One view of every branch and sales channel: how to ask AI about all your sales
Short answer
Most multi-branch companies keep sales in several places: an online store, one or more point-of-sale (POS) systems, an accounting system and spreadsheets. An AI assistant connected to only one of them can answer only part of a question, so a combined view needs the systems connected together, one set of definitions, and clear rules on who can see what. Start with the five questions you ask every week and check that each one can be answered from the systems you have.
Why your reports disagree
Ask three systems for last month's sales and you often get three numbers. Usually nobody is wrong. Each system answers a slightly different question.
The online store counts an order when the customer pays. The POS counts it when the cashier closes the bill. The accounting system counts it when someone posts the invoice. Before any AI can give you one answer, you need one definition for each word in the question.
| Word | How systems differ | What to decide |
|---|---|---|
| Net sales | Some systems take off discounts and returns, some do not. Some include delivery fees. | Which of these come off before you call it net sales? |
| VAT | The store may show prices with VAT while accounting records them without it. | One basis for every report, written on each report. |
| Returns and cancellations | A return may reduce the original day's sales in one system and show as a negative sale today in another. Cancelled and unpaid orders may stay in a report. | Which day does a return belong to? Are cancelled and unpaid orders in or out? |
| Order date | Date placed, date paid, date delivered or date invoiced. | The one date that every report uses. |
| Branch | "Riyadh, Olaya", "Olaya branch" and "RYD-01" can be the same place in three systems. An online order may belong to a "web" branch or to the branch that ships it. | One list of branches with one code each, and a rule for which branch an online order counts toward. |
| Product | The same product can have different codes or names in the store, the POS and the stock system. | One product code that links them. |
What you can ask once the systems are joined
These questions need more than one system. They are slow in a spreadsheet and easy to ask once the data is connected.
- Which branch sold the most last week after returns, and how does each branch compare with the same week last year?
- This month, what share of sales came from the online store and what share from branches, by product category?
- Which products sell well online but are out of stock in a branch?
- Which branch has the highest return rate, and on which products?
- Which invoices are more than 30 days unpaid, and in which branch did the sale happen?
- Where are we selling below the planned margin once discounts are counted?
You do not need every system for every question. Pick the five you really ask each week and start there.
Where an AI connected to one system falls short
Software vendors are adding AI to their own products, which helps for questions about that product. Salla's connector lets Claude or ChatGPT read your Salla store, and Salla's page gives examples such as recent orders, products near stock-out and a sales summary. Foodics AI answers questions about Foodics data such as sales, orders, discounts and inventory transactions, and its FAQ says it does not have loyalty points, customer names or Foodics Pay data. On 22 June 2026 Foodics also completed its acquisition of Norma AI, a plain-language analytics company whose technology is now part of the Foodics platform (Wamda).
Neither gives you a combined view. You can switch on several connectors in one chat: OpenAI's help article says ChatGPT can use several apps in one conversation, and Claude's page limits only the Free plan to one custom connector. But each connector has its own sign-in, its own permissions and its own definitions. The assistant joins the numbers itself, with no shared meaning for "net sales" or for branch names. Check any total it joins against a report you trust.
Options today
| Option | Good for | Watch out for |
|---|---|---|
| Manual exports to Excel or Google Sheets | A one-off or monthly check. No new tools to buy. | Someone repeats the export every time. Definitions drift between files, and files with customer data get emailed around. |
| A dashboard tool (Power BI, Tableau and similar) | Fixed reports that many people read the same way. | Someone must build and maintain the data model, every new question becomes a new report, and access rules are set up separately. |
| One AI connector per system | Quick answers about a single system. | Each covers one system, with its own sign-in, permissions and definitions. Nothing reconciles them. |
| A governed layer across systems | Questions that span the store, the POS and accounting, under one set of definitions, roles, approvals and logs. | Someone has to agree the definitions first, and it needs setup. Ask who runs it, what it logs and where the data is processed. |
Most companies use more than one of these. A spreadsheet is a fine way to test whether your definitions match before you build anything.
A step-by-step plan
- Write the five questions you ask every week. Note who asks each one (owner, finance, branch managers) and how often.
- List the systems that hold the numbers. Online store, POS in each branch, accounting or ERP, spreadsheets. Name one person who owns each system.
- Write one page of definitions. Net sales, VAT basis, returns, order date, cancelled orders. Use the table above as a start.
- Make branch and product names match. A small mapping table is enough: each branch and product with its code in each system.
- Decide who may see what. For example: branch managers see their own branch, finance sees every branch, only owners see margin. Use each person's own sign-in, not one shared administrator login.
- Start read-only with a one-week pilot. One or two branches and one online channel are enough.
- Check the answers against a report you trust. Compare three numbers, for example one day's sales, one branch's returns and one open balance.
- Split scheduled reports from questions asked on demand. A Monday summary for each branch manager is a scheduled report. "Why did Jeddah drop on Thursday?" is a question. Scheduling depends on the tool: ChatGPT's pricing page lists scheduled tasks on its Go, Plus and Pro plans, so check the plan and the vendor's page for the tool you use.
- Allow changes last. Prices, stock and invoices should change only after a named person approves, and every change should be logged.
Pitfalls
- Counting twice. If the POS posts a daily total to accounting, adding the two doubles your sales. Decide which system is the source of truth for each number.
- Online orders in two places. A website or delivery-app order may also be keyed into the POS by hand.
- Branch names that almost match. "Jeddah Tahlia" and "Tahlia Jeddah" become two branches in a report.
- Different dates. A sale just after midnight, or a payment settled days later, lands on different days in different systems.
- One shared login. If the assistant signs in as an administrator, everyone who asks gets administrator access.
- Trusting a total you have not checked. Compare with a known report for the first few weeks.
- Personal data. Customer names, phone numbers and addresses in sales data are personal data. Leave those fields out of the pilot unless you need them, and check with your legal adviser before you connect them. Saudi rules are published by SDAIA.
Keep it safe
- Read-only first. Allow changes only after you have watched the answers for a while.
- Access by person and role. A connector that signs in as each person shows that person's permissions in that system. Claude's Team and Enterprise plans work this way: after an owner adds a connector, each member connects individually. Whether a system can limit a user to one branch depends on the system, so check that before you rely on it.
- Approval before any change. Anthropic tells you to review tool approval requests carefully, and OpenAI says ChatGPT may ask for confirmation before write actions. Keep write access off until you need it.
- A record you control. Keep a log of who asked what, in which system, and when.
More in Is it safe to connect company data to AI? and AI that asks before it changes anything.
Questions people ask
Do we need to replace any of our systems?
No. Each system stays the place where its data is entered. The aim is to ask questions across them, not to move everything into one product.
Can we do this with Excel alone?
For a one-off check, yes. For a report you need every week, someone has to repeat the exports, and the definitions drift between files. A spreadsheet is a good way to test your definitions first.
Will the AI's numbers always be right?
No. It can join the wrong rows, misread a column or count something twice. Check totals against a report you trust, and ask it to show which systems and filters it used.
Which assistant should we use?
That depends on where your people already work and what your admin needs to control. See Claude, ChatGPT or Microsoft Copilot for company data.
Where Vootlabs fits
This is the case Vootlabs is designed for. It connects the systems a business runs on, such as the online store, the POS, accounting and spreadsheets, to AI tools like Claude, ChatGPT and Microsoft Copilot, so your team can ask questions across all of them, with access by role, approval before any change, and a full audit trail.
We are onboarding a limited number of companies in a design-partner programme, and we are looking for companies with several branches or sales channels. If that is you, we would like to hear which five questions you ask every week.
Sources
Plans, menu names and product features change. Check the live page before you act. All pages below were read on 17 October 2026.
- Salla: Connect Salla to AI
- Foodics Help Center: Foodics AI FAQ (article updated 6 May 2026)
- Wamda: Foodics completes full takeover of Norma AI (22 June 2026)
- Claude Help Center: Get started with custom connectors using remote MCP
- OpenAI Help Center: Connecting custom MCP servers in ChatGPT
- ChatGPT: Plans and pricing
- SDAIA: National Data Governance Platform (Saudi Personal Data Protection Law and regulations)