Conversational BI One question. Three-day wait.
Ask about sales or stock in WhatsApp. Get the figure, with its workings.We connect your apps, so staff ask about sales or stock in WhatsApp or Slack and get the figure, with its workings.
- Starts from the apps you already use
- Can ask which of two meanings you mean
- Can show how each figure was worked out
Sales means what?
One word. Two right numbers.
Ask for October’s sales and you might mean invoiced or paid. We set it to ask which.
What were sales in October?
S$61,400invoiced
S$48,200paid
S$13,200 apart
Did you mean invoiced or paid?
One question back. When a question could still mean two things, we set the tool to ask which before it answers.
Which word means two things in your business? Tell us one.
Get StartedHow answers look
Each answer can show its working.
We set each answer to open and show how it was worked out, so anyone can check it.
Ops numbersYou: Branch manager C
What were sales in October?9:41
Did you mean invoiced or paid?9:41
Paid9:42
October sales, paid
S$48,200
Worked out from
- ASales = paid invoices
- 1 to 31 Oct
- Read from: Xero
- BData as of 6:00 today
- A
What it counted. The meaning from the page you signed, the dates, and the app it read.
- B
How fresh it is. It can say when the data was last copied, so last night’s figure doesn’t pass for this minute’s.
- C
Who may ask what. We can set it so a branch manager sees only their branch.
Before you trust it
Three wrong answers. Three fixes.
A wrong figure in a bank meeting costs more than a late one.
-
Test orders counted.
ABC
1OrderStatusS$
2#1041Paid120
3#1042TEST85not a sale
4#1043Staff40not a sale
5#1044Refunded96not a sale
Your meaning sheet leaves them out.
-
One word, two meanings.
You meant: October
Nov
11831
Tool read: last 30 days
You sign the meanings. We set unclear questions to get one question back.
-
An outlet with no data.
Orchard sales: S$12,000I don’t have that yet.
We can set it to say so, and log the question.
Microsoft Learn on Copilot: “Without preparation, Copilot can misinterpret the data,” checked 17 Nov 2026.
The build
What we build. Six things you keep.
We start from Xero, your CRM, the till and your spreadsheets. No data warehouse needed.
What our numbers mean
- Sales = paid invoices
- Last month = 1 to 31 of the month
- Not counted: test, staff, refunded orders
- Outlets: Bugis, Orchard, Tampines
- The question list. What your team really asks about numbers, from chats and meetings.
- The meaning sheet. One page, signed by you, that every answer follows.
- Links to your apps. Read-only where your apps allow it, so the tool reads figures without changing them.
- The ask box. In WhatsApp, Slack or Teams. No new app to learn.
- Your test questions. Ones you already know the answers to, run before launch.
- Your own accounts. The tools run on your own accounts, so you can see what each one costs.
All your numbers in one app? Its own assistant may be enough. The questions below say when.
After launch
Unanswered questions get connected next.
Regular check-ins with your account manager decide what to connect next.
Question log, sample week
- 41Answered
- 6Asked back
- 3No data yet
Orchard outlet till
Next to connect
Time back for the person everyone asks. Fewer figures pulled by hand.
Decided in the meeting. The follow-up question can be answered there, instead of next week.
A new outlet shows up as a gap. We can log it under “No data yet” until we connect it.
Your team
The people who keep it right.
A figure looks off? Message one person, who brings in the right specialist.

AI specialist.
Puts AI where a step needs judgement, with a person checking what matters.

Integration engineer.
Keeps data matching and safe across systems.

Quality specialist.
Tests normal days and unusual cases, on real examples from your business.

A dedicated account manager.
One person who knows your business and owns the work.
- 1We review your details.
- 2We talk it through with you.
- 3We propose a solution.
- The systems, the code and the accounts are yours.
- Priority access to Luna. Partners go to the front of the queue with our AI assistant.
Before you ask
Can you trust a figure in chat?
BI means business intelligence: your figures. Start with the AI your apps already have.
What is conversational BI?
Conversational BI (business intelligence you can talk to) lets staff ask about sales, stock or spending in plain words, in WhatsApp, Slack or Teams, and get the figure back with how it was worked out. SuperToasty builds it on the business’s own accounts, and it answers from meanings the owner has signed.
- Starts from the apps you have
- Meanings signed by you
- On your own accounts
Isn’t Xero’s own AI enough?
Xero’s own AI may be enough if every number you ask about lives in Xero. Xero says its JAX chat has no additional charge for subscribers today, and its connectors let ChatGPT and Claude answer from live Xero data (Xero, checked 17 Nov 2026). A build is worth paying for when one question needs Xero plus your CRM, till or roster, such as sales per staff hour at each outlet.
Does our Power BI plan include Copilot?
Power BI Copilot is not covered by a Power BI Pro licence alone. Microsoft says it needs paid Fabric capacity (F2 or higher) or Power BI Premium (P1 or higher), and that trial and free versions don’t qualify (Microsoft Learn, checked 17 Nov 2026). Check what your Microsoft plan already includes before paying anyone, us included.
How do I know the number is right?
A conversational BI answer can be wrong in a way that looks right: test orders counted, refunds included, “last month” read as the last 30 days. Microsoft warns that without preparation, Copilot “can misinterpret the data” (Microsoft Learn, checked 17 Nov 2026). We guard against it with a meaning sheet you sign, a tool set to ask back when a question is unclear, and your own test questions with answers you already trust.
Will asking questions slow down our systems?
Asking questions does not need to touch your live systems. We connect read-only where your apps allow it, so the tool reads figures without changing them. We can run questions on a copy that refreshes on a schedule, not on the till or CRM your staff are using. Each answer can say how fresh that copy is, for example “data as of 6:00 today”.
Do we need a data warehouse first?
A data warehouse is not needed to start. A warehouse is one big store holding every figure in one place. Many small businesses have Xero, a CRM, a till and a few spreadsheets instead, and messy data is normal. The meaning sheet is where it gets sorted: one page saying what each word means as a number. Larger vendors call it a “semantic layer”.
Why not paste our numbers into ChatGPT?
Pasting an export into ChatGPT works once. The file is out of date the next day, nobody can see how the figure was worked out, and your numbers sit in one person’s chat history. Conversational BI reads your apps on a schedule, answers from meanings you agreed and shows its working. Developers call the core step “text-to-SQL”: turning a typed question into a request the database understands.
How is conversational BI different from an AI dashboard?
Conversational BI is something you ask; an AI business dashboard is something you look at. A dashboard shows the agreed numbers and says what changed without being asked. Conversational BI answers the question nobody built a chart for, such as last month’s sales per staff hour at one outlet, in chat, with its workings.
Who is conversational BI for?
Conversational BI suits a business of about 10 to 80 people where one bookkeeper or ops manager pulls every figure, and owners wait days for simple answers. It fits best when questions cross several apps. It is not worth building if your numbers are still on paper, or every answer you need already sits inside one app. Want the same figures sent to you every week instead? That’s Reporting Automation. Questions whose answer is written in your documents, not your numbers? That’s RAG Development.
Keep going
Not quite what you need?
See the rest of this group, or the service closest to this one.
