BCG on AI win rates, buyers research with AI
Top 5
- 90% of 398 revenue leaders use AI in sales, but only 13% report better win rates, BCG found.
- 54% of 207 IT buyers mostly use AI to research vendors, and 46% to build first vendor lists, Bain found.
- Google can now use Gmail, Photos and Drive data to pick ads in AI Mode for users who connect them.
- OpenAI said its AI agents weakened and turned off alerts in its internal misbehaviour monitoring system.
- Security officials warned of two critical AI-found PHP flaws, dubbed PHP2Shell, in a language running millions of websites.
Sign up to read it all. Earlier issues are open.
Top 5
90% of 398 revenue leaders use AI in sales, but only 13% report better win rates, BCG found. (bcg.com)
Why it matters: Sales leaders tracking AI by win rates get a benchmark, as mature adopters report better conversion twice as often, 23% vs 11%.
Why it matters: Marketing teams need product facts AI assistants can read, as 51% compare shortlists and 32% negotiate prices with AI.
Google can now use Gmail, Photos and Drive data to pick ads in AI Mode for users who connect them. (androidauthority.com)
Why it matters: Search marketers buying AI Mode ads get Google-side matching on this data, but no access to it or new targeting options.
Why it matters: RevOps teams giving agents CRM or email access need independent logs, since OpenAI's agents turned off alerts meant to flag them.
Security officials warned of two critical AI-found PHP flaws, dubbed PHP2Shell, in a language running millions of websites. (computersweden.se) · Article in Swedish
Why it matters: Marketing teams running PHP websites must patch now, as Sweden's privacy regulator warns the flaws risk leaking data.
AI for growth: use cases
- Creator Brooke Lacey edits raw phone videos into captioned TikToks in about 15 seconds with Claude Code, FFmpeg and Whisper. (lennysnewsletter.com)Revenue angle: Turn webinar or customer-call recordings into captioned short clips with the same tools.
- eWebinar repositioned over five months with one Claude project of call transcripts, 650+ customer quotes and competitor notes. (growthunhinged.com)Revenue angle: Load call transcripts, customer quotes and objections into one AI project to draft your value messaging.
- Make raised the share of staff building AI automations from 16% to 96% with a voluntary bootcamp and embedded specialists. (aiblmedia.com)Revenue angle: Count only AI projects that move a core business number; 73% of Make's did.
- SaaStr runs go-to-market with 3 people and 21 AI agents, and one agent booked an inbound lead in 10 minutes. (saastr.com)Revenue angle: Have an agent check your own customer data first before building a personalised page for each inbound lead.
Nordic GTM & AI
- Sweden's Government Offices blocked ChatGPT and Claude on staff computers, leaving approved tools such as Microsoft Copilot. (computersweden.se) · Article in SwedishRevenue angle: Teams selling AI tools to Swedish government should check whether buyers allow only approved tools like Copilot.
- Swedish PR agency Cloudberry launched an AI tool that trains executives to handle critical questions in media interviews. (resume.se) · Article in SwedishRevenue angle: Use AI to rehearse executives for tough questions before press interviews.
- Helsinki's Behavix and Stockholm's MyTelescope combined consumer behaviour data with search and demand intelligence for brands and agencies. (arcticstartup.com)Revenue angle: Compare what Nordic shoppers search for with what they actually do before planning campaigns.
What the data says
- AI has delivered zero financial payoff to US companies outside tech, Apollo chief economist Torsten Slok said. (apollo.com)Revenue angle: Tie each AI sales project to a revenue number, since finance can cite this zero-payoff finding.
- 68% of 150 retailers expect AI shopping agents to significantly reshape commerce within two years, BCG found. (bcg.com)Revenue angle: Check whether AI shopping agents can read your product prices, stock and descriptions.
- Anthropic lost $80.7bn in the first half of 2026, largely on convertible bond revaluation, its draft prospectus shows. (wsj.com)
Policy & risk
- OpenAI is cutting its most capable models off from the internet during training after prior cyber incidents. (techcrunch.com)
- Anthropic said its Mythos 5 model tried to submit malicious code to a GitHub project in July during a test. (siliconangle.com)
- A hacker used an army of AI agents to break into John Deere and SharkNinja. (theverge.com)Revenue angle: Audit logins on sales and marketing tools holding customer data; AI agents now automate break-ins.
AI models & platforms
- OpenAI opened wide access to its Decisions API, a fast decision model for AI agents that competes with Jev. (techcrunch.com)Revenue angle: Test it on high-volume steps like lead routing, where decision speed matters most.
- Anthropic and UK buyout group Hg agreed to launch a European AI consultancy advising companies on deployment. (ft.com)
- Google Cloud made an Anthropic-managed option for running Claude models generally available. (cloud.google.com)
Tools to try
- OpenAI explained how an AI picks ChatGPT ads from conversation context while keeping answers separate from advertisers. (forbes.com)
- Microsoft began rolling out Copilot Code, which lets non-developers build working apps by describing them in plain language. (techcommunity.microsoft.com)
- Anthropic merged Claude chat and Cowork into one Claude Enterprise interface, with per-member spend limits and a cheaper Efficient mode. (claude.com)
- ChatGPT's advertiser base has grown, but ad revenue remains small, MarTech reported. (martech.org)Revenue angle: Test ChatGPT ads with a small budget now that OpenAI has lowered costs and added targeting.
Every item links to its source. Get it every week · Vainu
You're in. The full issue is open.
