Prompt library

Copy practical prompts for 15 common commerce decisions. Every pack includes at least three scenarios and the workflow an assistant should follow.

These prompts are written in business language, not tool syntax. Replace placeholders such as {shopName}, {period}, and {productName} when you know them. If you do not specify a shop, the assistant should discover the shops available to your DataGlass account before continuing.

If you are new to the connector, use the interactive prompt lab to build a question and preview its tool route. The worked examples then show six complete answers with evidence and quality checks.

Start with one question. Ask for a deeper breakdown only when the first answer identifies a useful lead; this keeps answers focused and avoids unnecessary data calls.

Most of these prompts are answered by reading alone. A few end in a proposed change — a target ROAS, a price move, a reorder. Those only execute if your connector holds that domain's write scope and you approve the specific change; see Permissions and safety.

1. Get started and discover shops

Native playbook: getting-started

Workflow: retrieve get_playbook when guidance is needed, then use one or more of list_shopee_shops, list_lazada_shops, and list_tiktok_shops.

ScenarioPrompt
First connectionShow me all of my connected shops and group them by marketplace.
Choose a shopWhich of my shops can I analyze, and what region and currency does each use?
Understand coverageWhat can DataGlass analyze for my Shopee, Lazada, and TikTok shops?
Selected workspaceGive me a quick guide to the data available for {shopName}.

Good answer: show names and platforms first. Keep raw IDs as supporting metadata unless the user requests them.

2. Check shop health

Native playbook: shop-health-check

Workflow: discover the shop, then call diagnose_shop. Follow a signal's suggestedTool only when more evidence is needed. Use get_financial_overview for an all-shops rollup.

ScenarioPrompt
Selected shop snapshotHow is {shopName} doing over the last 30 days? Flag anything that needs attention.
Week-over-week checkGive me a health check for {shopName} this week versus the previous week.
Risk scanIs anything wrong with {shopName} right now? Focus on profit, GMV, margin, COGS coverage, and loss-makers.
Portfolio viewSummarize realized profit and margin across all of my shops for {period}.
Executive summaryGive me the three most important things I should know about {shopName} today.

Good answer: lead with a one-line conclusion, label warning or critical signals, and avoid dumping full product tables unless asked.

3. Explain a profit, GMV, or sales drop

Native playbook: why-profit-dropped

Workflow: start with diagnose_shop, then use the relevant product ranking, breakdown, cost, COGS, or ads tools. Compare with the immediately preceding equal-length period unless the prompt names another baseline.

ScenarioPrompt
Profit declineWhy did realized profit drop for {shopName} in {period}? Rank the biggest drivers.
Sales declineSales are down for {shopName}. Tell me whether the cause is fewer orders, lower revenue per order, ads, or product mix.
Margin compressionWhy is margin lower for {shopName} even though GMV is stable? Break down the cost lines that changed.
Product moversWhich products contributed most to the profit change for {shopName} in {period}?
Data-quality checkCould the apparent profit drop for {shopName} be caused by missing COGS or settlement lag?

Good answer: separate real business movement from missing-cost or settlement artifacts, and state the strongest driver first.

4. Improve low-margin or loss-making products

Native playbook: improve-low-margin

Workflow: use the platform financial overview and product breakdown or ranking, confirm get_product_costs, run the relevant profit simulation, then retrieve margin-defense or price-buffer recommendations when useful.

ScenarioPrompt
Find loss-makersWhich products are losing the most money in {shopName} over {period}, and why?
Thin-margin reviewFind products in {shopName} with dangerously low margin and rank them by profit impact.
Remediation optionsFor the three worst-margin products in {shopName}, compare a price increase, a smaller discount, and a COGS reduction.
Cost validationBefore recommending changes, verify whether the worst products have explicit or estimated COGS.

Good answer: do not recommend a price move based on inferred COGS without clearly stating the uncertainty.

5. Diagnose and optimize ads

Native playbook: debug-ads

Shopee workflow: use get_shopee_ads_performance, True or profit-adjusted ROAS, then the relevant ads, campaign-floor, target-ROAS, GMS, or boost recommendation tools.

TikTok workflow: use get_tiktok_gmv_max_campaigns, optional session drilldown, get_tiktok_campaign_roas, then budget or target-ROAS recommendations.

ScenarioPrompt
Weak campaignsWhich ad campaigns in {shopName} have the weakest true ROAS in {period}?
Hidden lossesFind campaigns that look good on marketplace ROAS but lose money after COGS, fees, cancellations, and returns.
Budget allocationWhere should I increase, hold, or reduce ad budget for {shopName}? Explain the profit evidence.
Bid or target guidanceShow the recommended bid or target ROAS for the weakest campaigns in {shopName}.
TikTok drilldownReview the GMV Max campaigns and product sessions for {shopName} and identify wasted spend.

Good answer: optimize on True or profit-adjusted ROAS, not headline ROAS. A recommendation is a proposal until someone applies it — in DataGlass, or through a tool such as apply_shopee_target_roas when PAT_ADS_WRITE is granted and the user has agreed to the specific change.

6. Defend margin from commission changes

Native playbook: defend-margin-commission

Workflow: retrieve the platform commission overview and impact, then the matching margin-defense actions and a simulation for selected products.

ScenarioPrompt
Upcoming increaseHow much will the upcoming commission change cost {shopName}, by product?
Preserve target marginWhat price or discount changes would preserve margin after the commission increase for {shopName}?
Prioritize actionRank the products in {shopName} most exposed to the new commission rate and suggest the smallest safe adjustment.
Compare tacticsFor affected products in {shopName}, compare raising price with trimming discount before I decide.

Good answer: include the effective date, current and new rate when available, profit impact, and the proposed margin-defense move. If the connector holds PAT_PRICING_WRITE, it can apply that move with apply_margin_defense_action — it should show the per-variant prices and get an explicit go-ahead first.

7. Plan inventory and restocking

Native playbook: plan-restock

Workflow: call get_inventory_alerts, get_reorder_plan, a forecast or demand chart, then get_open_purchase_orders to avoid double ordering.

ScenarioPrompt
Immediate riskWhich SKUs need restocking now, and how urgent is each one?
Reorder quantitiesBuild a reorder plan using current stock, reorder points, forecast demand, and inbound purchase orders.
Demand explanationExplain why {productName} needs its recommended reorder quantity using historical and forecast demand.
Overstock reviewWhich products appear overstocked or slow-moving, and what evidence supports that?

Good answer: subtract or explicitly show inbound stock and distinguish open alerts from resolved alert history.

8. Audit COGS and margin confidence

Native playbook: check-cogs

Workflow: read shop financial COGS coverage, resolve canonical products with list_products, use get_product_costs, and retrieve get_estimated_margin for Shopee shop-level inference when needed.

ScenarioPrompt
Missing costsWhich products in {shopName} are missing explicit COGS?
Profit confidenceHow much of {shopName}'s profit in {period} depends on estimated rather than entered COGS?
Cost auditVerify the stored cost basis for {productName} and explain how it affects margin.
Prioritize cleanupRank the missing-COGS products by how much they could distort reported profit.

Good answer: label explicit, inferred, and missing costs separately. Users enter costs in the DataGlass app, not through the MCP connector.

9. Simulate price, discount, or cost changes

Native playbook: simulate-price-change

Workflow: locate the product, read its current economics, then call exactly one of simulate_shopee_profit, simulate_lazada_profit, or simulate_tiktok_profit for each proposed scenario.

ScenarioPrompt
Price increaseIf I raise {productName}'s price by 5%, what happens to per-unit profit and margin?
Discount reductionCompare the current result for {productName} with reducing its discount from 15% to 10%.
COGS negotiationHow would {productName}'s profit change if unit COGS fell by {currency} 10?
Break-even targetWhat selling price would {productName} need to reach a 20% margin, based on its current fees and COGS?

Good answer: show current versus simulated values and list the assumptions. Never claim that the live listing was changed.

10. Analyze products and conversion

Native playbook: none yet.

Workflow: discover the platform shop, use a platform product list or search, then the matching conversion-metrics tool. Use fetch for a complete canonical product after search.

ScenarioPrompt
Product lookupFind {productName} across all of my connected shops and show its platform listings.
Funnel diagnosisWhich products in {shopName} have high traffic but weak conversion in {period}?
Platform comparisonCompare the conversion performance of {productName} across Shopee, Lazada, and TikTok where it is listed.
Discover opportunityFind products with improving conversion but low sales volume that may deserve more visibility.

Good answer: use YYYYMMDD windows for conversion-tool inputs, but display ordinary calendar dates to the user.

11. Review orders and fulfilment

Native playbook: none yet.

Workflow: use count_orders for a quick count, list_orders for rows, get_order for one normalized order, and count_unshipped_orders for backlog.

ScenarioPrompt
Volume checkHow many orders did I receive across all shops in {period}, broken down by platform?
Fulfilment backlogHow many orders are still waiting to ship, and how old is the backlog?
Order investigationFind the details and line items for order {order reference}.
Recent order listList my newest orders for {shopName} and highlight unusual totals or statuses.

Good answer: call the count tool before fetching a long list when the user asks only “how many.” Follow cursors only when more rows are necessary.

12. Reconcile marketplace payouts

Native playbook: none yet.

Workflow: discover the shop and use exactly one of get_shopee_payout, get_lazada_payout, or get_tiktok_payout for the requested payout or settlement dates. Do not substitute profit tools.

ScenarioPrompt
Amount paidHow much did {platform} pay {shopName} in {period}?
ReconciliationCompare DataGlass's expected payout with the marketplace-reported payout for {shopName}.
Investigate a deltaWhich orders or statements explain the payout mismatch for {shopName} in {period}?
Pending versus paidSeparate paid and pending settlement amounts for {shopName} in {period}.

Good answer: state whether the date is a payout or settlement date, show local currency, and treat a zero reconciliation delta as an exact match. Lazada exposes a net finance-statement settlement rather than the Shopee and TikTok reconciliation report.

13. Explore recommendations and growth opportunities

Native playbooks: debug-ads, improve-low-margin, and defend-margin-commission cover parts of this domain.

Workflow: begin with list_recommendations, then retrieve the typed recommendation tool that matches the decision.

ScenarioPrompt
Recommendation inboxWhat are the newest actionable recommendations across all of my shops?
Growth ideasWhich products should I boost or bundle to improve discovery and basket size?
Margin protectionShow price-buffer and margin-defense recommendations for {shopName}.
Ads actionsSummarize the highest-impact ads recommendations for {shopName} and the evidence behind each.

Good answer: state recommendation status, shop and platform, expected impact, and the evidence. Applying an Action happens in DataGlass, or through a tool such as rollback_recommendation or stop_recommendation when PAT_RECOMMENDATION_WRITE is granted — never without the user's explicit agreement to that specific change.

14. Troubleshoot connection or missing data

Native playbook: none yet.

Workflow: list the platform's shops. For Shopee, use list_shopee_shop_accounts to inspect authorization by app type. Explain authentication, access, plan, and data-timing failures without retry loops.

ScenarioPrompt
Missing shopWhy is one of my connected shops missing from DataGlass?
Missing Shopee dataCheck whether {shopName}'s Shopee ERP or Ads authorization has expired.
Empty resultI expected data for {shopName} in {period}. Check the shop, region, date format, authorization, and settlement timing.
ReconnectMy DataGlass connector stopped working. Tell me whether I need to reconnect, re-authenticate, or upgrade my plan.

Good answer: a 401 requires OAuth re-authentication; a 403 is an ownership or allowlist denial and should not trigger a re-authentication loop. A plan denial should link to DataGlass billing.

15. Explain query conventions

Native playbook: conventions

Workflow: retrieve get_playbook(playbook="conventions") when the model needs wire conventions, then translate the result into ordinary language.

ScenarioPrompt
Date helpWhat date range will you use for "last month," and what comparison period will you use?
Metric helpExplain the difference between realized profit, payout, headline ROAS, and true ROAS.
Identifier helpFind the correct shop and product identifiers for my request instead of asking me to look them up.

Good answer: human dates are YYYY-MM-DD, conversion datadate values are YYYYMMDD, timestamps are ISO 8601, and money uses the shop's local currency.

Prompt template

Use this structure when creating another prompt:

<business question> for <shop or all shops> over <period>.
Show <decision metric> and explain <desired evidence or next step>.
PlaceholderUse
{shopName}Human-readable shop name from the current workspace
{platform}Shopee, Lazada, or TikTok Shop
{period}Natural language such as “the last 30 days”
{productName}Product name or SKU; let the connector resolve IDs
{currency}The shop's local currency, when already known
{campaignName}Campaign name, or ask for the weakest campaigns