Bookkeeping agencies in 2030

Imagine a bookkeeping agency with five employees. Three of them spend much of their day entering supplier invoices, posting bank statements and organising documents. There is plenty of work, clients keep bringing in invoices, deadlines are approaching, and there is always something left for tomorrow.

Now imagine software taking over much of that data entry. Invoices arrive automatically, their details have been read, routine items have been processed, and anything unclear is waiting for review. Work that used to keep people late suddenly takes up a much smaller part of the working day.

Great. What does the rest of their day look like now?

That is where the story of AI and bookkeeping gets especially interesting to me. When you build software, it is easy to focus on everything the program can do. The agency owner also has to think about the people who used to do that work, the clients who pay for the service, and the competitor who will use the same technology in their own way.

I think bookkeeping agencies will still be around in 2030, but I expect substantial changes in how they are organised, what their employees do, and why clients choose them.

Blockbuster and Netflix

TLDR: Agencies that do not embrace AI trends quickly risk their survival

The story of Blockbuster and Netflix is useful here, provided we resist the temptation to act wise after the event.

In 2004, Blockbuster was a huge American chain with more than 9,000 video rental stores, 80,000 employees and annual revenue of $6 billion.

Netflix started mailing DVDs in 1998. It introduced online streaming in 2007.

According to Netflix co-founder Marc Randolph’s recollection, in 2000 they offered to sell Netflix to the then-dominant video rental chain Blockbuster for $50 million. No deal was reached because, at the height of its power, Blockbuster saw Netflix as an insignificant technological novelty. They literally laughed in their faces as they turned down the offer. Something similar happened in 2000 when Yahoo declined to buy Google for $1 million — almost the same story.

The point is that Blockbuster filed for bankruptcy protection to reorganise in 2010, while Netflix had revenue of $45 billion in 2025.

What interests me most about this story is the change in what a customer has to do to watch a film. Going to the video store, choosing something from a shelf and returning the disc were all perfectly normal parts of the process. When the film can reach you another way, a business model built around those steps needs to be reconsidered.

It is worth imagining something similar in bookkeeping. Over the past 18 months, 99% of the bookkeepers I have spoken to have reacted in a similar way when I say that “bookkeeping AI” is inevitable in the near future. That includes laughing in my face and the familiar “No chance, this is Serbia!” :)

Maybe they are right. Maybe this is all the fantasy of a programmer who does bookkeeping as a hobby and probably does not really understand it.

But what if it is true? What if tomorrow their competitors have AI software that lets them do twice as much work in 50% less time or even less? What if they are Blockbuster, laughing at Netflix?

The end of bookkeeping feudalism

TLDR: Competition becomes nationwide

Let us step outside the office and look at where the clients of our story’s heroes — a typical five-person bookkeeping agency — come from.

In my personal experience, the vast majority are within a radius of about ten kilometres. A local business, a local bookkeeper, and documents dropped off along the way every few days. There may be a client from another town, but they are always the exception: “that unusual client”.

With a little exaggeration, I would call this a bookkeeping fiefdom. Its boundary is set by how far a client can take a folder without too much inconvenience. That is a perfectly practical reason, especially when documents have to travel frequently between the business and the agency.

Except that an electronic invoice can already be downloaded from SEF, Serbia’s electronic invoicing system, directly into connected software. A bank statement can be delivered electronically, an invoice received by email can be forwarded, and a question about a document can be resolved without visiting the office. Every step completed that way makes the folder, and the distance it travels, less important.

Once paperwork no longer requires regular trips to the bookkeeper, it becomes perfectly reasonable for a client in Ćuprija to consider an agency in Subotica or Niš that specialises in exactly what they need.

That means more competition for the local agency. But it also means access to clients it could hardly have reached before. If it knows a particular type of business especially well, it no longer has to hope that ten more such businesses will open in its neighbourhood.

But how do you add ten more clients when the agency can barely keep up with the ones it already has? Of course, AI increases productivity and efficiency, allowing it to handle those “ten more clients” without difficulty.

A personal relationship can still be a good reason for a client to stay. Some people like dropping by, have known their bookkeeper for years, and place great value on that trust. Such relationships are worth nurturing. I would simply be cautious about assuming that a nearby office will, by itself, remain enough of an advantage.

What will people do if AI handles data entry?

TLDR: Reviewing AI’s work and developing domain expertise

Let us return to our agency. People who spend years entering invoices learn a great deal along the way that does not fit into a job description labelled “data entry”. They know what a particular client’s documents look like, what is usually missing, when a payment seems suspiciously small, and why a certain invoice calls for a conversation with the business owner first.

That experience can also be put to use in a differently organised workplace.

Suppose an employee opens a screen in the morning showing documents that AI has flagged for review. The original invoice is on one side, the proposed accounting entry on the other, and below them an explanation of why the item was held back. One purchase lacks information about its purpose. Another differs from the client’s usual expenses. A third needs a more experienced colleague’s attention.

The employee reviews the document, confirms the proposal, corrects it, or sends a question to the client. If the case is beyond their expertise, they pass it to someone who can decide. The invoice number and amount are already there; their time goes into the part that requires checking and understanding.

Practical experience with documents can be a perfectly good foundation for this kind of work. Training could start with familiar examples: here is what the program did, here is how you check it, this is where you ask for more information, and this is where you call a colleague. Telling someone to “learn AI” means very little when they have to sit down at a computer and get to work on Monday.

The software, of course, needs to support productive work without requiring a doctorate in computer science. If reviewing one invoice means showing an employee seven technical messages and asking them to compose the perfect prompt for an AI, both the software maker and its users are in trouble. The screen should clearly explain the issue in ordinary language and show what the person can do about it.

Review also needs a purpose. There is little value in having someone spend eight hours clicking “approve” on everything the program suggests. A broader set of documents might be reviewed at first, with more attention later devoted to exceptions, sensitive cases and a sample of routine processing. The extent of review would depend on results and risk, with a clear record of who did what.

As the system improves, even those checks may become less frequent. So I would not count on this screen providing enough work forever for all three people in our example. It is worth looking at what gets neglected today because nobody has time for it.

Another possibility is for an employee to become the person in the agency who knows a particular group of clients especially well. Take small businesses that sell services online and receive payments from abroad. They deal with different platforms, fees, reports and recurring questions.

For such a client, it matters a great deal that someone in the agency is familiar with PayPal, Stripe and similar systems. Someone who can read a platform’s report, knows which documents to request, and understands why the amount arriving in the bank account is not necessarily what the customer paid. They will involve the appropriate specialist for the specific tax treatment, but the client will not have to begin every conversation by explaining how they make money in the first place.

That person can onboard new clients, help them set up document delivery, follow up on open questions and notice when something new appears in their business. Another employee may know retail better, and a third construction. Over time, expertise deepens in the areas where the agency already has clients and experience.

This takes training, time and support from more experienced colleagues. It also requires the owner to talk to their people: some will prefer reviewing documents, some working with clients, and others developing the skills to handle more complex cases. Before looking for entirely new employees, it makes sense to explore what can be developed within the existing team.

How will new bookkeepers start their careers?

TLDR: The start of a bookkeeper’s career will change

What we look for when hiring will change.

Knowledge of regulations will still be necessary to judge whether what is on the screen makes sense at all. But computer skills, an understanding of how a document moves through several systems, and the ability to check a software proposal will become increasingly useful. By 2030, familiarity with AI tools will be as essential a hiring requirement as “Excel skills” are today.

Experienced employees can spot when something is wrong because they have seen so many invoices, statements, corrections and unusual clients. They had to acquire that experience somewhere, often through the very tasks we are now trying to automate.

Imagine a beginner who, in today’s agency, would have spent their first year or two on simpler tasks, with a colleague occasionally correcting them and explaining why things are done a certain way. Now they arrive at an agency where software handles most routine data entry by itself and leaves the harder cases to people. On their first day, they are given the things even the program was unsure about. We have hardly made their start easier.

If beginners no longer encounter those apprenticeship tasks in their everyday work, someone has to make time to go through them together. What used to be learned along the way will need to be planned.

I predict that there will also be AI that gives a beginner a set of fictional documents and asks them to propose a course of action, then compares it with the approach the AI identifies as correct. Next comes a case with missing information, where the beginner has to decide what to ask the client. In a third case, the program offers a convincing but incorrect suggestion.

Ideally, this should be part of education in business schools and universities.

Those exercises can be combined with supervised work on simpler real cases, exposure to different types of clients and a gradual increase in responsibility. An experienced colleague may find that the best use of some of the time freed up by automation is training the next colleague.

If this sounds like “Nikola is fantasising”, this sort of thing is already widely used in the United States for CPA training. Automated training has existed there for some time and is now being reworked into something roughly like what I have described.

A bookkeeper who knows my business

TLDR: Specialisation and personalisation

This brings us back to the agency employee who specialises in a particular group of clients.

Imagine the owner of a small business selling services to foreign customers. When looking for an agency, they gain little from hearing that it can handle another twenty different industries. What matters is that their way of doing business is not treated as an exotic case everyone has to ask around about first.

If they can choose that agency from anywhere in Serbia, there is more room for specialisation. An agency that already understands online sales could develop clear guidance for those clients, train its people to handle their most common problems, and organise document intake so the same uncertainties are not resolved from scratch every time. Another might do the same for construction companies or hospitality businesses.

Choosing one or two groups you already have experience with is enough to begin. Some of the more complex questions can be handled with an outside specialist or a partner agency. There is no need for every employee to become an encyclopaedia of every kind of business in Serbia.

In a market like that, it is not hard to imagine a kind of “Uber for bookkeepers”. A client describes what they do and what they need, and receives a shortlist of agencies that work with similar businesses, an explanation of their services, and recommendations and ratings from other clients.

Five stars alone do not tell you that the tax treatment is correct. But a platform that could verify providers’ experience and clearly show what the service includes could make choosing easier. Whether such a model will take hold here remains to be seen.

Even without waiting for that platform, an agency can describe its work more concretely: which clients it has experience with, what problems it solves for them, how the relationship works and what is included in the price. That gives a new client far more to base a decision on than yet another promise of “quality bookkeeping services”.

Bookkeeping in real time

TLDR: Clients will expect instant answers

Bookkeeping is an asynchronous activity today. Clients deliver paperwork during the month, the bookkeeper calculates VAT once a month for VAT-registered clients, and only in February do they draw a line under the previous year, prepare the year-end trial balance and analyse the results.

By 2030, clients’ expectations will be “I want a trial balance every day”. I do not mean that literally, of course. They will want to see their business results, analysis and metrics in real time, presented in a way they understand and can explore as they choose.

Achieving that without AI is impossible.

And how much will it cost?

TLDR: Market commoditisation

This brings us to the part of the story that probably will not be anyone’s favourite.

If routine processing for a client takes less time, one important part of the cost of providing the service falls. If that client can also easily compare several agencies, competitive pressure increases. In my view, that combination will push down the price of simple, standardised services.

That does not mean all costs disappear along with manual data entry. Software, review, training, communication and cases requiring substantial work remain. Nor are two businesses with similar numbers of invoices necessarily equally demanding to serve.

If the same five people stay at the agency, fewer hours spent on data entry will not, by itself, reduce the monthly payroll. It is worth checking what the freed-up time actually delivers: new clients, an additional service or less overtime.

Knowing what a particular group of clients really costs to serve will therefore become increasingly useful to the owner. How much time goes into routine processing, how much into requests for missing information, and how much into conversations and corrections? Only then can they assess what automation changes and what price makes sense.

One agency may decide to take on more straightforward clients with the same team, while checking that quality and deadlines can be maintained. Another will develop a more expensive service for a particular industry. A third may draw a clearer line between basic processing and additional work that is agreed and charged for separately.

Suppose a client wants someone to monitor outstanding items during the month, check which invoices remain unpaid, and talk through recurring problems with them. When the data is more up to date, such a service is easier to organise. The agency could try it with a few interested clients and see whether it is useful enough for them to pay for.

That is an important test, in my view. Having more free time does not mean the client wants to buy every new service we invent. We need to talk to them and see where our help actually solves a problem.

I would be especially careful to give the promise “we look after your business” concrete meaning: what we review, how often, when we get in touch and who gets involved if there is a problem. The client then knows what they are paying for, and the agency can organise its people and estimate the cost of providing that service.

Large agencies or many small ones?

TLDR: AI makes tools more accessible, creating opportunities for both small and large agencies

One possible outcome is the emergence of larger agencies with good software, consistent procedures and teams for different types of work. Such an organisation could serve many clients, charge less for each individual service and still do well.

It is easy to take that thought further and conclude that the big players will take over everything. Yet the same software can also be a tremendous help to a small agency.

If an experienced bookkeeper gets software that reliably handles much of the routine work, they could serve more clients with less administrative help. If they also work with a particular type of business, repeated tasks and familiar situations make organisation easier still. They will need cover when they are away and access to a specialist for questions beyond their expertise, but they do not have to employ all that support permanently.

It is therefore possible for a large number of small, independent agencies to use a handful of software platforms. Software consolidates while service providers remain numerous. Larger organisations and specialists in more complicated cases operate alongside them.

That is why I can quite easily imagine there being more agencies than there are today. Some people may use technology to start their own businesses, while others join forces or work in larger organisations. Which model suits whom will depend on their clients, knowledge and way of organising work.

Back to the office

At the beginning, we left five people in an agency where software had freed up part of the working day.

The owner could start quite modestly. Choose one group of documents or a few similar clients and, together with the employees, look at where time goes today. Then try a different process, record how many checks and corrections were needed, and compare the result. If time is freed up, agree straight away on what to use it for.

They will then have their own experience to base decisions on. Along the way, they will also discover what they need from the software, who in the team wants which kind of work, and whether clients notice any improvement at all.

I am really interested in what this looks like from your agency. If the time you spend on routine processing were cut in half, where would you direct your people first? Towards new clients, better support for existing ones, training younger colleagues, or a task you never seem to get around to today?

If you have already tried any of this, even better. I would love to hear what actually changed once the program had done its part of the job.