At a glance
- Generative AI is now being tailored for specific industries like finance and accounting.
- Early examples include Microsoft’s Copilot and Anthropic’s Claude for Financial Services.
- These platforms can slash hours of analysis and reporting work down to minutes.
- They could automate routine tax work by pulling data from various accounting systems.
One of the most striking aspects of ChatGPT is its versatility. It can compose a haiku, rustle up a recipe or whip up a marketing strategy in a matter of seconds.
But for accountants, the greatest impact of generative AI is likely to be an industry-optimised version of these AI platforms. If generative AI could be trusted to read financial data and write reports, it could shave hours off tax and compliance work.
The first of these sector-based versions have now emerged.
And the first industry on the block? Finance.
First steps
Early last year Microsoft launched Copilot for Finance. It is a specialised version of Microsoft’s consumer AI tool – one that accesses your finance data in Excel, Outlook and Teams. It then uses generative AI to reconcile records, flag anomalies, explain variances and turn findings into presentation-ready reports you can share straight from your inbox.
Eighteen months later, the product is still in preview. Microsoft has said little about the results of Copilot for Finance, suggesting that it is harder to replicate the runaway success of a ChatGPT in an industry-specific context.
But now Anthropic, the company behind AI tools like Claude Sonnet, has launched its own version – Claude for Financial Services. And while Microsoft’s attempt was announced with little more than a website, Claude for Financial Services’ July launch signalled greater focus. On day one, it boasted connections to enterprise data platforms and a string of high-profile financial institutions as early customers.
Claude for Financial Services is an enterprise-only edition. It can pull data directly from sources like S&P Global and combine it with a company’s own data in Databricks, Snowflake and other data warehouses. It is currently in testing with insurers like New York Life, Nippon Life and Verisk. It is also being tested by financial institutions such as Bridgewater, S&P Global and Norway’s trillion-dollar sovereign wealth fund, NBIM.
From hours to minutes
The concept behind Claude for Financial Services is simple. The AI chatbot will produce research by searching across public and private data and then generating models, memos and benchmarks, all with an audit trail. This type of work would typically take up to a full working day for a financial analyst. Claude can do it in minutes.
Anthropic claimed during the launch event that New York-based insurer AIG, which has begun using Claude for underwriting, was delivering claims five times faster. “What used to take weeks is now taking days”, it said. Anthropic also claimed accuracy in insurance claims at AIG had gone up from 75 percent to 90 percent.
Anthropic also quoted NBIM CEO Nicolai Tangen saying that Claude had fundamentally transformed how the fund worked, delivering the fund a 20 per cent productivity gain. That, he said, translated to 213,000 hours a year.
Of course, these are vendor-supplied metrics from a product launch, and warrant some scepticism. But at the very least, they show something of what a major AI vendor wants to do.
How generative AI speeds up finance
Anthropic walked through a typical example of how this specialist tool can be used. A fictional hedge fund analyst discovers that a company has posted bad results – revenue down 12% – yet its share price has jumped 17% to $71 before the market close.
“Claude has fundamentally transformed the way we work at NBIM.”
Nicolai Tangen, CEO, NBIM
With Claude for Financial Services, the analyst pulls data from various sources. These include S&P Global for transcripts, Morningstar for research, FactSet for estimates, and the company’s own research files in Box.
The analyst can then ask a single question and receive a clear summary synthesising insights from all the sources, with links back to originals. It flags key points from the earnings call, like a tariff-driven hit to profit margins, and shows an event-annotated price chart.
It also compares the company with peers to show the shares look expensive, then builds a plain-English valuation that estimates what the business is really worth today based on the cash it’s likely to make – landing closer to $54 than $71.
Finally, it drafts a ready-to-send investment memo using the firm’s template stored in Box, complete with sources, risks and next steps. It’s all done in under 30 minutes instead of 3–5 hours.
How could this affect accounting?
If Claude for Financial Services delivers on its promise for hedge funds and insurers, it’s worth speculating what it could likely do in an accounting context.
- For a small accounting firm, this technology could pull all of a client’s tax data into one workspace. This would include Xero, Sage and QuickBooks Online ledgers, bank feeds, payroll reports, receipt inboxes, HMRC portal downloads and prior-year workpapers.
- From there, it could turn that data into usable outputs, without manual rekeying. It could auto-classify source documents, extract key figures, and map them to income tax schedules, flagging missing invoices, VAT miscodings or payroll discrepancies for review.
- After that, it could generate reconciliations, lead schedules and an audit trail, draft a small business company return, and prepare client-ready summaries and cover emails with the right attachments.
The net effect could be to change a routine tax job from hours of clicking between systems to a focused review in minutes.
ChatGPT – with all its hallucinations, constant forgetting and lack of ability to learn – feels far from this potential reality. But the specialised versions have bigger context windows, so they can remember much more of every conversation. There are stricter controls over the data they can access. And they are trained to prioritise instructions from internal documents.
The AI in your software
While access to this special version of Claude is restricted to the big leagues, we will probably see it filter down to smaller organisations relatively quickly through software they already use.
Box is a document management platform popular in the insurance sector, and is functionally similar to Dropbox. Box has already integrated with Claude for Financial Services and its own AI engine to give wealth advisers a faster way to assess a client’s net position.
Advisers can point Claude at a curated “Wealth Hub” in Box – a hub that holds client financial statements, investment portfolios, estate planning documents, and goals – and then ask natural-language questions across that whole collection.
Claude can assemble an instant brief of a client’s position by reading every file in the hub. In practice, that brief looks like a clean narrative summary. It pulls in identity details, assets and liabilities, recurring transactions, and portfolio holdings pulled from PDFs, Word docs, and spreadsheets inside Box.
Autofill is where the time savings compound. Box AI extract agents and the Box AI metadata APIs can lift key fields from source documents – balances, account numbers, addresses, policy numbers – and write them into Box metadata templates. Those fields can then populate offer letters, application packs, or onboarding templates without retyping. In short, the documents you already store in Box become the data source for your forms.
While the promise is clear, these platforms are clearly in the early stages of development. There is not yet enough evidence that these tools can work with the accuracy and consistency required in accounting.
But the generative AI models are launching new and better models every eight months or so. With that speed of improvement, we may soon see a chatbot that can finally automate the grind.
Gain insights on implementing AI in your practice and the latest updates from industry specialists by revisiting the IFA AI and emerging technologies conference online available on demand here.
