Reviewing your AI subscriptions to get the best returns

Say a paralegal pays $30 for extra AI use so a piece of client work gets finished. It goes on a personal card and is claimed back at the end of the month. Small purchases like it can happen across a firm.

The main AI subscription is usually easy to name, along with what it costs. So what is the firm spending on AI in a month, once the extras are added in?

1The plan and the extras

Some of the extras sit inside the main plan. The models offered at each tier, the limits on each and what counts as an add-on move as vendors release new models.

With Microsoft, for example, Copilot Chat is available at no additional cost with eligible Microsoft 365 plans, and custom agents used in it are charged on a metered basis. Microsoft 365 Copilot Business, for organisations with fewer than 300 users, is an add-on to a qualifying Microsoft 365 plan. Microsoft's Australian list price is AU$31.40 per user per month paid yearly, excluding GST, with a first-year promotional price of AU$26.91 until 31 December 2026.

Many AI plans also include a set amount of use each month and charge for more, which is where the paralegal's $30 went. Small amounts add up: across 50 people, an extra $20 each is $1,000 a month.

The model a person picks can change usage costs a lot. On Anthropic's published developer prices, its most expensive current models cost ten times as much as Claude Haiku 4.5 for the same amount of text. Subscription plans count use in their own ways, so the plan's terms show what each model costs the firm. Defaults can be set to the most capable model on offer, and they stay that way unless the person using the tool changes them.

Outside the main plan, those extras reach the firm as a line on a card statement or an expense claim.

2Where the spend lands

Card statements and expense claims are where Zylo found AI tools coming in. Its 2026 SaaS Management Index, built on more than 40 million software licences and US$75 billion of spend, found AI-native applications are the fastest-growing category of software spend, up 108% in a year, and that they frequently enter organisations through employee credit cards and expense reports, long before IT teams can respond.

Expense-based software spend rose 267% in a year, and ChatGPT is the most expensed application. Business units control 81% of software spend, and IT directly manages 15%. Zylo's figures come from large organisations, and the same route through cards and expense claims is open to people in a firm your size.

The paralegal's $30 arrives on an expense claim. Card payments and expense claims are built for small one-off purchases, so AI bought that way can go around the way the firm chooses its software.

3Reviewing your AI subscriptions

One way to close that route is to stop AI purchases on cards and expenses altogether. Stopping them does not remove the reason people were buying the tools.

The review

Your firm can bring those purchases into one place in six steps.

1. Run the AI audit first. Ask your people which AI tools they use, for which work, and whether they pay for any of it themselves, including the tools they found on their own, known as shadow AI. Then pull the last few months of card statements and expense claims that mention an AI tool.

2. Write down what the main plan covers: which models, at which tier, with what limits, and what happens when a limit is reached. Then check whether the tier fits the people using it, and whether individuals are buying their own upgrades around it.

3. List the AI that comes with software the firm pays for. The practice management system may have AI features in its current licence, and so may the document management system, the accounting package and any Microsoft or Google subscription the firm holds.

4. Look for the same job bought twice. If several people are paying for different tools to do the same job, check whether one plan can cover the group. Marketing, finance and the fee earners may each need something different, so keep what is working for each team.

5. Put AI subscriptions through the firm's software approvals. A named person approves each one, such as the person who owns AI in the firm, and the price, the owner and the renewal date go on the approved list. Before anything new is bought, one question gets asked: does something the firm already pays for do this?

6. Agree which model suits which kind of work, for example the smaller model for routine drafting and summaries and the larger one for complex matter work, and set the defaults to match where the tools allow it.

The next time the paralegal needs more AI use for client work, there's a way to ask. The request goes to the person who approves tools, and once it's approved, the tool goes on the firm's account and onto the list. If a partner asks what the firm uses and what it costs, the answer is one list with the renewal dates beside it. Each month's AI spend comes through the approved accounts, so reconciling it can become part of the normal month-end.

On the approved list, people can see which tools the firm has chosen, which model suits which job and who to ask.

4Where we come in

Deciding which tools go on that list, and what structure suits a firm your size, takes a closer look at how the work is done. The audit and the subscription list are yours to build, and we can help with that decision.

Our investigation looks at how your people use AI across the firm and measures how long common tasks take, so the firm starts from a baseline of its own. The findings belong to your firm whether or not we go any further together.

If you would like to know what your firm spends on AI and where it goes, reach out to our team. The first conversation works out whether your firm is the right fit for the AI Uplift Program, and whether we are the right people to help.

Reviewing your AI subscriptions