Work in your own workspace

The Workspace is where your own finance or admin user can build and look after your company's reports, screens and dashboards, in conversation, without waiting for a partner. It works on your own company's data only.

Who sees it

AI Chat in the bottom of the main menu opens the Workspace; there is one place to chat, and it is this one. Users with the finance or admin role can build in it. Users with the operations or viewer role can open it too, to ask questions (see If you have the operations or viewer role).

How the page is laid out

The conversation is the page. A new chat starts in the middle of the screen, with a greeting by your first name and a raised box to type in. The box is lifted a little, with a firmer outline, and is ready for you to type, so you can start straight away. Under it are a few suggested questions, as small buttons, that you can click instead of typing. The firmer outline goes as soon as you start typing, and the box keeps a clear outline while you are in it. Once you send a message, the box moves to the bottom and your conversation fills the middle, in a column that is easy to read.

On the left, your chats. New chat starts afresh. Below it is the list of your own earlier chats in this company, most recent first, each with its title and when you last worked on it (Today, Yesterday, the day of the week, or the date). The chat you have open is highlighted, and clicking another opens it. Reports made here lists the reports this chat has made, and Older chats (before the workspace) opens the history from before the workspace. Once a chat is open the list folds away to a narrow strip, so the conversation and the report have the room; the strip keeps New chat, and its Show chats button brings the list back (Hide chats folds it again). A title too long for the list is cut short, and shows in full when you point at it. On a narrow screen the list is hidden.
On the right, a report opens beside the conversation when you ask for one and the assistant builds it. It stays closed until there is something to show. Close it with ×, and open it again from Reports made here or from the Open report button at the top of the conversation. On a narrow screen the report covers the conversation, and Close takes you back.
The divider between the conversation and the report can be dragged, moved with the left and right arrow keys (hold Shift for bigger steps), or double-clicked to put it back. Your width is remembered on this browser.

The report has three tabs: Report, Data, which lists the rows the report is made from as a table (figures on the right, totals in bold), and How it's built, which shows how the figures were worked out and anything the assistant found. The stages of the work and where it is saved are at the top of the report. For a screen or a dashboard, which read several sets of rows, Data says so and points you to How it's built.

A screen can carry controls above its figures, such as a month or a choice from a list. Ask for them in your own words, for example "let me pick the month from a drop-down". The preview opens with a starting value, and changing a control redraws the figures for the value you chose.

A screen can also open the detail behind a row. Ask for it in your own words, for example "when I click a category, show me its ranges". Clicking a row then shows the detail for that row alone, and you can ask for a further level beneath it in the same way.

Along the top of the report are three buttons:

Download gives you the report's rows as Excel (.xlsx) or CSV. The file holds every row the report returned, not only the rows you can see. Until the report is signed off, the file carries a line above the figures saying it is a draft, so it cannot be mistaken for a signed report.
Save keeps the report as a draft in My reports, and the button then reads Saved. If it cannot be saved, the reason shows under the buttons. Save is for a report; a screen or dashboard is already kept in the chat.
Publish sends the report for sign-off (see Sign-off below). It first asks you to confirm, in the page. If the report cannot be sent, the reasons are listed in that box, one to a line, so you can fix them and try again. Once it is sent, the button reads Sent for sign-off. Only finance and admin users see Publish. The assistant only says a figure was checked against your ledger when that check says so for that figure. For a count or a quantity, which is not money, it tells you no ledger check applies, and when a figure could not be checked it says what could not be checked and why.

Finance and admin users also have a More menu beside them, with Run review, Save as skill, Add block, Parameters, History and the other builder tools. A small number on More counts reports that were sent back, and History says the same, so a sent-back report is not hidden.

The page itself does not scroll: the list of chats, the conversation and the report each scroll inside their own box. Type your message in the Chat box, then press Enter or click the Send arrow at the right of the box (Shift+Enter starts a new line). The arrow is greyed out until there is something to send.

What you can do

Build reports, screens and dashboards by describing what you want.
Pull a table of your company's business data that you do not have yet.
Request a data feed to keep a table of your data up to date.
Upload files, such as Excel files with formulas, mapping tables and PDFs, and map your management accounts to the trial balance.
Find and fix slow reports.
Research and analyse your numbers. An admin, finance or operations user can start a background investigation for a bigger question and follow it to the answer.
Find out which tables and columns hold what, and whether a table is already in your data. To do the searching, the assistant may hand it to a quicker research step, which reports back what it found. The assistant still checks the figures itself before it gives them to you, and the steps line shows Researched your data when it did.
Schedule a published report to run daily, weekly, monthly or at month-end, if you are an administrator. Results appear in the Scheduled bundle. A draft cannot be scheduled; publish it first.

Your plan sets how much you can do each month and the quality of the answers. It does not change what the Workspace can do.

If you have the operations or viewer role

You can open the Workspace and ask questions about your company's data. You can:

Ask questions and get answers that show where the figures came from.
Run the reports that already exist and ask about them.
Start a background investigation for a bigger question, and follow it to the answer (operations users; a viewer can read answers but not start one).
Get help with using the product, and check whether your data is up to date.
Find out which tables and columns hold what, and whether a table is already in your data.

You cannot build, upload files, pull data, or send anything for sign-off, so those controls are not shown to you. Building is for your finance or admin users, and you can ask one of them to build what you need. Your questions use the same company allowance and counter as everyone else's.

Your questions in the Workspace are not kept as a piece of work, so leaving or reloading the page starts a fresh conversation.

The side panel on each screen

The assistant panel that opens beside any screen is a narrow view of your Workspace. It reads the screen you are on, so you can ask "explain this figure" where the figure is. Personal details on the screen stay masked.

If you build (finance or admin), the panel carries on with the piece of work you last had open, or your latest open one. If you have none, your first question starts one, without leaving the screen.
If you ask (operations or viewer), the panel answers in a conversation that is not kept.
Widen opens the full Workspace on the same piece of work.
New conversation starts afresh.
Earlier chats opens the list of your earlier conversations, which you can read but not continue.

The panel uses the same company allowance and counter as the Workspace.

The usage chip

Usage shows in one line of small grey text in the top bar of the page, beside the page's name, AI Chat. It appears only when there is something to show, and it is hidden on a narrow screen.

Your plan has a monthly allowance. The line says how much of the company's month has been used, for example "34% of the month used".
Your plan has no limit. The line says nothing about the allowance.
You are the billing owner. The line also shows what the month's AI turns have cost, and what the open chat has cost. Nobody else sees the cost.
Neither applies to you. Nothing is shown in the top bar.

Click the line to see the detail: how much of the month has been used and how much is left, and, when you have a piece of work open, how much of the allowance that piece of work has used. The billing owner's detail also has a table of the latest turns.

Close to the allowance. You can carry on, and the page tells you the company is close.
Allowance used, or the work's share used. That work stops until the next month, or until your administrator tops up the allowance. Reports and screens you have already saved stay open to read.

Everything you do in the Workspace uses the same company allowance as AI Chat. See Your company's monthly AI allowance.

Reading an answer

While a question is running, the progress line describes the current activity, such as looking for data, checking a result or writing the answer. It updates during research as well as during the final reply. These are activity updates, not a percentage complete or a promise about how long the work will take.

For a saved piece of work, reopening the chat restores its latest saved progress. If the connection drops, the chat checks for the saved result and tells you it is reconnecting. It does not send your question again. Wait for that check before repeating the question. A confirmed failure remains visible with its reason. Stop ends following the current response. For an operations or viewer conversation that is not saved, the chat cannot recover a response after a connection loss; it tells you when you need to try again.

Once you ask for an outcome, the assistant continues with the research and routine steps needed to finish it. It works through ordinary technical problems without asking you to approve each retry. Existing sign-off and confirmation steps still apply, including publishing a report and confirming an investigation method.

When a business choice needs your answer, it appears in a separate Decision needed card. The question is in bold, followed by why it matters and buttons for the available answers. A recommended option is marked when there is one. Choose a button, type its number in Chat, or type your own answer. The assistant then continues. The card and your answer remain in the conversation. General choices are kept in the chat; report judgement choices keep their existing sign-off record.

Each answer reads as part of the conversation, the way it does in the common chat apps: your messages sit on the right in a soft bubble, and the assistant's replies run down the left with the product's mark beside them. A message written by a colleague shows their initials and name above it; your own carry no name. The conversation and the Chat box share one centred column, and a refusal or failure appears as a small bordered box inside the reply.

The steps line. Above an answer, one line says what the assistant did to get there, such as Looked at 3 tables and ran 2 queries · 14 s. Click it to see each step in words; click it again to hide them. The counts come from the work itself.
While the assistant works. You see each step as it happens. The step being done now sits in the open under the steps line, with a ringed mark that turns, what it is doing (for example Reading the data…) and a clock that counts up in seconds. Inside a long step it also says which part it is on, such as 2 of 4. The steps line counts what has been done so far (4 steps so far) and, when the answer arrives, folds back to its usual summary. Below it, Working · 2 min 10 s shows how long the whole request has been going. Between steps it says what was just done and that the next step is being chosen, or Thinking about the next step… if nothing has been said yet. The clock stops when the answer arrives or the work cannot finish. If you reload the page mid-request, the step in progress and the clock come back where they were.
Notes awaiting review. When a step uses a business note that nobody has checked yet, the step says how many are still awaiting review, and the steps line says the total beside its count (for example 3 steps so far · 3 facts still awaiting review), so you can see it without opening the steps. The note is used, and flagged so you know where an assumption came from. It is not a fault with your answer.
Points noted for review. When the assistant sets something aside for someone to check, such as a part of your data it could not match fully, it does not interrupt the answer. It sits quietly with its step, in plain grey and without a box, as Noted for review followed by a short title; the answer itself explains the point in everyday words. The steps line counts them (for example Took 4 steps · 1 point noted for review), and puts them beside any facts still awaiting review. A caution from a step, such as only the first 500 rows are shown, is shown the same quiet way. A note is not a fault. Only something that has gone wrong, such as a request that was refused, is shown in a bordered box.
A limit met. When a step meets one of the workspace's limits, such as the 500 rows a draft's preview shows, a card with a warning mark sits in the answer. It names the limit, gives the rule with its number and lists the ways round it. The assistant never cuts the work short because of a limit without asking you. A request to our product team is only proposed until you choose to file it. See Limits in your workspace, and asking our team for something new.
Tables. A table in an answer is laid out like a statement: figures on the right, negative amounts in brackets and in red, and a bold total row. A wide table scrolls sideways inside its own box. When there are more than 15 rows, the answer shows the first 15 and Download table has them all.
Under a finished answer, Copy appears, and when the answer shows a table, Make a report from this and Download table appear beside it. They are not shown while the assistant is still working or waiting for your answer to a question. They stay in full view under the latest answer; under older answers they appear when you point at that answer or tab into it.
Make a report from this asks the assistant to turn its last answer into a one-page draft report: the table, a trend chart and a short commentary. It is the same as typing that request yourself, and it uses the same allowance. It appears only under an answer with a table of figures, and not under the answer that has already built the report. It is shut, with the reason shown, when the work is completed and read-only.
Copy copies the answer as plain text. A table pastes into a spreadsheet as columns.
Download table saves the answer's table as an Excel file. It appears only when the answer has a table.
The report card. When the assistant has built a report, a card for it appears in the conversation, after the answer that built it. It shows the report's title, Report · draft, and a badge for where it stands (Draft, Sent for sign-off or Completed). Click the card to open the report beside the conversation.

Sign-off

A report or screen built in the Workspace is not live for your company straight away. It is sent for sign-off to Report Approvals. An administrator sets the named approvers and the number of approvals under Admin → Report Approvals. Only a named administrator can sign it off. Until it is signed it is provisional.

Open the report's inbox notification to review it, or choose Review and sign off from the notification detail. This opens the submitted report and its review controls, beside the original request in the words it was asked in, what was asked along the way, the changes in this version, and anything still outstanding. When you send a report for sign-off you get a notification too, and it opens the same page. Any notification about a piece of work opens the report itself once it is waiting for sign-off, not the conversation. Review any missing-data warnings before signing off. A report approval does not connect a missing data feed; feed requests have a separate status under Admin → Data feeds.

If you are the administrator, you may sign off your own work. The record shows that you were also the author, and Report Approvals and the sign-off page say so.

Once it is live, a report built this way shows Built in your workspace in the Reports catalogue. Only an administrator can archive it.

Slow reports

A speed fix you propose does not go live by itself. It waits for your administrator on the Partners page, under Speed fixes, where it reads "A colleague proposed…" and names you. The administrator can also undo it later. A finance user cannot make a fix live or undo it. You can only look at and fix reports your own role is allowed to open.

Handing work to a partner

Sometimes the assistant cannot finish a piece of work for you. A line under the top of the page tells you why, and offers to hand the work to a partner. It appears when:

none of the layouts we have fits what the work needs;
data the work needs has not arrived yet;
a report is too slow to open;
the work has used its share of the allowance;
the check on the preview has failed twice.

Who can hand over. Your administrator sees the choices and can hand the work over. A finance user sees the same choices as Ask to hand over to… buttons. Asking does not move anything: the request waits for your administrator, who confirms it under Report Approvals. If the partner does not yet have your company's consent, the administrator gives it first on the Partners page, and a finance user is told so. Only the administrator can use Find a partner. Operations and viewer users do not see the offer.

Where it can go. The choices are shown in this order:

Your partner, if your company has one.
Our team.
Find a partner, which opens the Partners page to find a certified partner close to your company.

A partner that does not yet have your company's consent is sent to the Partners page first, where you give it (only your administrator can). The administrator then confirms the hand-over under Report Approvals.

What moves. The work moves to the partner with its conversation and its draft. You can still see it, read-only, with the partner's name and where the work has got to. From then on the partner's work does not use your company's allowance, and the partner sets a quote before the work goes for sign-off.

What is out of reach

Payroll is never reachable in the Workspace.
A report your role cannot open cannot be looked at or sped up from here.

How it relates to AI Chat

AI Chat in the menu opens the Workspace. You ask questions and read answers there, and finance and admin users can also make and change things. Operations and viewer users use it to ask. It reads your data under the same rules the old chat page did, so a number you see in one is the number you saw in the other.

Earlier chats

The conversations you had before AI Chat moved to the Workspace are kept. Open Earlier chats, from the side panel or from the link at the top of the Workspace, to read them. They are read only: you can read each conversation but you cannot add to it. Ask your next question in AI Chat.