AI Time Investment: Spend Time on AI Before It Saves You Time

*Part 4 of 6 in a series on the habits that make AI useful. Previously: [Learning AI: The People Who Learn Fastest Expect It to Fail](/insights/learning-ai-expect-it-to-fail/). Next: [AI Adoption Is a Management Challenge, Not a Technology Project](/insights/ai-adoption-management-challenge/).* "I haven't got time for AI" is the most honest objection I hear, and the hardest to argue with in the moment. Every AI time investment feels like a cost long before it becomes a saving. The person saying it is not being difficult. They are running a department, covering for someone on leave, and answering emails at nine at night. Being told that a new tool will save them time sounds, from where they're standing, like being handed more work. They're right about the short term. AI usually costs time before it gives any back. That's the trap, and it's worth naming properly, because the people who would gain most from the time saved are exactly the people who can't find the time to create the saving. ## The productivity paradox in one sentence People stay busy doing manual work because they're too busy doing manual work to look at how it could be done differently. Every business has a version of this. The finance manager who rekeys the same figures every month knows it's daft. There has never been a month quiet enough to fix it. The urgent work wins, every week, forever, because urgent work always wins by default unless someone decides otherwise. ## AI almost never saves time on day one The honest version of an AI improvement looks like this: - learn enough about the tool to be dangerous - understand the process properly, including the exceptions - test two or three approaches - build a template or set of instructions that works reliably - connect it to wherever the information lives - write down how it works - show other people Only after that does the saving start. Anyone promising instant transformation is selling something, and usually a licence. The first few weeks feel worse, not better. That's normal, and knowing it's normal is what stops people abandoning it in week two. ## Think about return on time, not time spent You don't need a business case with a discount rate. You need a rough sum. Take a task that takes 30 minutes and happens four times a week. That's two hours a week, roughly 90 hours a year. Say you spend four hours investigating it and get it down to 15 minutes. You've bought back an hour a week — the investment pays for itself in a month, and everything after that is free. Now run the same sum on a task that takes two hours but happens twice a year. Four hours of work to save one. Don't do it. That simple comparison — frequency times time saved, against the effort to set it up — is enough to sort a list of ideas into sensible order. Our [AI Strategy & Roadmap](/services/ai-strategy-roadmap/) work is largely doing this properly across a business, but you can do a rough version on a whiteboard in an hour. ## Don't try to transform everything "Implementing AI across the business" is a phrase that reliably precedes nothing happening. It creates a programme, a steering group, a procurement exercise, and no change to anyone's Tuesday. Look instead for work that is repetitive, frequent or frustrating. Reliable starting points in most SMEs: - meeting notes and follow-up actions - research and background summaries - routine reporting - first drafts of anything - classifying or tidying incoming data - recurring customer or supplier communications - comparing documents — contracts, specs, quotes - finding information buried in your own files None of it is glamorous. That's rather the point: it's the work that quietly consumes a day a week across a team. ## Protect the time or it won't happen Adoption needs a slot in the diary. One hour a week, one person, one real workflow. Whether that's Friday morning or the last hour of Monday matters less than it being booked and defended. Without protected time, the operational work will consume it. Not through anyone's fault — that's simply how a busy week works. ## Leaders have to make learning time legitimate Here's what quietly kills adoption in an SME: staff believe experimenting is not real work. If someone thinks they'll be asked why they spent an hour "playing with ChatGPT" while the queue built up, they won't do it. They'll do it at home, on their own account, with company data, which is a worse outcome for everyone. Management has to say out loud that sensible experimentation is part of the job, then ask about it in one-to-ones like any other objective. That single signal does more for adoption than a training budget. It's the theme the [next article](/insights/ai-adoption-management-challenge/) picks up. ## Measure whether it actually got better Don't assume the saving is real. Before you start, note how long the task takes, how often it's wrong, and how much rework it generates. Afterwards, check the same things — plus how much human intervention the new version needs. Sometimes the result is clear: three hours a week back, same quality. Sometimes it's less flattering. The drafting got faster and the checking got longer, so the work moved rather than disappeared. Occasionally the output looks better and is subtly worse, which is the one to watch for. Knowing which of those you've got is the difference between a business that compounds small gains and one that has a lot of subscriptions and a vague feeling of modernity — the [subscription graveyard](/insights/subscription-graveyard-ai-tools/) is full of tools nobody measured. ## The businesses that get ahead It won't be the ones that bought the most tools. It'll be the ones that gave a few people an hour a week, an explicit blessing to experiment, and a habit of checking whether the work actually improved. That's a management decision, not a budget one, and you could make it this week. ## FAQs ### How much time should I allocate to learning AI? Start with one protected hour a week per person involved, focused on one real task rather than general learning. It's small enough to survive a busy period and consistent enough to compound. ### Which tasks should I investigate first? Frequent, repetitive, text or data heavy tasks where you can quickly judge whether the output is right. Meeting notes, recurring reports, first drafts and document comparison are the usual early wins in an SME. ### How can I calculate whether an AI use case is worthwhile? Multiply the time the task takes by how often it happens to get the annual cost, estimate the realistic saving, then compare that with the hours needed to set it up and maintain it. Frequency matters more than task length. ### Should employees be given dedicated AI learning time? Yes, and it should be explicit. Without stated permission, people either skip experimentation or do it privately on unapproved tools, which creates risk without creating capability. ### How quickly should an AI experiment deliver a return? For a small workflow, expect signs of value within a few weeks and a measurable saving within a quarter. If it's still consuming time after that with nothing to show, stop and record why — that's a useful result too. ### How do I know if AI is genuinely saving time? Measure the same task before and after: time taken, error rate, rework, and how much checking a person now does. If the work simply moved from drafting to reviewing, you've changed the shape of the job rather than reduced it.

This article was written by Peter Lowe. The ideas and opinions are his own; AI was used to assist with drafting and editing.