AI Copilot: What an AI Copilot Actually Does Inside a Production Finance System

Most questions a producer asks the finance team are simple to say and slow to answer. "Can we afford another night shoot in block three?" sounds like one question. In practice it means checking actuals, open purchase orders, crew overtime that hasn't been posted yet, and whatever the location manager promised verbally last Tuesday.

That gap between question and answer is where an AI Copilot earns its place. This post covers what Copilot does inside a production finance system built on Microsoft Dynamics 365 Business Central, what it replaces, and where the line sits between AI assistance and a finance manager's sign-off.

What "Copilot" means in a production finance system

Copilot in Business Central is not a separate chatbot bolted onto your accounts. Microsoft describes it as a system feature that is part of Business Central itself. It runs on Azure OpenAI Service and works with the data already in your environment.

For production companies running Just-TV for production accounting, that means your budgets, cost codes, purchase orders and supplier invoices sit in the same dataset Copilot reads from. You are asking questions of your production's own numbers, not a generic AI tool working from a spreadsheet you pasted in.

What the copilot handles in practice

Surfacing the numbers behind a cost-to-complete answer

The most useful feature for production finance is analysis assist. It works with Business Central's data analysis mode, turning a plain-English request into a structured view of your data. Instead of building filters and pivots by hand, you type something like "show art department actuals and open purchase orders by episode against budget" and Copilot lays it out.

Chat with Copilot adds a conversational layer. You can ask questions about your data from anywhere in the system using natural language queries, and it works alongside analysis assist to produce analysis tabs directly from the chat pane.

This doesn't produce an estimate to complete on its own. It assembles the inputs, actuals and commitments against budget, in seconds rather than an afternoon. The forecast is still a finance decision, and we will come back to that.

Drafting reports and summaries

Once the figures are on screen, Copilot can help draft the explanation around them: a first pass at a budget variance note for the weekly cost report, or a plain-English summary for a producer who doesn't want to read a ledger. It is a draft, and treating it as one matters.

Flagging anomalies on request

Copilot doesn't patrol your ledgers looking for problems. What it does well is let you ask for exceptions in plain language instead of building the filter yourself. "Which cost codes are more than 10% over budget this week?" or "Show invoices from this supplier coded to more than one episode" are the kinds of checks that used to mean a manual export. Now they are a sentence.

What this replaces for the finance team

In most production offices, answering an ad-hoc question still looks like this: export the cost report, filter it in Excel, cross-reference the PO log, check for unposted invoices, then rebuild the view when the producer asks a follow-up. Each question creates a new spreadsheet, and each spreadsheet becomes a version-control risk.

Finance leaders expect this to change. In McKinsey's 2024 CFO pulse survey, 85% of respondents said they expect AI tools to generate insights that reduce the need for manual analysis. Among finance functions already using AI, 71% said it had boosted worker productivity.

For a production accountant, the gain is less about headline productivity and more about not rebuilding the same report three times in a week because the question changed slightly.

Before and after: an ad-hoc budget question mid-shoot

This is an illustrative scenario based on a typical scripted drama workflow.

The question. Week six of a ten-week shoot. At 4pm the producer messages the finance manager: the director wants a second unit day for a stunt sequence in episode four. Is there room in the budget, and where are stunts and locations sitting right now?

Before. The finance manager exports the latest cost report and filters it to the stunts and locations cost codes for episode four. She then opens the PO log to check what is committed but not yet invoiced, emails the location manager about a site fee that hasn't been raised, and builds a quick pivot to show the position. By the time the numbers are checked and a summary is written, it is past 6pm. The producer has already told the director "probably", which puts the finance manager on the back foot.

After. The finance manager opens Copilot and asks for stunts and locations actuals and open purchase orders for episode four against budget. Analysis assist builds the view in seconds. She follows up: "Which of these cost codes are over budget?" Locations is 8% over because of an earlier weather day. She asks Copilot to draft a short summary of the position.

Then she does the part Copilot can't. She adds the site fee she knows is coming but isn't in the system, factors in the stunt coordinator's quote, and judges whether the contingency can cover it. She sends the producer a checked answer before 5pm: yes, with a specific figure drawn from contingency, and a note on the locations overspend.

The difference isn't that AI made the decision. The finance manager spent her time on judgement rather than filtering.

Where AI assistance ends and human sign-off begins

This is the part finance teams care about most, and Microsoft is clear on it. Its analysis assist FAQ states that the feature organises how data is displayed without creating or changing business data, and that it doesn't provide conclusions about that data. It also notes that AI output is nondeterministic and may not be accurate, so users should review it before making decisions that affect others.

A few practical boundaries follow from that:

• Copilot reads and drafts. It doesn't post. Journals, payment runs and approvals still go through your normal workflows and approval limits.
• Access follows your existing permissions. Copilot features are governed by Business Central's permission sets, so a user can't see figures through Copilot that their role wouldn't otherwise show them.
• The estimate to complete is a judgement call. Copilot can show what the system knows. It can't know about the unpriced reshoot, the verbal deal with a supplier or the overtime that hasn't been submitted.
• Anything external needs a named sign-off. Cost reports going to a broadcaster, financier or completion guarantor should be checked and approved by a person, as they are now.

On data privacy, Microsoft states that it doesn't use your company data, prompts or responses to train AI models, and that its Copilot features are built under its Responsible AI Standard. For production companies holding confidential talent fees and commercially sensitive budgets, that is worth confirming before anyone asks.

A useful rule of thumb: Copilot answers "what does the system say?" Only your finance team can answer "what will this actually cost?"

Getting Copilot working on your setup

Copilot needs Business Central online, and an administrator activates each capability from the Copilot & agent capabilities page. Depending on where your environment is hosted, data movement settings may also need reviewing.

The bigger factor is data quality. Copilot can only give useful answers if cost codes, episodes and purchase orders are set up consistently. A production running clean coding through Just-TV gets far more from a plain-English question than one where half the commitments live in a side spreadsheet.

That is where Creative Total Media can help: enabling the right capabilities, setting permissions by role, and making sure your production data is structured so the questions your producers ask can be answered from the system.

Want to see what Copilot could do with your production data? Ask the CTM team about enabling AI Copilot on your Business Central setup.