Bookkeeper
A bookkeeper records and organises a business's money - the sales, purchases, bills and payments - so the accounts are accurate and up to date. It is careful, numbers-based work that suits people who are organised, trustworthy and like things to add up neatly.
How this job changes over time
Our best estimates, shown as ranges and grades, not exact predictions.
| WHAT CHANGES | Now | 5 yrs | 10 yrs | 20 yrs |
|---|---|---|---|---|
| Tasks AI is used for | 52% | 70% | 82% | 90% |
| Number of jobs | 90-100% | 70-90% | 50-75% | 25-60% |
| How hard to get in | C - hard | D - very hard | E - extremely hard | E - extremely hard |
| Job security | Weak | Weak | Weak | Gone |
Right now, AI is already used for about 52% of the day-to-day work in this job, and by 20 years from now that could be around 90%. There are likely to be fewer of these jobs over time, very roughly 25-60% of the 2024 number, 20 years out. Getting your first job here is moderately hard today, and it looks set to get harder. We have not yet classified what structurally protects this job, so treat the security row as our general model for jobs at this exposure level.
What happens, and when
At this horizon the model has AI used for about 70% of this job's day-to-day work, and the number of these roles is projected at roughly 70-90% of the 2024 total. Getting in is graded D - very hard, and job security here is rated Weak.
At this horizon the model has AI used for about 82% of this job's day-to-day work, and the number of these roles is projected at roughly 50-75% of the 2024 total. Getting in is graded E - extremely hard, and job security here is rated Weak.
At this horizon the model has AI used for about 90% of this job's day-to-day work, and the number of these roles is projected at roughly 25-60% of the 2024 total. Getting in is graded E - extremely hard, and job security here is rated Gone.
How to aim for this career
When you start, AI will do most of the quick, routine work. So build the parts that stay human: good judgement, working well with people, and knowing one subject deeply.
The people who do well will use these tools to get real results, not fear them. Start now: try AI tools on school projects and learn how to guide them to a good answer.
Aim for the part of a job where a person still decides what to make, checks the work, and owns the result. Keep your options open too, since the same skills suit nearby jobs that AI changes less.
A note on the security row: we have not yet classified what structurally protects this specific job, so the grades above use our general model for careers at this exposure level. The exposure score itself is a real, curated figure for this line of work.
Careers that use similar skills, and hold
Worth a look if you like the sound of this path. Each one has something this job does not: a moat.
Sources: exposure blends published AI-usage research (Anthropic 2026, observed real-world usage; OpenAI 2026) with our own UK reviews of labour-market moats, robotics evolution and scientific developments. Job-security forecast from OpenAI, "The AI Jobs Transition Framework" (Richmond, 2026, OpenAI Economic Research), CC BY 4.0. Grades and verdicts are Careermash editorial judgment: we show forecasts as forecasts and own our conclusions.