How to keep your skills current when you are billing full days
6 minute read. Updated 2026-08-08.
Billing full days leaves no built-in time to learn, unlike a permanent role with training days baked in, so the gap between what you can invoice for and what you actually know widens quietly. Ring-fence around 1 day in 10, fixed to the same recurring day, and spend it building something rather than reading about it. Increase that allocation around genuine shifts, not every model release, and use contract gaps for focused rebuilding rather than job search alone.
The structural problem
A permanent AI engineer gets training days, conference budget and slack time built into the job, whether they use it well or not. A fractional consultant billing full days has none of that by default. Every hour spent learning is an hour not invoiced, which makes learning compete directly with income rather than sitting alongside it.
This is not a complaint about the model, it is the trade you made when you went independent: a higher rate, no employer-funded development time. Ignoring the trade does not make it disappear, it just means the gap between what you can bill for and what you actually know widens without you noticing.
Why this bites harder in AI work than elsewhere
The absence of a stable market for some of the most current titles is itself the evidence. Nothing is published for AI governance, AI risk and compliance, AI safety and evaluation, AI product, or AI training and enablement as job titles in the UK advertised market; the work exists, but it is being bought under other titles because the field has not settled on what to call it yet.
If you are billing today under a title like Machine Learning Engineer or Generative AI, the skills that title implies in 12 months' time will not be the ones it implies now. Someone billing full days on last year's stack, with no protected time to look sideways, will not notice the drift until a client asks about something they have not touched.
What does not work
Treating learning as something you will get to between contracts does not work reliably, because the gap between contracts is also the moment you are most focused on finding the next one, not on deep learning. Relying on conference attendance or reading newsletters does not work either; it produces familiarity with terms, not the ability to build something with them under a client's deadline.
Following every model release as it happens is its own trap. It feels like currency but produces breadth without depth, and clients pay for depth.
A concrete allocation
Ring-fence 1 day in every 10, roughly 2 days a month on a typical working pattern, as protected, unbilled time. Put it in the calendar on a fixed day, the same day each cycle, so it is not the first thing sacrificed when a client asks for more hours.
Spend it building, not reading. Reproduce a paper end to end rather than summarising it. Run an evaluation against a model you have not used on client work yet. Contribute to an open source tool in the part of the stack you are weakest on. The test is whether you would have something to show a client afterwards, not whether you understood an article.
Pick a single adjacent skill deliberately each quarter rather than reacting to whatever launched that week. Depth in one direction beats breadth across all of them, and it gives you something specific to say when a client asks what you have been doing with your unbilled time.
When to spend more than the baseline
Increase the allocation temporarily around a genuine shift, not a routine release: a new regulatory obligation landing, a change in the tools a client base is standardising on, a gap you have noticed across several clients at once. That is a signal worth 2 or 3 days, not the usual amount.
Between contracts, use the time deliberately rather than only searching for the next one. A focused week rebuilding a specific capability is worth more to your next pitch than 2 extra weeks of applying with an unchanged offer.
Making the case for it commercially
If a client questions why you are less available on your ring-fenced day, the honest answer is the right one: it is what keeps the expertise they are paying for justified in 12 months' time, not just today. Most clients who have worked with contractors before understand this quickly, because they have seen the alternative, a specialist billing confidently on knowledge that quietly stopped being current.
The cost of the unbilled day is visible immediately. The cost of not taking it shows up later, as a rate you can no longer defend.
What to do about it
- Ring-fence around 1 day in 10 as protected, unbilled time, on a fixed recurring day.
- Spend it building something, not reading about it.
- Pick a single adjacent skill each quarter rather than reacting to every release.
- Use the gap between contracts for focused rebuilding, not only job search.
- Increase the allocation temporarily around genuine shifts, not routine launches.
- Be direct with clients about why the time is protected; most have seen the alternative.
Questions people also ask
How much unbilled time should I actually set aside for learning?
A reasonable baseline is around 1 day in every 10, roughly 2 days a month on a typical working pattern, protected and fixed to the same day each cycle. Increase it temporarily around a genuine shift, a new regulatory requirement or a change in the tools your clients are standardising on, rather than reacting to every model release as it lands.
Isn't the time between contracts enough to keep skills current?
Not reliably, because that time is also when you are most focused on finding the next engagement, which competes for the same hours. Use part of any gap deliberately for rebuilding a specific capability, but do not rely on gaps alone; billing full days for months at a time without any protected learning time is when the drift sets in and goes unnoticed.
How do I justify unbilled time to a client who's paying for my availability?
Be direct about it. Protected time is what keeps the expertise they are buying current, not a personal indulgence, and most experienced clients recognise this because they have worked with contractors whose knowledge quietly went stale. If a client genuinely cannot accept any reduction in your availability, that is useful information about the engagement, not a reason to drop the practice altogether.
What should I actually spend the time doing?
Build something rather than reading about it: reproduce a paper end to end, run an evaluation against a model you have not used with clients yet, or contribute to an open source project in your weakest area. The test is whether you would have something to show a client afterwards. Pick one direction each quarter rather than following every new release that appears.
Where the figures come from
Every rate and salary quoted in this article is a median or percentile of figures advertised in UK job postings over the six months to 8 August 2026. They are not rates paid, and the gap widens at the top of a range.
- IT Jobs Watch, UK contract rates, 6 months to 8 August 2026, read 2026-08-08.
- IT Jobs Watch, UK permanent salaries, 6 months to 8 August 2026, read 2026-08-08.