From Data Entry to Decision Support: How AI Is Transforming Financial Administration

AI can crunch the numbers in seconds, but can it tell you what they actually mean? Yvette Lamb explains why the Assistants who can interrogate the data will become increasingly valuable

There was a time when being “good with numbers” as an EA meant being able to reconcile wads of receipts from a multi-destination business trip. Or setting up an Excel spreadsheet to calculate the annual leave allocations of your entire team (which you likely filled in manually when your colleagues completed an annual leave request form – remember those!?). Now that most of these things are done for us, “good with numbers” no longer comes from being good at entering them and making sure they all add up. It comes from being able to interpret them once AI has input them for us.

I run an annual salary survey for EAs across four countries, and the discipline that the project demands is the same one this new part of our roles is asking for. We never fill a gap in the data with a guess. If a sample is too small to mean anything, we highlight it rather than let the number pass as fact (it can still be fact – it might just need a caveat).

A number can only be trusted once you know where it comes from and what it can’t tell you.

That’s what AI-generated financial data needs from us now. It’s fast and confident, so we need to know where we should be digging in. For example, when your expense tool alerts you to a potential anti-bribery and corruption instance, instead of handing it straight to Compliance, you look through to see what actually happened. Your salesperson in that area is leaving, and there has been a flurry of activity to hand their accounts over to other salespeople. This meant trips to see them, dinners, and introductory meetings. Perfectly explainable when you understand what that flag is actually for.

This is the same lesson that the survey data keeps teaching me. This year, we found that taking on more operational tasks barely moves an EA’s salary at all – with the correlation close to zero. However, actually holding an operations title does move the salary – by a real and consistent margin. The tasks and the title look like they should tell the same story, but they don’t. An AI flag works in the same way.

An AI flag shows you a pattern, but not the underlying reason behind it. Mistaking one for the other is how you escalate the wrong thing, or don’t escalate anything at all.

When such a large part of the job used to be data entry, the quicker you were at this, the better you were at your job. Nowadays, this is where AI and system integrations come into their own. Your CRM can link to your finance tool, and your HR system can link to your emails. There are apps to add receipts to expense systems, and AI can analyse your calendars. So, there’s a time saving there – and what we do with it is what sets us apart.

Monthly Budgets

We used to collate everything the monthly budgets needed ourselves. Pulling numbers from different sources, chasing Finance for their parts, and planning out what large expenses were coming up before being asked for them. Now that a lot of this gathering is done for us, our roles have moved to the first look to check that they make sense. Does this number look right? Does it make sense? Does it align with what the business is trying to achieve this quarter? What do I need to flag before this report goes anywhere else?

Expenses

Gone are the days of sticking receipts to pieces of A4 paper and scanning them before manually walking them to the accounts payable department (a normal procedure when I was in investment banking, and actually not the worst job to do on a Friday afternoon). Some of us have executives who actually scan the receipts into the app on their own. Yes, there’s still the “Oh, I lost my hotel receipt” response, and you have to chase it down, but I doubt that will ever go away. Now we are looking at those expense claims which are against policy, and we question the policy itself. Perhaps it’s out of date or no longer fit for purpose with how our business now operates. We don’t simply stop there, either. We rewrite the policy, we get it approved through the correct channels, and then we go and update the systems that use it.

“Learn the Tools” Is No Longer Good Advice

We don’t need to be fluent in these tools anymore. For starters, a lot of the new ones look the same, and for anything we don’t know, we can write in a prompt in our AI tool of choice to teach us.

You are a senior level Executive Assistant working in healthcare. The company you work for has just implemented a new invoice management system called New Invoice Corp. Please design me a user guide on how to use it, highlighting anything that might be similar to or very different from the last system we had called Old Invoice Corp. Design me some questions to act as a test for me to see how well I know it.

So, whilst a lot of job applications will still ask for years of experience on a specific system, that requirement is becoming decorative. What should get you hired now – and promoted – is evidence that you can pick up a new system fast and know what to do with what it tells you.

The Next Step

Once you’ve mastered the systems and how to interpret the data they produce, the next step is to build the process around which they work. This is where our roles step directly into strategic support.

Start noticing where numbers move manually from one system to another, or where reports take a manual effort to build each month.

You might not have the skillset or the permissions to build the solution yourself, but you do have something just as useful: the ability to run a full discovery of what’s actually needed and turn it into a complete brief that your technical team can build from.

I did exactly this with my company CRM. I automated the creation of a weekly report that used to eat up at least three hours of my week, and then I built an alert on top of it. If the value of an Opportunity changed by more than £100k, a message landed automatically in a specific Slack channel, tagging the Opportunity Owner. That meant their manager always knew where to go with a question, and they weren’t waiting a week for the new report to be generated before they found out about large deal movements. I wasn’t asked to build that; I noticed a gap and I closed it.

That’s the shift our roles are taking with the use of AI: we are designing the system that the data is in, rather than simply inputting the data and stopping there.

The Bottom Line

The job was never about data entry. Even at times when that was a large part of it, we were already questioning what we were inputting. What’s changed is that we’ve got the time back to do more of the interrogation, rather than it being a quick afterthought.

How you feel about this shift matters more than you might realise. In the salary survey, one of the strongest predictors of job satisfaction this year wasn’t pay, title, or hours. It was whether someone saw AI as a threat to the profession or a tool that supported them. The threatened group scored the lowest on job satisfaction than anyone else in the data. The supported group scored the highest.

We are being given an opportunity to actually think about data – financial or otherwise – and that makes us more strategic than ever.