Short answer
Per-seat AI tools improve the speed and quality of individual work. They rarely show up in the accounts, because minutes recovered across a day get absorbed rather than reinvested — which is a reason to buy them deliberately, not a reason to avoid them. In practice a small company needs one general assistant on a company plan, plus at most one or two specialists where a general tool is weak. Once a task is repetitive, rule-bound and happens whether or not anyone is at their desk, it has stopped being a productivity question and become an automation one.
Productivity is not automation
These two get sold together and they behave nothing alike, which is why budgets for them go wrong.
A productivity tool returns time to a person. Whether that time becomes value depends entirely on what they do next, and the cost recurs every month for every seat. An automation removes work from the company: it runs on a schedule or an event, produces the same output at three in the morning, and costs a build plus a small running fee. The first improves how work is done; the second changes whether it is done by a person at all — the territory covered in our list of automation workflows to deploy.
| Productivity tool | Automation | |
|---|---|---|
| What it changes | How a person works. | Whether a person does the work at all. |
| Cost shape | Per seat, every month, for as long as you keep it. | A build, then a small running cost. |
| Where the return depends | On what people do with the time returned to them. | On the process itself — it runs at three in the morning either way. |
| Visible in the accounts | Rarely. | Where it works, yes — and measurably. |
Both are worth having. Confusing them produces the two classic mistakes: expecting a subscription to deliver a structural saving, and building a bespoke workflow for something a general assistant already handles.
The categories that earn a seat
Stripped of the marketing, the tools that get used past the first month in companies of ten to a hundred people fall into a short list.
- A general assistant for drafting, rewriting, summarizing and thinking out loud. This is the one that gets daily use and the one worth choosing carefully, because it absorbs most of the demand the others would serve.
- Meeting capture — recording, transcript, and structured actions. The clearest case where a specialist beats a general tool, mostly because it removes the step of remembering to do anything.
- Search across your own documents, once finding things is a recognized daily irritation. The method behind it is described in our guide to retrieval over company documents.
- Function-specific tools where the work is genuinely specialized — code being the clearest example. Everywhere else, a general assistant with a well-written prompt usually matches the specialist at a fraction of the cost.
What consistently does not earn its place: a second general assistant bought by a different team, and any tool whose demonstration is impressive but which nobody can name a weekly use for.
Why the saved hours disappear
Individual time savings are easy to observe and hard to bank. Twenty minutes recovered from writing an email does not accumulate into a day; it gets absorbed into the other work waiting, or into the slack that made the day survivable.
This is not an argument against buying the tools — quality and speed of individual output are worth paying for on their own. It is an argument against justifying them with a projected headcount effect, which is the claim that will be checked and will fail. Where recovered time does convert into something visible, it is because it was deliberately pointed somewhere: a role that stops needing a contractor for overflow, or a service level that improves because responses now go out the same day.
A practical test before signing: if this tool works exactly as promised, what will be different in three months that a person outside the team could notice? If the only answer is that work feels easier, buy it for that reason and stop building a business case around it.
Tool sprawl, and the shadow version
Left alone, a company this size accumulates AI subscriptions the way it accumulates project trackers. Marketing buys one, an engineer expenses another, someone signs up for a note-taker in a meeting. Each is individually defensible and the aggregate is a set of overlapping subscriptions, several places company information now sits, and no view of who is using what.
Underneath that runs the unapproved version: people using personal accounts because the sanctioned option is missing or takes three weeks to request. Prohibition is the response that reliably fails, because the underlying need is real and the workaround is a browser tab. What works is making the approved path fast — a tool available on the first day, a short written rule about what may be pasted into it, and a named person who can say yes to a new one.
Choosing one assistant rather than five
Since a general assistant absorbs most of the demand, this is the decision worth spending time on. Four criteria matter more than a feature comparison.
- Where the data goes and for how long — a business plan should say plainly that your inputs are not used to train the provider’s models. Confirm it in the terms rather than in the marketing page.
- Administrative control: accounts you can create and remove centrally, and a view of usage. This is the concrete difference between a company plan and a pile of expensed licenses.
- Where it already sits. An assistant inside the tools people have open all day gets used; one behind a separate login gets forgotten in week three.
- Whether you could leave. Prompts and instructions you can export, and no critical process depending on one provider — the same portability logic set out in our comparison of open-source and proprietary models.
What leaves the company through these tools
Daily tools are where confidential information most often goes somewhere unintended, and it is almost never malicious. A contract pasted in for summarizing, a client list dropped into a spreadsheet assistant, a recruitment note run through a rewriter.
Two measures handle most of the exposure, and both are cheap. First, one page in plain language stating what must never be pasted into an external tool — typically personal data about clients and staff, anything under a confidentiality clause, and credentials. Second, company accounts under business terms rather than personal ones, so the rules governing your data are ones you have actually read. The wider picture is in our note on AI and cybersecurity for small business.
A rollout that survives month three
Across LYVIA’s own rollouts the same pattern repeats — enthusiastic launch, a training session, high usage for the first couple of weeks, then a slow decline until only the people who would have found the tool anyway are still using it. That is an observation from client work rather than a published benchmark, but it is consistent enough to plan against.
What changes the curve is anchoring the tool to specific recurring tasks rather than announcing a capability. One team, three named jobs it should be used for, a shared place where people put prompts that worked, and a review after a month that is willing to conclude the tool is not for everyone. Skills matter here more than licenses, which is the subject of our guide to training your team.
When the answer stops being a subscription
There is a recognizable point where a productivity tool is the wrong instrument. It arrives when the same task is being done repeatedly, the rule for doing it is explicit, and it needs to happen whether or not a particular person is available.
At that point paying per seat for someone to do the task faster is more expensive than removing the task, and it keeps the work dependent on a human being present. Recognizing that boundary early is most of what separates companies that spend well here from those that renew a growing stack every year — the sequencing question is handled in our AI strategy roadmap.
Frequently asked questions
What is the difference between AI productivity tools and AI automation?
A productivity tool gives time back to an individual, who then decides what to do with it. An automation removes a piece of work from the company, whether or not anyone is at their desk. Both are legitimate, but they behave completely differently on a budget: productivity is a per-seat cost that shows up every month, and its return depends on what people do with the recovered hours. Automation is a one-off build with a running cost, and its return is measurable in the process itself.
Do AI assistants really save hours per week?
Individual time savings are real and easy to observe. What is much harder is turning them into anything visible in the accounts, because saved minutes spread thinly across a day are usually absorbed rather than reinvested. The honest framing is that assistants raise the quality and speed of individual work — which is worth paying for — and that anyone promising a headcount effect from a per-seat license is selling the wrong thing.
How many AI tools should a small company have?
Fewer than it will naturally accumulate. One general assistant that most people use, plus at most one or two specialists where a general tool is genuinely weak — meeting capture and code are the common exceptions. Beyond that, each additional tool adds a subscription, a place company data can sit, and a thing to explain to new joiners, while overlapping heavily with what you already pay for.
What is shadow AI and why does it matter?
It is employees using AI tools you have not approved, usually with personal accounts, because the sanctioned option is missing or too slow to get. It matters because company information leaves through a door you cannot see, and because you lose any record of how the work was produced. Banning it rarely works; the reliable fix is making an approved tool easy enough that the unapproved one is not worth the trouble.
Should we buy a company plan or let people expense individual licenses?
A company plan, in almost every case, once more than a handful of people are using anything. Individual licenses on personal accounts mean company data sits under terms nobody has read, access does not end when someone leaves, and there is no administrative view of who is using what. The per-seat price difference is small next to those three problems.
How do we tell whether these tools are worth the money?
Not by surveying perceived time saved, which is consistently generous. Look at usage — how many licenses are actually used weekly, three months in — and at one or two concrete artifacts, such as whether first drafts of proposals now arrive faster. As a rule of thumb from LYVIA's own engagements rather than a published benchmark, weekly usage that drops off sharply after the first month is the signal to reduce seats rather than to run more training.
If you are already paying for several AI subscriptions and cannot say which ones changed anything, that audit takes very little time and usually pays for itself. Book a call.
