Before you buy another AI tool, use the one you have.
Most organizations we work with already pay for more AI than they use. An assistant came bundled into the email and document suite at the last renewal. Two people expense a chat subscription. The accounting software added a "draft with AI" button last spring, and the client database grew a summary panel that nobody has clicked. Then someone sees a demo at a conference and asks for a budget line for a new tool.
That request deserves a fair hearing, but it should come second. The cheapest AI you will ever get is the AI already sitting in your contracts, and the first job is finding out what it is and whether anyone can use it.
Count what you already have.
Start with the money, because the software list will mislead you. Pull a year of card statements and vendor invoices and mark every line that includes an AI feature, whether it is sold as a separate add-on or folded into a plan. Office suites, email, accounting, scheduling, the help desk, the CRM, the e-signature tool and the video meeting platform all belong on the list. A town office or a small clinic is often surprised by how long it gets.
Next to each line, write two things: how many seats you pay for and how many people actually open the feature. You may not have exact usage data, and an honest guess from the person who manages the account is good enough. Say a team of twelve pays for fifteen seats of an assistant that four people use each week. That gap is the first finding, and it costs nothing to close.
Why the features go unused.
AI features tend to arrive in ways that make them easy to miss. Some ship switched off and wait for an administrator to turn them on. Others are on but tucked behind an icon that looks like decoration. The announcement email went to whoever set up the account years ago, and that person may have left.
The bigger reason is time. Nobody was given an hour to try the thing on their own work, so it stayed a button. People also hold back when they are unsure what happens to the data they paste in, and in our experience that worry stops more use than any missing feature does. A one-page answer from whoever manages the account (what the tool can see, where it keeps it, what the contract says) clears a surprising amount of hesitation.
The cheapest AI you will ever get is the AI already sitting in your contracts.
Many of these tools can already do the ordinary jobs teams ask about: summarize a long email thread, draft a first reply to an intake form, turn meeting notes into a list of follow-ups, pull figures out of a stack of invoices. The question is whether anyone on your team has tried.
Test the tools against your own work.
Pick three or four tasks that happen every week and that people would gladly hand off. A school district office might choose answering routine parent questions and summarizing board packets. A manufacturer might choose drafting quotes from a spec sheet and cleaning up supplier emails. Write each task down in a sentence, with an example of the real input.
Then give one person a short, fixed window, say two weeks, to try each task in the tools you already have. Ask them to keep notes on what worked, what nearly worked and what failed outright. Choose someone who wants the tool to succeed and who knows the work well enough to judge the output. A skeptic will find the failures fast, which is useful later, but the first pass needs someone who will push past the first clumsy result.
At the end of the window, the notes will tell you more than any vendor comparison. Some tasks will be handled well by a feature you already own. Some will be handled badly. A few will turn out to need something your current tools cannot do, and those are the only ones that justify a new purchase.
When a new tool earns its place.
Sometimes the trial confirms that you need something new. A specialized tool for legal review, transcription of field interviews or reading engineering drawings may do a job the general assistants cannot. That is a fine outcome, as long as you can say what the gap is in one sentence and point to the notes that showed it.
Before you sign, check three things. First, confirm the new tool does not duplicate a feature you are already paying for and simply haven't switched on. Second, name the person who will own it, set it up and answer questions about it. Third, decide what you will cancel or downsize to make room. Organizations that add tools without retiring any end up with a long invoice list and the same unused features, one layer deeper.
A tool audit you can run this month.
The audit checklist
- List every subscription that includes an AI feature, from invoices and card statements rather than memory.
- For each one, record seats paid for, seats in use and the name of the account administrator.
- Ask each administrator whether the AI features are switched on, and turn on the ones your policy allows.
- Write a one-page note per tool on what data it can see and where that data goes, and share it with the team.
- Choose three or four recurring tasks and write each as one sentence with a real example of the input.
- Give one willing person two weeks to try those tasks in the tools you already have, and ask for written notes.
- Sort the results into three groups: handled well, handled badly and not possible with current tools.
- Cut unused seats at the next renewal, and move them to the people whose notes show real use.
- Consider a new tool only for the "not possible" group, with a named owner and something cancelled to make room.
- Put the next audit on the calendar for six months out, since vendors add features to existing plans all the time.
