Your competitors announced far more AI than they run. The gap is your opening.
Three figures with their samples and dates, and the three questions that read any competitor announcement.
Three figures with their samples and dates, and the three questions that read any competitor announcement.
Written by Artem, chief executive, commercial and delivery at Ryon. Published 24 September 2026. Last reviewed 24 September 2026. Version 1.0. This page is versioned rather than rewritten, and changes are listed in the changelog at the foot.
In one sentence. Announced is not deployed means a public AI announcement describes a purchase, a pilot or a plan far more often than it describes an agent doing work that carries money, a client relationship or regulatory exposure.
What the announcement did not say
In the Bank of England and Financial Conduct Authority joint survey of 21 November 2024, 75% of UK financial services firms were using AI. Only 2% of their AI use cases were fully autonomous. The announcement you read last week came from somewhere in that gap.
An announcement can be one of three things. A licence rollout, where software was bought and switched on. A pilot, where a small group tests something in a quiet corner. An agent doing work in production, carrying real consequence every day. Almost none say which, and often the writer does not know either.
Question one. Is it a licence, a pilot, or an agent doing work
Nearly nine in ten respondents report regular use of AI in at least one business function. McKinsey, 1,719 participants across 97 nations, fielded May to June 2026, published August 2026. McKinsey sells AI services.
Near universal use is real. That is not the claim you are afraid of.
In the same survey, 37% attribute at least some EBIT impact to AI, essentially unchanged from 2025, and only 6% are high performers attributing 5% or more of EBIT. Use is nearly everywhere. Attributed profit is not.
So put the first question. Does it name a licence, a pilot, or an agent doing work in production. Name none of the three and it is a plan.
Question two. Who signed for it
In the Bank of England and Financial Conduct Authority joint survey of 21 November 2024, 55% of AI use cases in UK financial services involved some degree of automated decision making and 2% were fully autonomous. That survey came from a regulator with no commercial interest in the answer. Its fieldwork is older than almost every press release it explains. In the same survey, 84% of firms had a person accountable for their AI framework.
Read that against the announcement in front of you. Automated decision making was common. Full autonomy was rare. An announcement that names nobody describes something nobody signed for.
Autonomy should scale with how reversible the action is, and never past it. That is the Four Autonomy Tiers, live 29 September 2026.
What thirty four initiatives in a queue look like
One client had a 34 initiative backlog and nothing moving, because thirty four things in a queue is the same as nothing in a queue. We did not do thirty four. We picked five, built them, and trained forty two of their own people alongside the build. The first invoice agent removed 90% of manual entry. Measured, baselined against that client's own entry volumes before the work started. Eight agents are in production. First value arrived in 4 to 8 weeks against a previous estimate of 9 to 18 months. Client withheld at their request. Delivered, deployed or contracted in the past twelve months, across Europe, the UK, the US and the Gulf. Results vary by client.
The work had moved back into inboxes. Nothing in that engagement was announced while it was being built.
Question three. What share of the money went to agents at all
AI agents and assistants account for roughly 1% of the worldwide AI spending Gartner forecasts for 2026, while AI infrastructure accounts for over half of it. Gartner, forecast published 16 September 2026.
A forecast is a projection rather than a measurement. The money is going into capacity.
Announcements get written about capacity. Capacity is what a firm can buy inside a quarter. A competitor announcing an AI platform has told you what they purchased. They have not told you what it does on a Tuesday.
What this means for your own estate
Put the same three questions to your own announcements. Most firms doing that honestly find a licence rollout, two pilots and nothing in production. On the maturity ladder that is copilot. The self audit in the Ryon Standard takes you through the rest.
Most firms are honestly at copilot and for most of them that is the right rung this year.
Before anything of yours acts on a customer, settle three things. What it may touch. What gets logged. Who can switch it off. Least privilege, an audit trail and a kill switch. Ryon runs on two verticals. Client staff build the simple agents. Ryon's engineers build the sensitive and complex ones. Neither side does everything. What that work costs is published in the Ryon price ladder, live 1 October 2026.
Are our competitors' AI announcements actually deployed
Most are not, in the sense you mean by deployed. A public announcement usually describes a purchase, a pilot or a plan, rarely an agent doing work that carries money or regulatory exposure. The three questions above separate them in about five minutes.
Are we behind on AI compared with our competitors
Probably not in the way you fear. Near universal use of AI sits alongside a minority of respondents attributing any profit impact to it. The field is flatter than the announcements make it look. The question is which of your competitors has anything in production.
The opening
The gap between an announcement and an agent doing work is the only period in which a slower firm arrives first. It closes on the day somebody else's agent starts doing the work.
Book the free 2 hour diagnostic. You get one page naming where an agent would pay back and where it would not.
How to read the numbers on this page
Every number about Ryon on this page carries one of five labels. Measured means observed in a delivered engagement. Contracted target means written into a contract as an acceptance criterion. Designed target means designed and not yet measured. Published product claim means a forward looking claim about a Ryon product, not a client outcome. Comparable build means observed on a similar build rather than this one.
Every Ryon engagement is baselined against the client's own numbers before the work starts. Every figure on this page is measured against that client's starting position, not against an industry average.
Where a number below comes from a firm that sells the thing the number is about, this page says so in the same sentence. Where it comes from a regulator, a university or a standards body with no commercial interest in the answer, it says that too.
Notes
Ryon is an engineering firm, not a law firm. Nothing on this page is legal advice. Every regulatory date and obligation here is reported from the primary source named beside it and linked from it, and it was accurate on the date this page was last updated. Whether any of it reaches your business depends on facts we do not have. Take your own legal advice before you rely on it.
Client names are withheld. The architecture, the guardrails and the measured or contracted targets are exactly as they were. Where a number is a target rather than a result, it says so.
Sources
- Bank of England and Financial Conduct Authority, third joint survey on AI in UK financial services, UK financial services firms, published 21 November 2024. A regulator with no commercial interest in the answer. https://www.fca.org.uk/publications/research-notes/ai-uk-financial-services
- McKinsey, The State of AI in 2026, on the road to ROI. 1,719 participants across 97 nations, fielded 4 May to 8 June 2026, published August 2026. McKinsey sells AI services. https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/the%20state%20of%20ai/the-state-of-ai-in-2026-on-the-road-to-roi.pdf
- Gartner, worldwide AI spending forecast, published 16 September 2026. A forecast, not a measurement. https://www.gartner.com/en/newsroom/press-releases/2026-09-16-gartner-forecasts-worldwide-ai-spending-to-grow-49-point-5-percent-in-2026
Changelog
Version 1.0, 24 September 2026. First publication.