How Restaurant Owners Can Leverage AI: Managing Food Delivery, Site Selection, and Reviewing Daily Reports—All Explained in One Article

A Practical Guide to QwenWork for Catering Business Owners

How to Use QwenWork

Running a catering business involves much more than just food production. Before peak dining hours, you need to check if your stores are open for business. During downtime, you must reply to negative reviews and calculate profits. When planning to expand, you have to scout locations and oversee renovations. Various pieces of information are scattered across different backend systems, and even if the owner spends time browsing through them all, it's not always easy to figure out what the top priority should be.

This article summarizes three catering applications of QwenWork: the Delivery Operations Assistant handles delivery management; the Site Selection Advisor integrates with Amap Cloud Map to evaluate locations; and the Keruyun Catering Operations Analysis Assistant reviews point-of-sale and business performance. It is recommended to first identify the specific problems you are facing and follow the corresponding steps, rather than installing the full suite of tools all at once.

QwenWork serves as the entry portal, while the Expert Suites handle specific tasks. Only by connecting your store accounts can you access the data you are authorized to view.

Step 1: Enable the Tools and Connect to Your Stores

On your computer, open qwenwork.cn/app/ and log in using your phone number or by scanning the QR code with DingTalk. The web version requires no installation; if you prefer the desktop client, you can download and install it from the "Download Desktop Version" link on the QwenWork web page. Desktop DingTalk users can also update to the latest version and access it from the "QwenWork" option in the left sidebar.

The Expert Suites are a set of execution tools, while the Connectors access other systems within the scope of your authorization. After installing the suites, you still need to complete the connection and authorization steps. For the first time, it is recommended to select a familiar store and check one day's data to confirm that the store name and figures are correct before expanding the scope.

For delivery: Connect "Taobao Flash Sale Catering Merchant Edition" and "Browser Automation," then install and check the "Delivery Operations Assistant."

For Amap: Connect "Amap Cloud Map" and "Browser Automation," then install and check the "Site Selection Advisor."

For Keruyun: Enable "Browser Automation," install and check the "Keruyun Catering Operations Analysis Assistant," and log in with an authorized account on the actual page. When merging Flash Sale data, be sure to use it in conjunction with the Delivery Operations Assistant.

Managing Delivery: 14 Scenarios, Choose as Needed

The following scenarios all use the "Delivery Operations Assistant." Please replace the bracketed content in the prompts with your own store name, date, product, or group name.

01 It's Opening Time: Which Stores Are Not Yet Open on the Platform?

The kitchen is preheated, and the staff are all in place, but the notification sounds from the delivery platform are yet to come.

Prompt: Help me check today's operating status for [Store Scope]. Based on each store's individual business hours, identify stores that should be open by now but are not, excluding normal rest days. List the store name, city, current status, and duration of the anomaly; if the start time of the closure cannot be verified, please state this explicitly and do not estimate. First, just check and list them.

The top priority is to ensure customers can place orders; only then should you consider how to increase sales.

02 Why Do Stores of the Same Brand Have Different Minimum Order Amounts and Delivery Fees?

Customers select their meals, only to find at checkout that they are far short of the minimum order amount, which might cause them to leave.

Prompt: Compare the business hours, minimum order amount, delivery fee, and free delivery threshold of [Store Scope] against the [Headquarters Standard]. If there is no headquarters standard for an item, refer to the settings of the majority of stores and note this. Output the stores with issues, current values, benchmark values, and inconsistent items. First, just check and do not make modifications.

Clarify the differences first, then decide which settings need to be unified.

03 With So Many Negative Reviews, Which Ones Should Be Prioritized Today?

Owners worry most about two situations: severe complaints hidden among a pile of reviews, and ordinary negative reviews left unaddressed for days.

Prompt: Query the new negative reviews in the past 30 days for [Store Scope], as well as currently unreplied negative reviews. Categorize by store, rating, and issue type, listing the review content, time, and reply status. Provide a list of those I need to prioritize, with food safety-related feedback pinned to the top and marked as pending verification. First, just check; do not auto-reply.

Reviewing negative reviews aims to find the issues that must be resolved today.

04 How to Handle Backlogged Negative Reviews One by One?

The lunchtime rush has just passed, and a new batch of reviews has accumulated in the backend.

Prompt: Filter all unreplied negative reviews for [Store Scope], categorize them by food quality, missing/wrong items, delivery time, packaging, and pricing/promotions, and draft a reply for each based on the original text. List food safety-related feedback separately, mark it as pending verification, and investigate the situation offline first. Do not fabricate compensation promises. Show a preview first; after I confirm, reply in bulk and output the results.

You can process them in bulk, but you shouldn't treat every customer's issue as the same.

05 Sales Are Down This Week: Which Data Should the Owner Check First?

Checking daily revenue only tells you whether you sold more or less.

Prompt: Generate this week's operations weekly report for [Brand or Store], specifying the statistical time range, counting only up to yesterday and excluding today's unfinished business data. Summarize revenue, income, order volume, exposure, store visit and order conversion rates, and average spend per customer, comparing them with the same period last week, yesterday, and industry averages. Explain the main changes, priority areas for investigation, and missing data in plain language.

The value of a weekly report is to help you choose the first task to tackle next week.

06 Feeling Like Business Is Slowing Down: Is It Really Because Customers Visit but Don't Buy?

Conversion rate sounds complicated, but it simply means "the proportion of people who move to the next step."

Prompt: Analyze the business changes for [Store Scope] during [Time Period]. First, verify if the order conversion rate has indeed dropped, then check one by one: which cities, regions, and stores have seen a significant drop; which products have decreased in sales; whether there have been changes in prices or promotions; whether there are out-of-stock items or slower delivery; and what customers are mainly complaining about in their reviews. Tell me the main issues, corresponding data, and recommended initial actions. Please note where data is missing; first, just analyze without adjusting operational settings.

Prove where the problem lies first, then decide where to allocate funds and manpower.

07 Dishes That Should Be Sold Out Are Still Selling, While Restocked Ones Show as Sold Out. What to Do?

Dishes lacking ingredients are still listed online, making it easy to receive orders that cannot be fulfilled; dishes that have been restocked still show as sold out, causing orders to be lost in vain.

Prompt: In [Designated Stores], take down [Products to be Discontinued] and restore [Confirmed Restocked Products] to appropriate inventory levels. First, list the stores, products, specifications, current values, target values, and restricted items, and give me a preview of the changes. Execute only after I confirm, and return the details of successful, failed, and unexecuted items.

The dishes available for order on the menu must match what the kitchen can actually prepare.

08 How to Unify Price Increases Across Multiple Stores Without Missing Any Store or Changing the Wrong Dish?

When multiple stores change the price of the same signature dish separately, it is easy for some to miss the change or change the wrong specification.

Prompt: Uniformly increase the price of [Specific Products or Combos] in [City and Store Scope] by [Amount] yuan. First, list the stores, products, current prices, and adjusted prices, while checking for promotion conflicts and explaining any items that cannot be executed or are unclear. Execute in bulk after I confirm, output the details of successful, failed, and unexecuted items, and keep a backup of the original prices.

Identify the targets before adjusting prices, and verify the results one by one during execution.

09 Before Opening in the Morning, How to Understand Yesterday's Business in a Few Minutes?

Problem solved: The owner browses multiple pages every day but still cannot clearly explain what went well and what went poorly yesterday.

Prompt: Every morning at [Time], summarize yesterday's business performance for authorized stores. Compare revenue, valid order count, average spend per customer, and order conversion rate with the previous day and the same day last week. List exposure, store visits, order count, and the two conversion rates, pointing out the weakest link. List the top three stores by revenue, the 1-2 stores that fell behind the most, as well as medium/negative reviews, unreplied reviews, and rating changes. Write in short sentences that store managers can easily understand. First, generate a draft; after the DingTalk connector is authorized and the [Receiving Group Name] is confirmed, send it as scheduled.

The purpose of reading the morning report is to decide which store to manage first today and which task to tackle first.

10 How to Detect Dropped Orders, Refunds, and Food Preparation Backlogs as Early as Possible?

When orders are backlogged, the food is not yet ready, but delivery riders are already crowding in the store waiting to pick up.

Prompt: Every day at [Check Time], check yesterday's data for authorized stores, and send an alert if any of the following conditions are met: 1. Revenue or order count drops by more than 20% compared to the previous day or the same day last week. 2. Order conversion rate is 3 percentage points or more below the 7-day average. 3. Refund rate reaches 10% or above; stores with fewer than 20 orders yesterday are excluded from this criterion. 4. Rating is below 4.5, or at least 3 new medium/negative reviews are added in a single day. 5. The ratio of the store's actual receipt to the user's actual payment is below 70%. During [In-Operation Check Time], additionally check: stores that should be open but are not, excluding rest days and planned suspensions; orders in transit that have exceeded the preparation time or will exceed it within 5 minutes. List the actual figures, triggered rules, and initial investigation direction by store. Pin food safety feedback, widespread closures, and order backlogs to the top; report all clear if normal. First, provide a draft; after completing DingTalk authorization and confirming the [Receiving Group Name], send it on schedule.

Alerts must include figures, store names, and next steps, so they don't just become noise in the group chat.

11 When Weekly Business Rises and Falls, Who Is Actually Playing the Key Role?

Problem solved: Dropping prices as soon as revenue falls, and adding promotions as soon as it rises, without knowing the real reasons.

Prompt: Review the last two weeks every Saturday. By default, look at the order conversion rate, first verifying the number of store visits and orders; when revenue fluctuates by more than 15%, focus on revenue, then break down contributions by exposure, store visit conversion, order conversion, and average spend per customer, keeping definitions consistent. Trace the stores and links with the biggest impact; summarize all medium/negative reviews, distinguishing between merchant-side and delivery-side issues with typical original texts; check the effect of subsidies and the abnormal order rate and preparation timeout rate on peak days. Give the top 3 priorities for next week. First, review the draft; after completing DingTalk authorization and confirming the [Receiving Group Name], send it on schedule.

Find the most impactful cause first, then arrange next week's actions; don't just slash prices across the board at the first sign of a drop.

12 Customers Clearly Paid a Lot, So Why Is the Store's Actual Take-Home So Little?

Problem solved: Orders look bustling, but the month-end account settlement is less than ideal.

Prompt: Every morning at [Time], reconcile the total actual amount paid by users and the total actual amount received by stores for all authorized stores yesterday. Only recognize the explicitly stated "Store Actual Receipt" in the backend; do not substitute it with amounts of unconfirmed meanings. Calculate the actual receipt/actual payment ratio for each store and the brand as a whole, and compare it with the previous day. List stores below 70% separately; do not judge stores with fewer than 20 orders yesterday. List the top 5 orders with high actual payment and low take-home ratio, detailing how much was paid, how much was received, and what was purchased. Every Wednesday, additionally check: ongoing product promotions, delivery fee reductions, targeted red envelopes, and platform invitation opportunities. Sample a few high-amount, low-ratio orders, detailing how much of the discount is borne by the platform, the merchant, and the red envelope, to find stores with abnormal merchant burdens; determine if there is evidence that subsidies brought in new orders, and note it if evidence is insufficient. Read-only, do not change promotions or prices, and do not display customers' personal information. If data retrieval fails, retry once; if it still fails, log it in the uncovered list without affecting other stores. First, output a draft; after completing DingTalk authorization and confirming the [Receiving Group Name], send it on schedule.

First, calculate exactly how much is taken home, then check who bears the discounts, and finally decide how to adjust promotions.

13 Among the Dishes on the Menu, Which Should Be Promoted More and Which Should Be Changed First?

Problem solved: The menu is long, but the store manager just picks a few dishes to discount based on gut feeling.

Prompt: Every morning at [Time], analyze yesterday's exposure, store visits, orders, and the two conversion rates for authorized stores, comparing them with the 7-day average to find the weakest links and underperforming stores. Every Friday, check dish exposure, clicks, click-through rate, sales volume, and sales revenue, categorizing them into 4 types: maintain those with high exposure and conversion; for high exposure but few orders, check the main image, price, and reviews; for low exposure but good conversion, list as potential items and suggest increasing exposure; for low exposure and low sales, list as slow-moving candidates. Additionally, list dishes with high exposure but low clicks, prioritizing checking the main image; for dishes with many clicks but few orders, prioritize checking price and reviews. Only provide analysis and suggestions; first, output a draft, and send it after DingTalk authorization and confirming the [Receiving Group Name].

First, see where customers are getting stuck, then decide whether a dish should be promoted, changed, or kept under observation.

14 Now That the Data Is Reviewed, What Exactly Should the Store Manager Do First Today?

Problem solved: There are reports and alerts in the group chat, but the store manager doesn't know which one to do first.

Prompt: Every morning at [Time], organize yesterday's data for authorized stores into "What the Store Manager Should Do Today." For each store, first summarize the status, then assign tasks: 1. Items to improve in platform business health, sorted by platform weight, with official benchmark standards; if standards cannot be found, mark as pending verification and do not fabricate them. 2. Number of medium/negative reviews and main complaints. Pin food safety feedback to the top, handle offline first, and do not use template replies; note delivery-side responsibilities. 3. Rating factors below 4.2; do not draw rating conclusions if there are fewer than 5 reviews for the item, but severe feedback still needs to be handled. 4. Volume of canceled, refunded, and preparation-timeout orders; zero sales, out-of-stock, or delisted main dishes; which link is weaker between store visits and orders. First, schedule tasks that must be handled that day, then those affecting this week's conversion, and finally minor tasks that can be done on the go. Attach data basis, specific actions, and completion criteria for each item. First, provide a draft; after completing DingTalk authorization and confirming the [Receiving Group Name], send it on schedule.

A qualified briefing should allow the store manager to take action immediately after reading it and verify each item one by one when closing the store.

Amap Site Selection: Compare Two Locations First, Then Verify

Suppose you are preparing to open a noodle shop, with one candidate location near office buildings and another near a residential area. First, prepare the specific address, area, monthly rent, as well as the category, target customers, average spend per customer, and budget. Just saying "help me find a good location" will hardly yield a useful comparison.

The Site Selection Advisor will organize the analysis from five aspects: whether there is heavy foot traffic nearby, whether they are your target customers, whether there are many similar stores, whether transportation is convenient, and what the surrounding commercial facilities are like. What you need to see is not just a score, but also the basis for each judgment.

Prompt: Use the "Site Selection Advisor" to compare [Address A] and [Address B]. I am in the [Category] business, my main customers are [Demographics], and the average spend per customer is about [Amount]. Please combine the available data from Amap Cloud Map to compare the customer base, competition, transportation, and commercial facilities, and then list the pros and cons combined with the rent. Note the sources and time; for items that cannot be found, write "to be supplemented" and list things that need to be verified on-site.

After getting the table, visit the site separately during lunch and dinner: see where the people come from, whether the entrance is easy to find, and whether it is convenient for customers to enter the store. Add your observations to make the analysis closer to reality. If assumptions are used for calculating foot traffic, transaction ratio, and revenue, they cannot be treated as guaranteed income.

After selecting the location, you can also use the Multi-user Workspace feature to follow up on renovations, licenses, and expenses. For small stores, four items are enough: progress, bottlenecks, licenses, and expenses. Clearly write down the person in charge, planned date, and current status for each item, so the owner can see who is handling it and where it is stuck.

Prompt: Use the "Site Selection Advisor" and Multi-user Workspace feature to set up a preparation workspace for [New Store]. First, collect the person in charge, planned opening date, budget, and license status, then organize the four sections: progress, bottlenecks, licenses, and expenses. Remind me to fill in missing information; do not guess that it is already completed.

Alert conditions like a 7-day delay or 10% overspend in the original case can be adjusted according to your own situation. They are management rules, not industry standards. The progress in the workspace also needs to be updated by the person in charge; creating a table does not mean the work is already being pushed forward.

Keruyun Daily Report: Look at Dine-in and Delivery Together, Align the Numbers First

Looking only at delivery orders makes it easy to ignore the other half of the store's business. The Keruyun Catering Operations Analysis Assistant can help organize point-of-sale operational data; combined with the Delivery Operations Assistant, it supplements the analysis of traffic, conversion, and promotions on the Flash Sale platform.

For the first time, select a familiar store and one day's data. Let it list the stores, statistical time, sources, and missing items. After verifying with the backend, expand to more stores. If a store is missing, check permissions first; do not write "all stores" in the report.

Prompt: Use the "Keruyun Catering Operations Analysis Assistant" and "Delivery Operations Assistant" to analyze the business of [Store Scope] on [Date]. First, verify the data sources and statistical scope to avoid double-counting delivery revenue. Organize total operations, channel composition, store performance, promotion effects, abnormal changes, dish gross profit, and payment reconciliation, and finally give the top three priorities for today. Explain if data is missing; do not modify prices, promotions, or financial records for now.

Look at four things in the report first: how much was sold yesterday, which stores had big changes, which dishes need attention, and what to do first today. Small stores can use the simplified version first; when there are many stores and complex issues, add detailed content such as delivery metrics, conversion, and repurchase.

When merging the two systems, the most important thing is to avoid double-counting. Keruyun may already include delivery revenue, so you cannot just add the Flash Sale amount directly. The meanings of gross revenue, operating income, and store actual receipt must also be distinguished. List uncertain amounts separately; do not force a merge just to make up a total number.

Dish gross profit also requires reliable cost data. You cannot just take the selling price minus commission as the complete profit. If you see a combo with high sales volume but low gross profit, first verify the ingredient costs and who bears the discounts before deciding whether to adjust. The 30% gross profit alert line in the original case is not applicable to all catering stores.

Suppose delivery orders have decreased, but dine-in has not dropped simultaneously; you should first investigate the delivery channel rather than adding promotions for the whole store. The reasons proposed in the report must also be distinguished: which are supported by data and which are just guesses pending verification.

Final Step: Get It Right Once, Then Execute Daily

First, manually generate a report and verify the stores, time, and data sources. Once satisfied, set it to execute daily.

Prompt: Set the previous task to execute every day at 8:00 AM Beijing Time, counting the same batch of stores from the previous day, and save the results in the conversation first.

To send it to the store manager, configure the DingTalk connector and specify the group name.

Run it once immediately after creation to see if the report can be generated and if the message is received. When local browser operations are involved, the computer needs to stay awake and connected to the internet. Do not treat unfindable data as zero, and do not treat speculated reasons as confirmed conclusions.

The owner doesn't need to run all the features every day. Before opening tomorrow, pick the one thing that gives you the biggest headache, copy the corresponding paragraph into QwenWork, and try it once.

Leave the backend to it, and keep the customers for yourself.

We dedicated to serving clients with professional DingTalk solutions. If you'd like to learn more about DingTalk platform applications, feel free to contact our online customer service or email at This email address is being protected from spambots. You need JavaScript enabled to view it.. With a skilled development and operations team and extensive market experience, we’re ready to deliver expert DingTalk services and solutions tailored to your needs!