The reason most monthly management packs take six to eight hours to produce isn’t the numbers. The numbers come out of Xero, QuickBooks, or Sage Intacct in seconds. The bulk of the time is the narrative: writing the commentary that explains what changed, why, and what the client should care about. That’s the part AI can genuinely halve, when you split the job correctly between specialised reporting tools and a general-purpose model.
The split that works is the one most firms haven’t operationalised yet. The reporting platform produces the figures, the visualisations, and the variances. The AI produces the commentary, the executive summary, and the client-friendly explanation. Put them together and a senior accountant can finalise a board pack in roughly the time it used to take to write the cover memo alone.
Stage one: let the dedicated platform do what it’s good at
Before AI gets involved, your figures should already be calculated, formatted, and visualised by software designed for it. Fathom, Spotlight Reporting, and Syft Analytics all connect directly to the GL and produce variance tables, KPI dashboards, and consolidated reports without you doing the maths.
This matters because LLMs are unreliable arithmeticians on raw text. Letting Fathom calculate “gross margin moved from 38.4% to 35.1%” is far safer than asking ChatGPT to derive it from a CSV. The reporting tool is your source of truth for figures. The model is your source of language.
The output you’ll feed to the AI is the variance table and KPI summary the reporting tool already produced, not the raw GL. Export it as a PDF or paste it directly into the prompt. You’re handing the model facts, not asking it to compute them.
Stage two: the narrative prompt that produces a memo a CFO would sign
Here’s the prompt that does most of the heavy lifting. Adapt the bracketed sections to your client and reporting period.
“You are a senior management accountant writing the commentary for the [PERIOD] management report for [CLIENT NAME], a [INDUSTRY] business with annual revenue around [REVENUE]. Below is the variance table and KPI summary from our reporting platform. The figures are correct, do not recompute. Produce: (1) a 60-word executive summary stating the period’s headline result and the single most important driver, (2) a ‘what changed and why’ section of three to five short paragraphs covering the most material variances (do not list every line, only the ones that move the result by more than [MATERIALITY THRESHOLD]), (3) a ‘what to watch’ section flagging two forward-looking risks or opportunities visible in the data, (4) a one-line recommendation for the client on the most useful action they could take this month. Tone: confident, advisory, written for a business owner with operational understanding but not financial training. No jargon, no hedging language, no marketing buzzwords. Maximum 600 words total.”
What lands in your inbox is a draft that’s roughly 80% of what you’d send. The remaining 20% is your sector judgment, the things you know about the client that aren’t in the table (the lost contract, the new hire who started Tuesday, the supplier renegotiation in progress). Adding those takes about 15 minutes. The total elapsed time from “the data is in” to “the memo is sent” is about 25 minutes for a small client, an hour for a complex one.
The failure mode: letting AI invent a reason
The single most common mistake on these prompts is letting the model speculate about causes the data doesn’t actually show. If gross margin dropped 3 points, the variance table will show whether it’s a price issue, a cost issue, or a mix issue. The model will, if asked, confidently write “this was driven by rising input costs” even when the data shows the input cost line is flat and the issue was actually a sales-mix shift.
The fix is in the prompt. Tell the model explicitly: “If the data does not show the cause of a variance, say ‘cause not visible in the data, requires client conversation.’ Do not speculate.” That single constraint is what separates a memo a CFO will sign from one they’ll quietly rewrite. Add it to your prompt template and never remove it.
A worked example: the gross margin sentence
Suppose the variance table shows gross margin fell from 38.4% to 35.1% in the month. Without the constraint, the model produces something like: “Gross margin compressed during the period, likely due to inflationary pressures on raw materials and supplier pricing renegotiations that are common in the current macro environment.” Plausible. Wrong. Generic. And impossible to defend in a client meeting if the actual cause was that you mis-coded the freight account.
With the constraint, the model produces: “Gross margin fell 3.3 points to 35.1%. The variance is concentrated in cost of goods sold (up 8.2% on revenue growth of 3.6%). Cause not visible in the data: recommend confirming with the client whether there were one-off freight, supplier, or stock write-down items in the period.” That’s defensible, it points to the next conversation, and it doesn’t put words in the client’s mouth.
What this changes for the firm
Wolters Kluwer’s Future Ready Accountant 2025 reported AI adoption inside firms jumped from 9% to 41% year over year. The firms in that 41% figure aren’t all using AI for the same thing, but the cluster reporting the biggest practical wins are using it exactly here, on the narrative layer of management reporting and the explanatory layer of client communication.
The counterintuitive part is that the time saved doesn’t end up as fewer hours billed. It ends up as more reports per partner, more frequent reporting cycles offered to clients (monthly instead of quarterly), and a better mix of work for the team because the writing gets done in 25 minutes rather than two hours. The senior accountant’s afternoon goes from grinding through variance commentary to actually talking to clients about what the variance commentary means.
If you’re a small or mid-size firm that has already moved your reporting onto Fathom or Spotlight, the missing piece is the narrative prompt above and a 30 minute team session on how to use it consistently. That’s the upgrade that quietly halves the production time for monthly packs without changing your tooling stack at all.