
Silverfin boots up AI assistant
byPost-bookkeeping platform Silverfin has launched an AI assistant tool that can be embedded into its core system, running automated checks on accounting data and flagging outliers or opportunities to accountants.
The Silverfin Assistant tool has been operating in Silverfin’s home market of Belgium for the past two years, in part thanks to its acquisition of machine learning and artificial intelligence (AI) specialist Boltzmann.
Now the accounting platform has rolled the AI assistant into its UK offering. When purchased and activated, the tool sits in the background as an ever-present assistant, initiating more than 50 individual checks on accounting data and highlighting potential issues to users from within the software itself in context as a pop-up. These could include unusual balances, missing transactions or any outliers compared to previous periods.
“Accountants may not have time to manually trawl ledgers looking for things like this,” Silverfin director of product market Andy Perrin told AccountingWEB. “The AI assistant does a lot of the gruntwork so preparers and managers don’t spend hours searching for anomalies.”
Assist, don’t automate
The tool’s prompts offer the user the ability to flag any issue as an action, add it as a “to do” or create a task for a colleague to follow up.
“We’re aiming to put the human first,” said Nick Gianitsis, Silverfin’s UK head of product. “It’s up to the judgment of accountants whether something is relevant or not. The checks just become part of accountants’ to-dos and they can simply dismiss them if they’re not relevant.”
The assistant is designed to display its analysis to accountants in a clear way so any flag is easily understandable.

“Any AI tool needs to be transparent about how it comes to its results if users are to have confidence in it,” added Perrin. “This means how the tool presents the information and analysis is vital.”
Gianitsis also outlined how the new AI assistant could be used to proactively start conversations with clients across accounting firm portfolios.
“Take something like section 455 tax as an example,” he said. “If a director owes the company money and hasn’t paid the loan back within nine months they may be liable to pay more tax. The assistant can flag ‘this loan has been outstanding with the company for six months, the director might want to pay it back’.”
Accounting firms already using the Silverfin Assistant have also found the tool beneficial for training more junior staff. As per the example above, a first-year trainee might not have come across section 455 before, so this is a cue to start an in-context conversation.
Data quality an AI advantage
Launched in the UK four years ago, Silverfin targets the automation of core accounting work. It concentrates on a select number of key functions for the post-bookkeeping arena, including digital working papers, accounts production and management reporting.
Data is imported from the firm’s chosen bookkeeping tools or spreadsheets and mapped into Silverfin’s own standardised chart of accounts. When accountants prepare working papers, this data then flows into year-end accounts and tax computations and schedules. If the source bookkeeping or work paper data changes, this will be carried through automatically into Silverfin’s accounts production and corporation tax templates.
According to Perrin, boosting data quality in this way gives Silverfin a competitive advantage as more operators in the accounting tech industry move towards AI-driven tools, as ultimately any system is only as good as the data it’s connected to.
“We’re also not like a number of more general large language models on the market,” added Gianitsis. “Our assistant never takes data from one client’s database and compares that with another – we’re not using ChatGPT.”
Customers can submit feedback on the performance of Silverfin Assistant from within the tool. For a product tour and pricing details visit the Silverfin website.
You might also be interested in
Tom is AccountingWEB's technology editor, providing unbiased news and analysis from the accounting tech universe. Got a tech tale to tell? Message me on the site or email [email protected]
Replies (4)
Please login or register to join the discussion.
Where's the AI in all this?
The (useful) comparator/exception reporting routines were easy enough to set up in SQL 20 years ago, as I'm sure they can be in Excel; and bundling them together as a 'library' (which now gets called an app) is hardly tricky.
So the innovation ... importing data from your chosen bookkeeping tool and mapping it into Silverfin’s own 'standard' chart of accounts ... remains Silverfin's greatest strength & weakness.
Strength: because it enables their 'library' to be independent of any specific bookkeeping software;
Weakness: because it relies on the accuracy of the mapping routine for your bookkeeping software!
As I said, I'm probably missing something here, but where does AI enter the picture?
> Where's the AI in all this?
Hi Hugo! Ken from Silverfin here. Great question!
You already provided part of the answer yourself: indeed the CoA mapping is crucial, which is why we have an AI language model to assist our users with. I can assure you that it is a full-fledged NLP with a large embedding and multi-layer neural network looking at the account names and the underlying transactions. No if-thens here ;-)
For our AI Assistant, it's a broad set of machine learning techniques, but as well expert expert-driven checks as you point out. We've found that approach brings most value. AI turns out to be critical to ensure an out-of-the box functionality without any manual configuration.
Just a few AI examples: for each transaction amount, we model if it is not an outlier based on the type of account and other typical transaction values in that client file (actually many different statistical models combined in an ensemble). Also, we use bayesian modelling (similar to spam filters) to automatically detect unexpected number of transactions in a period. And for some accounts, we model if its balance is not deviating outside an estimated range based on other accounts using so-called quantile regression forests.
I hope this answers your question! I want to thank you again for asking it, because it made me think we could be more forthcoming about this, let me see if I can make a blog post out of it :-)
Note that we have an experienced AI team of more than 10 people working on this with top data scientists (several phd's in physics, nuclear engineers etc.). Feel free to reach out to me directly at [email protected] if you want to have some more insights or even if you have some more ideas, I'm always happy to talk!
As Hugo says, this doesn't sound like AI. It just sounds like an IF...THEN instruction, which has been a part of software for decades.
If director's loan overdrawn and and accounts more than 6 months ago then raise alert.
A potentially helpful tool but not a great advance on a simple checklist. Does slapping "AI" on it allow them to charge more?
"AI" is this years buzzword
Bit like "blockchain" about 2 years ago.
Be a new one in 6 months.