekick talks to your data — pair this post with our companion guide on how to use Shopify Sidekick for inventory management. This one assumes you already know the UI and want the prompt library.
The Six Prompt Categories Every Merchant Needs
Forecasting is not one job. It is six related jobs that each require a different prompt pattern. Muddling them together is why most Sidekick sessions drift into vague advice.
The forecasting job map
| Category | What you are answering | Frequency |
|---|---|---|
| Reorder timing | When do I need to place a PO for each SKU? | Weekly |
| Seasonal forecasting | How much should I buy before peak season? | Quarterly |
| Dead stock | What is not selling and costing me cash? | Monthly |
| Contribution margin ranking | Which SKUs actually make money after costs? | Monthly |
| Supplier lead-time planning | How does lead time change my reorder math? | Per supplier |
| Vendor diversification | Am I too dependent on one supplier? | Quarterly |
Every prompt below belongs to one of those buckets. Use the table to pick which job you are doing before you start typing.
The five variables you always supply
Sidekick cannot see outside the Shopify admin. That means every forecasting prompt needs you to supply the context Sidekick does not natively store:
- Sales window: "last 30 days," "last 90 days," or "same 30-day window last year"
- Lead time: in days, per supplier or per SKU
- Safety buffer: how many extra days of cover you want on top of lead time
- Cost basis: COGS, landed cost, or wholesale price (pick one and label it)
- Location filter: which warehouse, country, or retail location
Skip one and Sidekick guesses. Guesses become bad POs. For a wider view of where Sidekick's data access stops and where Pulse or the Inventory API takes over, Shopify's official Sidekick help documentation is the source of truth.
Reorder-Timing Prompts (Listicle)
Reorder timing is the single most expensive decision most merchants fumble. Order too late and you stock out during a sales spike; order too early and cash is stuck on pallets. These prompts close that window.
1. The base reorder calculation
"Using the last 30 days of sales, calculate days of stock remaining for every SKU in the [COLLECTION] collection. Flag any SKU with fewer than [LEAD_TIME + 10] days of cover. Return a table sorted by days-of-cover ascending, with columns: SKU, current stock, 30-day units sold, average daily sell-through, days of cover, reorder recommended (Y/N)."
Example output: a 14-row table showing three SKUs at red (under 7 days), four at yellow (under 21), and the rest green. The bottom row totals units needing reorder.
Tweak: swap "30 days" for "60 days" when you are approaching a seasonal ramp, or change COLLECTION to a vendor name to reorder by supplier.
2. The multi-location reorder split
"For my top 20 products by 60-day revenue, show current stock at each location. For any SKU below 14 days of cover at a single location, suggest a transfer quantity from the location with surplus. Assume I can transfer in 3 business days."
Sidekick returns a transfer plan before it returns a PO plan, which is usually the cheaper move. According to Ecorn's 2026 inventory forecasting guide, transfer-first policies reduce emergency reorder costs by double digits for multi-warehouse merchants.
3. The fast-mover early warning
"Find products whose last 7-day sell-through is more than 2x their trailing 30-day average. Rank by absolute unit velocity. For each, calculate days of cover using the 7-day rate, not the 30-day rate."
This prompt is your viral-product detector. A TikTok mention usually shows up in 7-day rate before your standard 30-day dashboards notice it. Tweak by replacing "2x" with "1.5x" to catch softer lifts earlier.
4. The variant-level gap finder
"For my 10 best-selling products by units, show which variants are in stock and which are out of stock by size or color. Flag any where the out-of-stock variant accounts for more than 15% of historical sales for that product."
Variant-level stockouts silently kill conversion. This prompt surfaces them in one table instead of drilling into each product page.
Seasonal Forecasting Prompts
Seasonal prompts have to lean on year-over-year data. If your store is less than 13 months old, swap historical year comparisons for proxy data — an industry benchmark or a comparable product category.
5. The year-over-year buy plan
"Compare unit sales by product for the period [START_DATE] to [END_DATE] this year versus the same window last year. For any product up more than 20% YoY, calculate a recommended buy quantity to cover the next 90 days assuming the trend continues and my supplier needs [LEAD_TIME] days."
This is the quarterly planning workhorse. Always include the explicit date range — Sidekick handles relative dates ("Q4 last year") but gets more reliable with ISO dates.
6. The peak-week concentration check
"For the [HOLIDAY/EVENT] period last year, show what percentage of that month's total revenue happened in the peak 7-day window. Apply the same concentration to this year's projected month to estimate peak-week units needed."
Black Friday / Cyber Monday is the classic case — a month's sales can compress into five days, which breaks standard 30-day averaging. Shopify's BFCM planning resources document that concentration repeatedly in case-study form.
7. The pre-season ramp buy
"I expect my [CATEGORY] sales to triple in [MONTH] based on last year. Current inventory is [X] units of the top 5 SKUs. Calculate how many units I need to buy and by what date to avoid a stockout if my supplier lead time is [LEAD_TIME] days and I want a 14-day safety buffer after the season starts."
This prompt turns a vague "stock up for summer" conversation into a dated PO plan with safety buffer baked in.
Dead Stock Identification Prompts
Dead stock is the hidden tax on most merchant P&Ls — capital stuck on shelves that should be moving, reducing, or liquidating. Sidekick will not volunteer this information. You have to ask.
8. The 90-day no-movement scan
"List every SKU with zero or fewer than 3 units sold in the last 90 days, where current stock is at least 10 units. Include columns for last sale date, current stock, and cost tied up (current stock × cost). Sort by cost tied up descending."
Example output: a ranked list of cash-draining SKUs. The top entry in most stores is a product the founder personally loved. That is not coincidence.
Tweak: change "90 days" to "180 days" for a more aggressive cull, or filter by collection to audit one product line at a time.
9. The slow-mover vs. seasonal-dormant split
"For SKUs that sold fewer than 5 units in the last 90 days, show how they performed in the same 90-day window last year. If last year's units were more than 10x higher, label as 'seasonal'. Otherwise label as 'true slow-mover'."
This prompt stops you from liquidating a Christmas SKU in April. According to PageFly's dead stock management guide, the seasonality misread is the most common merchant mistake when running a dead-stock report.
10. The markdown and clearance prompt
"For the 20 SKUs with the highest cost tied up from my dead-stock list, suggest a markdown percentage that would move the inventory in the next 30 days based on price elasticity observed in past promotions for that collection."
Sidekick's elasticity estimate will be rough, but it is a starting point. Review before publishing the sale.
Contribution-Margin Ranking Prompts
Revenue ranking lies. A product can top your best-seller list and still lose money after ad spend, payment fees, and pick-and-pack labor. Contribution-margin ranking fixes that.
11. The true profitability pull
"For the last 90 days, calculate contribution margin per SKU as (selling price − cost − [FEE_PERCENTAGE]% fees − [FULFILLMENT_COST] per order). Rank the top 25 SKUs by total contribution margin, not units or revenue. Include columns for units sold, revenue, contribution margin per unit, and total contribution margin."
Pass in your real payment fee percentage and fulfillment cost per order. Sidekick does not know them. This prompt is almost always a wake-up call — merchants routinely find their #3 revenue product is their #18 profit product.
12. The margin-weighted reorder priority
"From yesterday's reorder list, re-rank the SKUs by contribution margin × days-of-cover-risk. Show the top 10 to reorder first if I only have budget for 50% of the list."
This is the cash-constrained version of reorder timing. When you cannot buy everything flagged as low stock, this prompt tells you what to buy first.
13. The low-margin audit
"Show me SKUs where contribution margin per unit is under [$THRESHOLD] or under [X]% of selling price. Include units sold in the last 90 days so I can see how much volume is tied to thin-margin product."
Thin-margin, high-volume SKUs are fine when they are loss-leaders into a funnel. They are poison when they dominate your warehouse. Bloomreach's ecommerce margin analysis covers why margin-first ranking changes procurement priorities.
Supplier Lead-Time Planning Prompts
Lead time is the variable most merchants treat as a constant. Treating it as a variable is where real forecasting starts.
14. The lead-time stress test
"For my top 30 SKUs by revenue, recalculate reorder-by dates under three scenarios: current lead time [X] days, delayed by 14 days, and delayed by 30 days. Return a table showing which SKUs would stock out under each scenario and the revenue impact."
This prompt prepares you for port delays, Chinese New Year, and customs inspections before they happen. The same structure adapts to tariff shocks — substitute "lead time" with "landed cost" and see where margin compresses.
15. The sourcing-switch simulation
"If I moved all [CATEGORY] production from [COUNTRY A] to [COUNTRY B] and lead time changed from [X] to [Y] days, recalculate my safety stock requirement for the next 12 months based on the last 12 months of sales. Show the increase or decrease in working capital tied up."
16. The MOQ and lead-time trade-off
"For SKUs with supplier MOQs over 500 units, calculate how many months of current sell-through that MOQ represents. Flag any where MOQ is more than 6 months of stock. Suggest whether to negotiate MOQ down, split the PO with another merchant, or drop the SKU."
Moosend's ecommerce inventory guide covers the MOQ-vs-lead-time trade-off in more depth — this prompt is the Sidekick-native version of that analysis.
Vendor Diversification Analysis Prompts
Single-vendor concentration is a silent risk on almost every ecommerce balance sheet. These prompts quantify it.
17. The vendor concentration audit
"Group my last 12 months of inventory purchases by supplier. Show each supplier's percentage share of total COGS and their percentage share of my top 50 SKUs by revenue. Flag any supplier over 40% of COGS or over 50% of top-SKU coverage as concentration risk."
Example output: a short table where one or two suppliers usually appear in red. The next prompt is how you start unwinding the concentration.
18. The backup-supplier mapping
"For SKUs sourced from my top supplier by revenue, list any SKU I have ordered from a secondary supplier in the last 18 months. For SKUs with no backup supplier, flag them as single-source. Rank single-source SKUs by revenue so I know which to re-source first."
This prompt is the foundation of a disaster plan. You cannot diversify what you have not mapped. Pair it with your broader product management routines to turn the list into quarterly sourcing goals.
How to Tweak and Iterate Prompts That Almost Work
A prompt rarely lands the first time. Iteration is cheap inside Sidekick — the whole thread persists, so you refine in follow-ups instead of rewriting from scratch.
The three-move iteration pattern
- Narrow: "Now restrict this to my US location only."
- Re-sort: "Sort by contribution margin per unit instead of revenue."
- Export: "Turn that into a CSV I can paste into Google Sheets."
Most forecasting questions land in two or three moves. When you catch yourself on move five, the original prompt was probably fighting the wrong question — stop and rewrite it from scratch rather than layer more filters.
Best practices vs. common mistakes
| Do this | Avoid this |
|---|---|
| Specify exact sales windows ("last 30 days") | Vague windows ("recently," "this season") |
| Label cost basis (COGS, landed, wholesale) | Letting Sidekick assume which cost you mean |
| Ask for tables with explicit columns | Accepting paragraph answers for numeric data |
| Re-run the same prompt weekly | Running a one-off and calling it forecasting |
| Supply lead time and safety buffer | Asking "when should I reorder?" with no inputs |
Re-running the same prompt on a schedule is the one habit that separates merchants who actually forecast from merchants who talk about forecasting. The Replo team documented this cadence discipline in their breakdown of AI-assisted operations for Shopify stores.
Common Mistakes That Break Forecasting Prompts
Every merchant who starts prompting Sidekick for inventory hits the same four walls. Knowing them in advance saves weeks.
Mistake 1: Asking about external data
Sidekick cannot read your supplier's portal, your 3PL's spreadsheet, or your ad platform's demand signal. Any prompt that implicitly requires that data produces a confident-sounding but fictional answer. Fix: paste the external data into the prompt itself, or move the analysis to Pulse or a dedicated forecasting app.
Mistake 2: Mixing units and revenue without saying so
"Top products" is ambiguous. "Top products by units" and "top products by revenue" can return completely different lists. Always specify the ranking metric explicitly.
Mistake 3: Forgetting time-zone and window edges
"Last 30 days" starts when you send the prompt, not at midnight UTC. For month-end reports, pin the window to ISO dates: "from 2026-03-01 to 2026-03-31 inclusive." Optimonk's Shopify analytics guide covers the window-alignment gotcha in more detail.
Mistake 4: Treating one prompt as a report
A single prompt answers one question. A report is a set of prompts run in sequence and compiled. Write a four-prompt morning routine, run it daily, and paste the outputs into an ops doc. That is what "AI forecasting" actually looks like in practice.
When Sidekick Runs Out of Road
Sidekick is a conversational layer on top of your admin. It is not a dedicated demand-forecasting engine. There are three moments you should stop prompting and hand off:
- Statistical forecasting (ARIMA, Prophet, seasonality decomposition): move to a forecasting app, Pulse reports, or a notebook pulling the Inventory API.
- Bill-of-materials (BOM) and assembly logic: Sidekick does not natively understand that Product A consumes 2 units of Component B per sale. A manufacturing app does.
- Supplier negotiation documentation: Sidekick can draft the email. It cannot track the commitment. Use a vendor management tool.
For the handoff patterns and exact Pulse prompts that extend this list, see our Shopify Sidekick Pulse walkthrough. And if you are comparing Sidekick against bolt-on AI tools for the same job, the Sidekick vs third-party AI apps comparison lays out the trade-offs.
Start With One Prompt, Not Eighteen
The fastest way to get value from this library is not to adopt all 18 prompts on day one. Pick one prompt — usually Prompt 1 (the base reorder calculation) — and run it every Monday morning for four weeks. The muscle memory of specifying sales window, lead time, and safety buffer will transfer to every other prompt in this guide.
Once weekly reorder prompts feel automatic, layer on dead stock (Prompt 8) and vendor concentration (Prompt 17). That three-prompt routine is already more forecasting discipline than most seven-figure Shopify stores have in place today.
Want to compare notes with other merchants running Sidekick forecasting routines? Drop into the Talk Shop community and share the prompt that moved the needle most for your store. And for the full picture of what the assistant can and cannot do before you deepen your prompt practice, revisit Shopify Sidekick's AI assistant overview.
Which of these 18 prompts will you run first — and what will you change about the way you buy inventory once the output hits your screen?
About Talk Shop
The Talk Shop team — insights from our community of Shopify developers, merchants, and experts.