Save content
Have you found this content useful? Use the button above to save it to your profile.
 a variety of electronic components on a dark background
istock_blacklight_trace_AW_computing components

Fix data and connectivity problems before AI accounting tools can deliver


While new generative artificial intelligence tools have the potential to transform accounting, until the age-old issues of messy data and vendor cooperation are fixed we may find that the promised revolution doesn’t arrive.

25th May 2023
Save content
Have you found this content useful? Use the button above to save it to your profile.

We’ve recently been blitzed with new releases trumpeting what the latest generation of artificial intelligence (AI) software will do for accountants, and each new day your fearless technology editor is greeted with news of a new development that will revolutionise the accounting industry.

The likes of Pixie, Karbon, Client Hub and Canopy all get first-mover bonus points for finding accounting firm efficiencies, but is email sentiment analysis the revolution we’ve been promised?

Having lived through innumerable hype cycles, it’s tempting to dismiss any new boom as marketing puff. As our editor at large John Stokdyk recently pointed out, we’ve had at least one version of artificial intelligence play out before us recently, with Pegg, Wanda and co all now consigned to the recycling bin of digital history.

Connecting and embedding

However, what sets this breed of tools apart from previous attempts is their inter-system connectivity – the fact they don’t exist in a vacuum. In the past, what was billed as AI turned out to be an isolated spreadsheet packed full of IF-THEN formulas, or even a less-than-artificial team of outsourced humans. 

This time around, AI merchants are using not only their models but also connecting with and embedding into existing technology, such as cloud accounting systems, document management and practice management software, and putting them to practical use. Who wouldn’t want to log on in the morning and find their correspondence neatly dealt with and ready for review in the drafts folder?

But eventually, the technology hype express must meet the cold, hard buffers of accounting reality. In this case, it’s the realisation that the system is only as good as the data it’s connected to. 

Sure, you can now draft correspondence to clients about a networking event in the style of a Gilbert and Sullivan light opera, but if your records aren’t up to date how will you know if your libretto has arrived in their inboxes?

Will tech firms risk their defensible moats? 

Then there’s the red signal of connectivity. To fully realise the potential of this new tech, you need to find ways to make the systems talk to each other and pass over data.

As anyone who’s had dealings with technology companies will know already, getting them to talk to each other is hard enough, let alone agree on connectivity that could risk their defensible moat for the greater good. Vendors talk a good ecosystem game, until one of the players encroaches too far onto their turf, or they spot a way to move in on theirs.

For a recent parallel, there’s Open Banking, which promised faster and safer access to a rich set of financial data. In some ways, it was a miracle that standards were agreed in the first place, and there have been plenty of efficiencies gained, despite the high-street bank incumbents dragging their feet every step of the way. However, these gains have been hard-won, and herald more of a quiet revolution than a big bang.

While there is huge potential in the new tools coming to market, until these age-old accounting issues of messy firm and client data are fixed and the majority of vendors agree to cooperate, we may well find ourselves in a similar situation to the Open Banking era, with a few new flashy toys that save time rather than the promised whole-scale transformation.

Replies (6)

Please login or register to join the discussion.

By Tornado
26th May 2023 00:25

In this case, it’s the realisation that the system is only as good as the data it’s connected to

Another great reality than describes the whole MTD project perfectly.

Thanks (0)
John Toon
By John Toon
26th May 2023 09:57

One advantage of LLMs is of course that as a lay person, with no coding experience, you can now find the appropriate code to run SQL queries, connect APIs or analyse data in a spreadsheet. So whilst accounting tools may continue to lack interoperability out of the box maybe now is the time for apps to start shouting about their open APIs again, which were of little use to those unable to make use of them.

As for email sentiment analysis - you've been able to do this with Power Automate/Outlook for a good few years now...

Thanks (1)
Replying to johnt27:
By Hugo Fair
26th May 2023 12:04

OK I'm intrigued ... and out of my depth (in terms of practical experience I've never played with LLMs or any aspect of AI).

But are you saying there's finally a solution to the problem with which I grappled for over 30 years (in different companies and for different applications) - namely that every need for what used to be known as interoperability is broadly solvable in one of three ways:
1. Fixed data dictionaries at both ends (basically hard-coded);
2. Same technology but with new 'hub app' in between (better protection);
3. APIs (whether push or pull variety) which you might term soft coding.

However the problem is that each 'interface' (including the APIs themselves) is still 'owned' by someone, who is (or should be) responsible for maintaining it.

So when one or other of the database schemas is changed, anyone else in the chain needs to be informed and to take appropriate action (including testing) ... which might sound obvious but was either unknown or ignored by HMRC on many occasions.

I'll give a really simple example (which paints me as the incompetent albeit many moons ago):
* I 'owned' a large HR system used by major organisations and it included a data item for NINO (as part of the recruitment/onboarding process and to make feeds to any Payroll simpler);
* When we built our own integrated Payroll (single shared database), I split NINO into 3 data items in order to offer full data-entry validation (e.g. validate prefix);
* No problem for our clients, but helpdesk went into brief meltdown with calls from angry 3rd-parties (some unknown to me) whose data uploads didn't work!

I guess you're going to point out that the better APIs are now 'open' (which is an improvement but brings its own problems) ... I'm just curious as to how LLMs can solve this core conundrum?

Thanks (1)
Replying to Hugo Fair:
John Toon
By John Toon
26th May 2023 14:12

I'm not saying LLMs fix the problems with APIs, of which there are many or wider access to SQL databases (which some vendors keep under tight control). But, over the years the various cloud vendors have paraded around their open APIs as a useful feature, whilst the reality was most accountants have no coding experience themselves, or within their staff/consultant base, to make real use of them.

There were ways around this using tools such as Zapier, but that required both Zapier and the cloud app to have built a connector which are often developed based on demand, perceived or otherwise.

Now though you could go to Chat GPT, for example, and ask it to provide the necessary code to pull data from, or push data to, an API. It would also provide all the necessary info to make this useful. Products like Microsoft Dataverse also make it incredibly easy to create new data lakes/warehouses (with all the trials and tribulations this could bring). Of course, if said API is amended, the code may no longer work so this doesn't replace the need to keep on top of API documentation/changes (assuming this is published)

Thanks (1)
Replying to johnt27:
By Hugo Fair
26th May 2023 14:28

It's good to see that my sardonic perspective and your positive approach are still rooted in the same technology world (which I had assumed was the case - but do sometimes doubt now that I no longer inhabit that world on a daily basis)!

Thanks (2)
By MarkRyan
26th May 2023 10:34

I agree with the data-issues sentiment

CynareLink has been delivering this integration for over 15 years

Firms can't rely upon any traditional vendor (particularly the moribund ones) to help them access
their own data, as this is perceived as a risk to their client-base


Thanks (0)