In the News: AI Spend Runs 24% Over Budget, a Claude Sonnet 5.5 Outage and Facial Recognition Treated as Evidence

AI overspend hits cash flow, Claude Sonnet 5.5 errors across the API, UK police treat facial matches as evidence, a new Mexico tariff threat and RFID stock coun

By Avik Ghosh, Managing Founder, illuminis

Five stories from the past week that matter to anyone buying, running or regulating AI, with our take on each.

Companies That Overspent on AI Ran 24% Over Budget, and Cash Flow Took the Hit

On November 12, the Capgemini Research Institute released a survey of 500 senior executives at companies with more than $1 billion in revenue. Roughly 60% said AI spending over the past 12 months was significantly higher than expected or budgeted. Among those that overspent, spending averaged 24% above expectations, 21% went more than 30% over, and two-thirds reported negative financial consequences such as lower free cash flow. Six in 10 said they lack a full view of AI costs, and 61% expect to adopt a multi-model strategy within 12 months. Capgemini Invent CEO Volker Darius said organizations need to "use the most appropriate model for each task." Source.

Our take: Most companies responded by tightening budgets or cutting scope. That treats the symptom. Overspend comes from sending every task to the same model and paying again for rework, and a multi-model strategy only helps if someone picks the right model for each step and measures the cost to finish the work. That is the case for CompletionPrism™.

A Claude Sonnet 5.5 Fault Took Out the API and Claude Code for 73 Minutes

On November 16, Anthropic reported elevated errors for Claude Sonnet 5.5 and Claude Fable 5.2 from 07:32 to 08:45 UTC, affecting claude.ai, the Claude API, Claude Code and Claude Cowork. Anthropic deployed a fix for Sonnet 5.5 and other Claude models and marked the incident resolved at 09:04 UTC. Source.

Our take: An hour and a quarter is short for a status page and long for a customer who cannot use a product. A software vendor whose features run on one model passes every provider fault straight to its customers. Failover to another model should happen automatically and be part of the product, not a ticket. See the Embedded License.

UK Police Inspectors Found Officers Treating Facial Recognition Matches as Evidence

A report by His Majesty's Inspectorate of Constabulary and Fire & Rescue Services, covered on November 13, found that investigators often treated retrospective facial recognition results as evidence rather than intelligence, that some did not carry out enough corroborating enquiries before acting on matches, and that in a small number of cases this led to inappropriate investigative action. The inspectorate called governance "fragmented and underdeveloped," asked for a statutory code in the forthcoming Police Reform Bill, and found forces keeping custody images they should have deleted. Big Brother Watch said police are not logging their use of the technology. Source.

Our take: An AI output is a probability, and the inspectorate is asking forces to show that a person checked it before acting. That needs a record of what the system returned, what was checked and who decided, made at the time and not open to later editing. The same test is coming for AI used in hiring, credit and healthcare. See InferTrust™.

A Threat to Double Mexico Tariffs Shows Why Sourcing Plans Need Live Scenarios

On November 12, President Trump threatened in a Truth Social post to immediately double tariffs on Mexico if it does not stop migrant caravans heading to the United States, without saying how the increase would be applied. Mexico is currently subject to a 10% tariff following a forced labor probe launched in March. Mexican President Claudia Sheinbaum said no caravan has reached the border under the current administration. Three bills in Congress would restrict the president's tariff power if they become law. Source.

Our take: Procurement teams now face a tariff that could double overnight, and the power to impose it could be cut back in Congress. Neither outcome is certain, so the useful work is knowing in advance which suppliers, parts and contracts each scenario affects and what the alternatives cost. See SourceMind™.

JD Sports Reports 98% On-Shelf Availability After Moving to Frequent RFID Counts

Supply Chain Dive reported on November 13 that JD Sports, working with Zebra, rolled out RFID-enabled mobile computers to 414 stores across the UK and Ireland in about three months, and reports 98% on-shelf availability and 20% faster replenishment. Stores can now count stock twice a week or daily, keeping shelves and inventory systems in sync, and teams can spot loss sooner instead of waiting for a 90- or 180-day audit. Source.

Our take: Accurate counts are where good stock decisions start. Once a retailer knows what it holds, store by store and day by day, the next gain comes from moving stock to where it will sell before a shortage or a markdown forces the issue. See StockBalancer™.

Get This Every Week

In the News comes out every week. Subscribe from the Insights page to get it by email.