Budgeting for AI as the price model shifts

Setting an AI budget has become one of the harder planning jobs.

The technology is moving fast, most firms are already experimenting across several types of AI at once, and the commercial model underneath it all is shifting from a fixed annual licence to a charge based on how much you use.

Why the shift matters

That shift matters more than it first appears. A great deal of what firms pay for AI today is widely believed to be priced below its true cost, as providers compete hard for position. The reasonable planning assumption is that unit prices, and the real cost of heavy usage, will rise as the market matures. A budget built on today's prices, for usage that grows through the year, risks a shortfall precisely when the service has become something your teams rely on.

What this paper covers

This paper is a practical planning aid for that problem. It sets out why AI budgeting is harder right now, what it means specifically for financial services and insurance, the levers that protect your budget, and a phased way to plan through this cycle and into next year.

A budget set on today's rates may not survive contact with next year's.

FORTAY CONNECT

Three things have changed at once.

Software here was historically simple to budget for. That simplicity has gone — and understanding why is the first step to planning through it.

Software was historically licensed for each seat or each year, which made budgeting simple. AI capabilities are increasingly charged on how much you use, often measured in tokens — the small units of text a model reads and produces. Cost now moves with volume, with the length of each interaction, and with how the service is designed. It is no longer fixed at the point of purchase.

Current pricing is widely expected to be a floor rather than a ceiling. Much of it appears to be subsidised while providers compete for share, so the prudent assumption is that the cost of running AI at scale will climb, not fall, over a two year horizon. Budgets set on today's rates may not survive contact with next year's.

Most firms are not running one AI project. They are experimenting across declarative and generative AI, conversational tools and process automation, often in different teams. That is a healthy way to learn, but it scatters cost across the business so that no single person can say what the firm is spending in total or where it is going.

None of the usual duties pause when AI takes on the work.

In a regulated setting, the cost of getting AI wrong is higher — which is all the more reason to plan the economics carefully rather than discover them later.

The duties still apply

Treating customers fairly and delivering good outcomes, operational resilience, protecting personal data, and keeping records that can be reviewed and explained all still apply. Accountability stays with named people, and any automated decision that affects a customer needs to be explainable with human oversight where it matters.

Insurance adds its own edges

Claims handling, advisor and claimant productivity, and the quality and completeness of records all sit close to cost and to risk. Fragmented record keeping is a known driver of professional indemnity exposure.

The cost of getting AI wrong is higher here — plan the economics carefully rather than discover them later.

FORTAY CONNECT

Why economics come first

Because the stakes are higher, the discipline has to be higher too. Getting the cost model right is not a finance exercise sitting to one side of the programme — it is part of running AI safely in a regulated business.

Most of the control sits in your design, not a price list.

The largest avoidable risk in an AI programme is consumption that grows faster than the value it creates. The good news is that most of the control sits in your design and governance. The main levers are these.

Sending simple work to a premium frontier model is one of the most common and expensive habits. A smaller, cheaper model often does the job just as well for a fraction of the cost.

Compact models, some no larger than a set top box, are now capable enough for many everyday tasks and can cut token cost sharply while keeping data close to home. They rarely need to replace frontier models for the hardest work.

Verbose designs consume far more than lean ones for the same result, so efficiency is a design choice made early — not a saving found later.

A pilot should test a business question and have a clear point at which you stop, continue or scale. A pilot without limits drifts, and drift is where cost accumulates quietly.

Across every model and tool in use, one picture of spend, one set of guardrails, and alerts that make unexpected growth visible early rather than at the next invoice.

Avoid designs that lock you into a single provider's consumption model with no fallback, so that if terms or technology change, you can change with them.

Keep spend tied to evidence, phase by phase.

A budget cycle is the natural moment to bring order to this. A simple three phase plan keeps spend tied to evidence and gives you defensible numbers to take into the year.

Baseline, headroom, quick wins.

Standardise and govern.

Scale proven value.

Independent advice on the cost side.

We are not tied to any single platform or provider, so our guidance sits on your side of the table — which matters most where cost and commercial models are involved. We work day to day at the intersection of AI, customer experience and technology in financial services and insurance. Our focus is practical and increasingly on efficiency: helping organisations plan, build and maintain these systems, and just as importantly keep them affordable as pricing shifts. The first step is not a demonstration or a sales engagement — it is a short, no obligation conversation, around 20 to 30 minutes, on where you are, where the market is heading and how to shape the cost side of your plan.

Book a 20-minute conversation → Email Fortay Connect

Mark Taylor · Fortay Connect mark@fortayconnect.com · 0161 240 3411 Independent advisory for regulated contact centre organisations. Produced by Fortay Connect, 2026. This document is provided for general information and does not constitute regulatory, legal or financial advice. Firms should confirm current supervisory expectations before deployment. www.fortayconnect.com

Fortay Connect Ltd

Independent AI advisory for regulated firms