The AI Efficiency for Accountants Newsletter V1 E3 by Marcie D Terman | Business Development Director 24 August, 2026
Turning accountancy practice data into better workflow, cash efficiency and client service
For many businesses, the first experience of artificial intelligence was fairly limited. ChatGPT could draft a well-structured email or summarise a document. Copilot could search for information. A chatbot appeared in the corner of a website and attempted, sometimes rather inelegantly, to answer customer questions.
Sometimes it worked very well. Sometimes it left you raging because you needed information quickly and the answer you received was wrong, unhelpful or simply dumb.
If your experience of AI has largely involved an irritating chatbot refusing to understand what you are asking, it would be reasonable to wonder what all the excitement is about. But that is now an extremely narrow picture of what AI can actually do.
The more interesting development for accountancy practices is not simply that chatbots have become smarter and more useful. AI can now become part of the actual operation of a business, removing much of the daily grind of repetitive work that requires accuracy but little innovation or professional judgement. It can monitor processes, analyse patterns, compare information, identify exceptions, work across different software systems and, within boundaries determined by the practice, take action.
That changes the question. Instead of asking “What task can AI do for us?”, we can begin asking “What parts of our practice could work better in collaboration with AI?”
Start with the problem
An established accountancy practice already has technology in place. Its clients may use Sage, Xero, QuickBooks or other accounting packages. The practice may have its own practice-management software, payroll applications, tax systems, document storage, CRM, Microsoft 365, client portals and connections into HMRC reporting processes.
The problem is rarely that no software exists. More often, the inefficiency lies in the gaps between systems and in the amount of human effort required to make the connections.
Someone checks the accounting package. Someone analyses an email to gather information and updates the practice-management system. Another member of staff notices that something is still missing and schedules outreach to the client. Has the client responded? Who is responsible for checking? The job is now getting close to deadline, so somebody needs to make a call by close of business — TODAY.
Repeat variations of that process across all of your clients and pages of apparently small inefficiencies begin to accumulate.
The opportunity for AI is not necessarily to replace those existing systems. One use is to reduce the friction between systems, clients and staff, using professional time primarily when something falls outside the normal process.
A useful AI project therefore does not begin with a model, an agent or a chatbot. It begins with understanding how the practice actually works: how information moves, where delays occur, which systems matter, what information is sensitive and what outcome would genuinely improve the business.
What is actually happening inside your practice?
Every accountancy practice generates reams of operational data simply by going about its normal business. How quickly do clients respond? Where does work repeatedly stall? Which kinds of jobs consume more staff time than expected? Which clients regularly miss deadlines? Where is work being returned for correction? Which clients require disproportionate amounts of chasing? Are there particular stages where delays repeatedly occur?
Individual members of staff may already know some of these things. Someone will say, “That client is always late.” A manager may know that a necessary type of job always seems to take a great deal of staff time despite requiring relatively little original thought.
But there is a difference between anecdotal knowledge and being able to look across the practice and identify patterns systematically.
This is one of the areas where AI becomes particularly useful. It can analyse large volumes of operational information and identify patterns that are difficult to see ‘with the naked eye’. That could mean recognising that a group of jobs is consistently taking longer than expected, that a particular client is absorbing substantially more staff time than their fee covers, that a certain communication method produces poor response rates, or that one stage in a workflow repeatedly causes delays.
That information can then be used to improve the way the business operates.
This is not simply automation. It is operational intelligence.
From information to action
Once a system identifies something unusual, the practice can decide what should happen next. That does not mean giving AI complete autonomy. There is a wide range of possibilities.
The system might simply alert management: “Client response times have deteriorated significantly over the last three months.”
It could identify the pattern and recommend a response: “This client historically responds more quickly to WhatsApp than email. Consider changing the reminder sequence.”
This could be delivered practice wide or to specific clients, whichever was deemed more effective. The software could prepare that change and wait for somebody to approve it. Or, where the practice has established clear rules and is comfortable doing so, the system could make the adjustment automatically.
The same principle can apply throughout the practice:
Alert me. Recommend something. Prepare it for my approval. Handle it automatically within these rules.
The practice decides which level is appropriate. Efficiency does not require surrendering control.
This is also where AI agents become interesting. A chatbot waits for somebody to ask it something. An agent can be given a defined objective and permitted to carry out a series of actions towards completing it.
For example, an agent could monitor whether expected information has arrived, check information quality against criteria established by the practice, identify what remains outstanding, prepare or send an appropriate request, update the status of the job and continue monitoring it. If everything proceeds normally, much of that work can happen without a member of staff repeatedly checking it. If something falls outside the rules, the agent can stop and bring the matter to the attention of staff.
A simple way of thinking about the difference is:
A chatbot can answer a question. An agent can help move the work forward.
The agent, however, is only one component. It still needs to know which systems it may access, what the practice regards as normal, what it may do automatically and where a person must remain in control. The process design around the AI is every bit as important as the AI itself.
Build around what already works
This may be one of the most important points for an established practice considering AI: you do not rebuild your technology estate in order to use it effectively.
Clients using Sage or Xero or any other package need to be catered for. If the practice has software performing core accounting, payroll, practice-management or statutory reporting functions well, the sensible approach may be to retain it. Existing connections into HMRC and other required reporting processes probably should remain in place, unless they are performing poorly.
The opportunity is often to create an additional layer that makes those systems work together more intelligently.
An application might take permitted information from a client’s accounting system and compare it with another source or a previous period. A change may be perfectly reasonable — seasonal variation in a construction business, for example — or it may indicate something that warrants checking. Rather than requiring staff to search through everything manually, the system can bring the exception to the attention of the appropriate person.
The same application could draw permitted information from several systems and give management one operational view of the practice without replacing the underlying software.
The important questions become:
What must remain? What can we connect to? Where are the gaps? And where would a new layer make the process work better?
Some systems provide excellent APIs and are relatively straightforward to connect. Older or more specialised software can require a different engineering approach. That is part of the development work.
The value of bespoke software is often not replacing the systems you already have. It is making those systems work together more intelligently.
Bespoke development becomes particularly interesting when the inefficiency lies in the way your practice operates rather than in a standard accounting task that an off-the-shelf product already solves well. Different firms have different approval procedures, client-service standards, reporting requirements, specialist areas, staff structures and attitudes to risk. The gaps between standard products are often where spreadsheets, email chains, manual checking and workarounds begin to multiply.
A bespoke system allows the practice to decide what should be monitored, what constitutes an exception, who receives an alert, what can happen automatically, what requires approval and what should never be automated.
The objective is not to put AI everywhere. A good solution may combine conventional software, existing applications, databases, business rules, integrations, agents and AI models.
The objective is to solve the problem.
And because the technology is changing quickly, a well-designed system should ideally allow the technologies underneath it to be reviewed and improved over time rather than unnecessarily locking the practice into one model or one technical choice forever.
Security and control still come first
Greater capability makes control more important, not less.
Accountancy practices are quite rightly cautious about client information. Introducing AI does not mean indiscriminately sending identifiable data into public models. For some tasks, identifying information may not be required and data can be anonymised before processing. Other applications may use commercial AI services with appropriate contractual protections. For more sensitive work, a controlled or walled architecture may be more suitable.
The same principle applies to access and actions. If an agent only needs to read one category of information, it should not automatically have access to everything. If an action carries greater risk, human approval can remain mandatory.
The underlying rule is straightforward:
Give the system only the information, access and authority it actually needs.
We covered security in much greater depth in Issue 2 of the newsletter and will look at other aspects of security in later issues. Here it matters because the wider possibilities for AI do not change the fundamental principle: the practice remains in control.
The bigger opportunity
AI efficiency is not really about making one administrative task 20% faster. The larger opportunity is to understand how the practice itself is operating and improve it deliberately.
Where is time being lost? Where are clients experiencing unnecessary friction? Where is staff capacity being consumed by work that does not require professional judgement? Which processes repeatedly create problems? Where are several pieces of software working well individually but badly together? Where are profitable jobs becoming less profitable because of the way they move through the practice? Where would better information allow management to intervene earlier?
Some answers may involve a chatbot. Some may involve agents. Some may already be available inside software the practice owns. Some may require integration between existing systems. And some may require a bespoke application designed around the way the practice actually wants to work.
In many of the most useful cases, the person benefiting from the system may barely notice that AI is involved.
And perhaps that is the point.
The first generation of widely encountered AI encouraged businesses to ask:
What can AI do for us?
The more interesting question now is:
What could make our accountancy practice work better?
Could one of your practice processes be more efficient?
If there is a process in your practice that consumes too much staff time, creates repeated client friction, gives management too little visibility or simply involves far too much manual checking, it may be worth looking at what could now be automated, integrated or redesigned.
And if the opportunity involves several existing systems, a workflow particular to your practice, sensitive information or something that does not fit neatly into an off-the-shelf product, that is the kind of problem we work on at DATAFORT.
We start with the process and determine what combination of existing software, integration, AI, agents or bespoke development makes practical sense. Let’s have a brief meeting to see how we can help your efficiency.
