Cathy Brown, ACA

Founder, Cathy Brown Advisory

While I'm building out the advisory service, I've created two free tools that give you a starting point for thinking about AI and automation costs in your business.

Two free tools to get you started

Before you build any AI automation, most people want to know what it will cost to run and whether it will actually be cheaper than what it replaces. The honest answer is that even a straightforward two-step customer service automation is genuinely difficult to cost accurately, because token costs vary with every interaction in ways that human labour costs simply don't. Extrapolate that across multiple automations running simultaneously and the uncertainty compounds quickly.Whilst these tools won't give you a precise number, they will show you:

  • the right questions to ask;

  • the costs most vendors never mention;

  • some of the variables that will determine whether an automation makes financial sense for your business before you've committed to building it.

AI Token EstimatorModel the running cost of an AI automation step by step and understand why the number is harder to pin down than anyone selling you the build will admit.

AI Automation Cost EstimatorThe all-in cost comparison covering salary, on-costs, build, maintenance, emergency risk, and the customer churn that almost nobody factors in.

Other AI cost risks to consider

The two tools above model the costs of a specific, well-defined customer enquiry automation. The risks below apply more broadly, to any business deploying AI in any form.Deploying AI tools to employees without usage capsWhen AI tools are made available to employees, usage is driven by employee behaviour, not by a controlled automation pipeline. Without usage limits, a single month of uncapped access can exhaust an annual budget.Agentic AI and the token multiplierStandard automations, like the two-step pipeline modelled above make a fixed number of API calls per transaction. Agentic AI works differently, taking autonomous multi-step actions and looping back on results. Each task can trigger ten to a thousand times more API calls than a simple automation. The cost model for agentic AI requires separate analysis before any build decision is made.Falling token prices do not guarantee lower billsToken prices have fallen significantly in recent years. The reasonable assumption is that cheaper tokens mean lower costs. However, as tokens become more affordable, usage tends to increase further increasing budgeting complexity. Current token pricing also reflects a heavily subsidised market that is unlikely to hold indefinitely - business cases built on today's rates should account for that.Employee resistanceEmployees who fear job displacement, distrust AI outputs, or feel a tool has been imposed without adequate support will work around it or disengage. Change management and honest communication about what the system does and doesn't do are essential.Choosing AI because of hype rather than fitAn honest assessment of whether AI is the right answer is worth more than any cost model. The question isn't just what will this cost to run, but what problem does this solve, how will we know if it's working and what happens if it does not.

About

Twenty years in finance across a broad range of industries, from audit at KPMG through to senior finance roles in private equity-backed and listed groups with revenues up to £1.3bn.A chemistry degree means I see businesses as process flows naturally, with inputs, outputs, dependencies, and reaction points. An ACA qualification means rigour and integrity aren't negotiable. An IBM Data Science Professional Certificate means I understand the technical layer.I've watched this pattern repeat since the dotcom era. A new capability arrives, vendors rush to sell it, and businesses spend money before anyone has properly diagnosed the problem. This wave is bigger than all of those combined, and getting it right matters more than ever.Cathy Brown Advisory exists for businesses that want someone in their corner who understands the technology, the financials, and the operational reality.

[email protected]
© Cathy Brown Advisory 2026