Where AI is actually Moving Revenue

Yes, for existing Microsoft 365 Copilot customers. Message Center MC1476200 (updated October 1, 2026) says a new Copilot Business seat deploys usage-based billing across all Microsoft 365 Copilot subscriptions in the tenant. Direct Microsoft purchases switch on October 19, 2026. CSP purchases switch on December 1, 2026. The default spending limit is 4,000 Copilot Credits per user per month. Create your own spending policy before the next seat purchase. Full decoder

On Oct 8 2026 OpenAI put Ultrafast on gpt-6.1-sol. Short Ultrafast list is $12 input / $0.60 cached / $15 cache write / $60 output per million tokens, 6x Sol Standard $2/$0.10/$2.50/$10. Long Ultrafast (>272K) is $24/$1.20/$30/$90. Separate Ultrafast TPM: Build 1M / Launch 4M / Grow 40M. Full decoder

No. OpenAI says Meetings works for online calls or in-person conversations without a bot joining the call. It uses the Mac's microphone and system audio. Audio is deleted once notes are ready. Decide consent, admin settings, and an evidence rule before reps turn it on. Full decoder

On 100 "best B2B software" queries, Google AI Overviews cited 38% fewer vendor self-ranking listicles in September than in June. The brand that ranked itself #1 was left out of the recommendation 83% of the time (69% in June). In 70% of cases the listicle helped a competitor get recommended. Score cited, recommended, and competitor recommended as three lines before you renew. Full decoder

Google's Gemini API lists Nano Banana 2.1 image output at $0.0336 per 1K image, $0.0504 per 2K, and $0.113 per 4K, half of Nano Banana 2 at 1K and 2K and 25% less at 4K. Input rose from $0.50 to $1.50 per million tokens and text and thinking output from $3 to $7.50, so a full ad-variant request saves less than the headline. Full decoder

Claude Haiku 5.5 lists $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 over 100,000. At 1,000 input and 50 output tokens, a million lead decisions costs about $125, the same as GPT-6 Luna and a tenth of Haiku 4.5. Full decoder

Slack's pricing page now lists Enterprise+ at $45 per user per month billed annually, 3x Business+ at $15. Enterprise+ removes the Business+ cap of 15 Slackbot messages per user per week. Over the cap, Business+ pays Flex Credits: 40 per standard Slackbot action, about $0.20 at list. Full decoder

OpenAI opened the Decisions API in public beta on Oct 6 2026. It bills input tokens only: $0.10 per million on gpt-6-luna, with no output or cache charges. A 1,000-token lead decision costs about $0.0001. Ask vendors who charge per AI action to show the model cost under each routing, scoring, or tagging step. Full decoder

OpenAI cut API usage tiers from five to three: Build ($5 in purchases, $500 a month), Launch ($100, $5,000 a month), and Grow ($500, $200,000 a month). At $500 the monthly ceiling rose 40x from $5,000. Set your own hard spend limit before agents run. Full decoder

HG Insights launched its Contextual Intelligence Platform on Oct 6, 2026, priced on Intelligence Credits with HG Copilot modules from $22,000. Its pricing guide runs two meters: Intelligence Credits to export data (0.1 credit per firmographic and hierarchy record, 3 credits per buyer intent data point, 30x apart) and AI action credits for compute such as research briefs and CRM row enrichment. Only 20% of a month's unused credits roll over. Standard term is two years; one year and quarterly billing cost more. Seats come in bundles of 5, 10, 25, or 50. No free trial. Write credits per dollar, the full rate card, the meter for Copilots, Agents, MCP, and Fabric API, rollover, term, seats, AWS commit treatment, and a paid-pilot exit into the Order Form. Full decoder

Across 26 fixed natural-language prompts, Landbase reports 76.7% average precision for GTM-3 Omni, 47.9% for Clay, 36.0% for Apollo, and 23.2% for ZoomInfo, with Landbase winning 18 of 26 prompts [Verified primary]. Full decoder

Neither KACI nor WashU sets a fixed number. KACI ran two and found same-provider overlap between 0.319 and 0.664. Ask for three or more runs per prompt per period so a spread exists on your own panel, then count only changes above that band. Full decoder

SemiAnalysis (Oct 5, 2026) limit-tested both labs' plans: on Claude Opus 5.5 vs GPT-6.1 Sol, Anthropic's $200/month plan offers about 5x the API-equivalent value of OpenAI's $200 plan. Top tier is close: $2,897 of GPT-6 Astra exhausts the OpenAI plan; $2,485 of Fable 5.1 uses half the Anthropic plan. OpenAI halved its $200 plan value last week and added a $500 tier; eligible Pro 200 subscribers keep the old allowance through Oct 29, 2026 (OpenAI Help). Anthropic raised Opus allowances ~20% on Max and ~50% on Pro. A user maxing Opus 5.5 runs the plan at a modeled -369% gross margin, so allowances will keep moving. Write workload profile, allowance per model, allowance floor, change notice, grandfather term, premium-model cap, a metered-API route for heavy agent users, and a monthly re-test into the Order Form. Full decoder

GPT-6 Astra Ultrafast is broadly available in the OpenAI API (OpenAI docs retrieved 2026-10-02; NVIDIA Oct 1, 2026). Short-context list is $60 input / $6 cached / $75 cache write / $300 output per MTok, 6x Astra Standard ($10/$1/$12.50/$50). Call model gpt-6-astra with service_tier ultrafast. Default TPM 500k (tiers 1-3), 1M (tier 4), 5M (tier 5). US residency and global only; no EU regional. Up to 8x speed is a vendor claim. Name tier, rates, TPM, region, chargeback, and accepted-outcome cost before agents default to it. Keep Pro 500 seat packaging separate. Sol Ultrafast is now live as of Oct 8 2026; see https://www.therevenueaireport.com/blog/gpt-6-1-sol-ultrafast-6x-api-order-form. Full decoder

Gong Activate and Apollo's ApolloNEXT stack both pack data + priority + action into one vendor loop (PR Newswire, Sep 30, 2026). Put holdout design, named human owner, rollback rehearsal, export of context/rules/action history, and permission boundary on paper before agents get write or send permission. Gong Activate is future-tense - not clearly GA; no GA date. Apollo: Intelligence Layer live; Builder Studio and Messaging OS "available soon." Soft-open Activate GA, Builder/Messaging GA, meters, SKUs, buyer incremental revenue, coaching-era Gong numbers as Activate proof. Checklist is buyer advisory. Companions: Enrich/Deep Mode and ZoomInfo DoubleO (cross-link only). Full decoder

CTech (Meir Orbach) reports Oct 1, 2026 that Israeli startup enso raised a $15 million Series A led by MoreTech Ventures, with NFX, Phenomen VC, and Gil Hirsch participating; total funding $23 million. Model: AI agents for marketing and sales plus an engineer and a marketer embedded per customer, measured against a business and pipeline baseline. Put fully loaded cost, embedded labor covered, baseline locked, confound log, attribution limit, playbook vs bespoke, handoff demo, named human owner, permission boundary, accepted-outcome + stop rule, and renewal reverse test on the Order Form and pilot before you credit the agent alone. Headcount locked at 26 (CTech). Soft-open seat price, meter, SKU, overage, site tiers, Dynamo/Wiliot, experiment lifts as buyer ROI, valuation beyond disclosed totals, GA. Checklist is buyer advisory. Full decoder

ZoomInfo announced Sep 30, 2026 that it acquired DoubleO.ai, an agentic orchestration platform for multi-agent automation (Business Wire; VentureBeat reprint). Vendor plan: fold DoubleO orchestration into ZoomInfo GTM context so agents can run GTM motions from signal to action. Put export of workflow definitions, rules, and action history; named human owner; rollback rehearsal; messy-input tests; audit trail; permission boundary; accepted-outcome blank; and meter blank for AE on the Order Form and pilot before agents get write permission. Champion-tracking and lookalike flows are vendor examples, not buyer proof. Soft-open price, GA, separate SKU, headcount, named customers. Exit/audit language is buyer advisory, not already contractual. Companion credit-path cite: /blog/zoominfo-mcp-bulk-vs-ai-action-credits-order-form. Full decoder

Enrich is dual-wallet: BYOK partner results do not burn Gong credits (partner bills you); Gong-provided partners burn 1 Gong credit per successful email or phone return (per type), still charged if email fails validation. Empty lookups, skipped partners, connection errors, and account enrichment do not charge. Flow ruleset auto-enrich uses the same rules and sits outside personal monthly caps. Deep Mode estimates 1,000-3,000 credits per research on the same company pool (2,000 credits/year per paid core seat; purchased no rollover; $ top-up blank for AE). Name Enrich / Deep Mode feature caps, company pool, chargeback owner, and accepted-outcome blank before flow auto-enrich goes live. Prefer help.gong.io Enrich/Deep Mode docs over PR provider-credits-first marketing. Distinct from Sep 28 Trackers/MCP shared-pool cite. Full decoder

API model id is gpt-6.1-sol. Standard short list is $2/MTok in and $10/MTok out, one-fifth of Astra $10/$50. Cached input $0.10; cache write $2.50. Input over 272K bills the full request at long Standard ($4 / $0.20 / $5 / $15). Fast is 2x Standard. Batch/Flex is 50%. Regional is +10% where eligible. Sol Ultrafast is now live as of Oct 8 2026. See https://www.therevenueaireport.com/blog/gpt-6-1-sol-ultrafast-6x-api-order-form. Per-task benches including Terminal-Bench Science 0.1 (Sol about $5.47 vs Astra about $23.80 vs Opus 5.5 about $23.21) are VENDOR / UNPROVEN. Name gpt-6.1-sol on the Order Form, leave the Bedrock inference profile blank until docs name it, and name the accepted-outcome owner before volume. List fifth is not bill fifth. Full decoder

Sonnet 5.5 list is flat vs Sonnet 5: $2/MTok in, $10/MTok out; cache read $0.20; cache write $2.50 (5m) / $4 (1h); Batch $1/$5; model id claude-sonnet-5-5. Vendor "up to 30% less per task" / "30%+ faster" = VENDOR/UNPROVEN until buyer eval. Prove bill move with tokens/task (API effort default High; apps Medium; recalibrated), cache write vs read, Batch vs real-time, serving model after fallback, pre-output refusal category. First Sonnet with cyber safeguards: cyber + frontier_llm fall back to Sonnet 5; bio + reasoning_extraction + general_harms block. Apps auto-switch DEFAULT ON; API fallback NOT default. Until configured, API returns 200 + stop_reason refusal. Fallback billed at Sonnet 5 rates + cache-miss credit; pre-output refusal billed for bio/frontier_llm/reasoning_extraction, not cyber/general_harms. Chargeback = serving model + refusal category. Name apps auto-switch + API fallback owners before volume agents. Full decoder

Effective Mon Sep 28, 2026 MT, a Copilot code-review effort value of Default resolves to Balanced. Explicit Lite is not auto-switched. Private-repo reviews burn BOTH meters: AI Credits (tokens; 1 credit = $0.01) and Actions minutes (copilot-pull-request-reviewer / dynamic/agents/copilot-pull-request-reviewer). Public Actions minutes stay free. Pool: Business 1,900 / Enterprise 3,900 AI credits per user/month pooled; no rollover; reset 00:00:00 UTC on the 1st. Completions / next-edit do not burn AI Credits. Code review model auto-selected and not disclosed; do not invent Balanced vs Lite $/review or fixed minutes/review. Kill switches: paid-usage OFF (default-on otherwise), explicit Lite, AI Credits budgets, Actions budgets, named policy owner. Meter != merge quality / defect escape / review outcome. Full decoder

Each paid core seat includes 2,000 credits/year in a company-wide shared pool that resets on the contract year. Question-based AI Trackers, MCP (ask_account / ask_deal / generate_brief), APIs (ask-entity / get-brief), and Deep research burn that pool. MCP burn matches equivalent API requests. Does not burn: pretrained trackers, manual UI briefs, viewing already-processed data. Rate card (credits, not $): Email 0.1; Call ≤10 min 0.3 / >10 min 1; Brief web search 10. Purchased credits: explicit buy only; no auto top-up documented; expire end of term; no rollover; $ rate blank for AE. Monthly caps = member-initiated only; workspace allocation BETA = ceilings from shared pool, unallocated unused. Exhaustion: processing stops; MCP/API error; AI Trackers need manual resume after top-up. Put pool math, top-up, caps, workspace ceilings, kill/resume owner, and accepted-outcome blank on the Order Form. Meter ≠ accepted pipeline/revenue. Full decoder

MCP is included at no extra fee but burns existing pools. Bulk data credits: Enrich + audience ingest. AI action credits: Account/Contact Research, Conversation Intelligence (min 9), Audience Analysis (~5-15 typical). Free: Search/Lookup/Find Similar/Recommended Contacts. MCP requires bulk data credits (not recurring monthly). No cost preview. RUM 12 months org-level. Copilot viewing is not MCP enrich. Replit = interactive research only (no bulk export / CRM write via MCP). Press: 100 free AI credits on first MCP connect = press claim until AE. Put credit type, tool×surface map, per-user caps, dashboards, and kill path on the Order Form. Full decoder

List $/MTok fell. GPT-6 Sol $2/$10 and Luna $0.10/$0.50 (short context); cache read 10% of input; cache write 1.25x; over 272K input reprices the whole request (2x input+cache, 1.5x output); Batch/Flex 50%; Fast 2x; regional +10%. Opus 5.5 $4/$20; cache read $0.20 (down 60% vs Opus 5); cache write $5/$8; Fast $8/$40; Anthropic ~40% typical task cost is a vendor claim. Require OF/chargeback lines for effort caps, long-context ownership, Standard vs Flex vs Fast, cache write vs read, Anthropic fallback / serving model, and usage.iterations. Full decoder

On HubSpot Limited Release / flexible seats-and-credits, one shared monthly Credits pool powers AI agents and automation. Content Agent (Beta) 1,000, Revenue Agent (Beta) 500, Customer 50, Prospecting 100, Nurture (Beta) 10, plus email and workflows, all burn the same pool. Included: Starter 5,000 / Pro 10,000 / Enterprise 15,000. Monthly reset; unused expire; no rollover. PAYG €0.010 (increments of 10); Capacity €10 / 1,000 / mo; Agents Packs Base €25 (2.5k) / Growth €400 (40k) / Scaled €1,500 (150k). Put overage path + pause on the Order Form before Content and Revenue. Limited Release / Nordics-Benelux pilot; EMEA expansion intent; UK date not published. Full decoder

Everyday Copilot (Chat, Office apps, built-in agents, Auto) is the Microsoft Copilot USL seat line. Cowork, Code, Autopilot, and Managed Runtime run on UBB Copilot Credits (PAYG $0.01/credit; Credit Pack $200/25k billed annually; P3 unused expire; MACC). Cowork burns models + runtime + context + tools. Chargeback: tenant → BU → agent → accepted outcome. MC1470425 Auto-apply new UBB services is ON by default (go-live gate). Home/Code Frontier; Autopilot private preview. Full decoder

You only know if you measure citation share on each of six surfaces (ChatGPT, Claude, Gemini, Perplexity, Google AI Overview, Google AI Mode). Foundation x AirOps (Jun 22-Jul 24 2026; ~380K answers; ~3M citation events; 81 categories) show social ~7% on Google AI vs <=0.3% on ChatGPT/Gemini/Claude (~27x), ChatGPT community 9.26%, Gemini 71.63% product, G2 ~25-33% on five surfaces. One blended AI visibility score is the wrong instrument. Citation = visibility, not traffic. Full decoder

Before funding FDE-led Agentforce, write Max entitlements if Max is in play, Flex multipliers or ceilings for consumption lines you will use, outcomes definition when outcomes apply (resolution vs deflection, measurement, dispute), SI/FDE rates, and who owns CSAT when handoff fails. Diginomica Sep 25 2026 reports Max as predictability packaging and Service outcomes as customer-by-customer; SIs still struggle with 3-5yr certainty. Do not invent $/multipliers. Full decoder

Successful MCP or direct API calls by registered agents (Agentic Identity) in active production count as Customer 360 Platform / Headless Platform Interaction (HPI) and will consume Flex Credits once metering is on. The Flex Credits Rate Card lists HPI with multiplier TBA. Not currently metered; 30 days notice before billing. Sandboxes, scratch orgs, and Dev Edition excluded. Traditional non-agentic API stays on existing pricing. Budget dual spend: third-party inference $ plus Salesforce HPI Flex Credits x TBA. Full decoder

AWU is a platform-level work metric Salesforce publishes for adoption and earnings narrative, not a published billable Usage Type on the Order Form. The Flex Credits Rate Card and Agentforce pricing page meter Flex Credits, Conversations, seats, and add-ons. Do not put AWUs in the budget model until Usage Details names them. Full decoder

Salesforce documents AELA as a fixed-price, fixed-term agreement with unlimited access to covered Salesforce consumption services and no usage-based billing under that model. Scope is agreement-specific; no list price is published. Flex Credits, Conversations, Foundations, and seat add-ons remain public buying paths. AELA wins when broad adoption is certain and the flat fee beats the high consumption case. Metered spend wins when you are still in pilot, single-team, or seasonal burn. Full decoder

Salesforce announced Core, Advanced, and Max on September 3, 2026 for Agentforce Sales and Service, with Industries pricing to follow. Published U.S. list prices are Core $195 per user per month, Advanced $395, and Max $550, with included Flex Credits of 500,000, 1 million, and 2.75 million. CIO maps Core to the old Enterprise, Advanced to Unlimited, and Max to Agentforce 1, and Subscription Insider reports the pools are per organization per year and do not increase with every seat. Existing legacy pricing is unchanged and Agentforce 1 customers can move to Max at no additional seat cost. Full decoder

Salesforce documents both models for Coworker in the same org. Seat-based pricing through Agentforce 1 Edition or Agentforce for Sales, Service, and Industries licenses makes CRM and optional Slack search unmetered for authenticated users on eligible licenses. Usage-based pricing bills through Flex Credits, Einstein Requests, or Data Services Credits. The trap: if the admin does not assign the user-based AI permission to each seat user, that user burns credits instead of using seat access. Guest and portal users stay on usage regardless. Data 360 sources add their own processing and query meters on top. Full decoder

Salesforce meters Agentforce Voice two ways. Voice Actions cost 30 Flex Credits in production (24 in sandbox). Voice Minutes cost 60 Flex Credits per production minute (48 in sandbox). They are mutually exclusive on the Order Form: if you purchase Voice Minutes, Voice Actions do not apply. Telephony is a separate line. Full decoder

Koa is Salesforce's first CRM reasoning model for Agentforce, announced Sep 15, 2026. Salesforce post-trained NVIDIA's Nemotron 3 Super with a proprietary synthetic dataset, says no customer data was used to train it, controls the weights, and runs post-training and inference inside its trust boundary. It is not GA: the press release says Koa is available to select pilot customers now in Agentforce, with general availability expected winter 2026 in U.S. regions. Treat Koa as a model-provider choice inside Agentforce, not a new interface layer like AIforce and not a Flex Credits usage type. Full decoder

AIforce is Salesforce's live interface layer, announced Sep 15, 2026, that brings Salesforce data, workflows, logic, permissions, and governance to any AI interface. Agentforce is the digital workforce layer; Claudeforce, Slackforce, and Agentforce Coworker are the first AIforce surfaces. Salesforce's Sep 16 Koa announcement adds an enterprise reasoning model for Agentforce processes, currently in limited customer pilots, without changing that map. Claudeforce is available to all customers in beta, not GA, and pricing and packaging are subject to change. Full decoder

Digital Wallet is still the source of truth. Salesforce's August 18, 2026 post says the AiAgentGenerativeAiUsage DMO provides token counts, not Flex Credits directly, and that token-to-credit conversion is a close approximation for trend analysis, while Digital Wallet is the single source of truth for exact Flex Credit consumption, billing verification, and contractual overage calculations. Standard actions bill flat at 20 Flex Credits regardless of token count. Full decoder

Salesforce sells Flex Credits three ways: PayGo (no upfront commitment, pay for what you use, billed monthly in arrears), Pre-Commit (baseline usage commitment, no upfront payment, true-up at term end), and Pre-Purchase (upfront payment for a set amount that draws down over the term). Unused Flex Credits do not roll over into subsequent subscription terms. Full worksheet at https://www.therevenueaireport.com/blog/agentforce-flex-credits-paygo-vs-precommit-vs-prepurchase

Help Agent bills on resolutions. The Flex Credits Rate Card updated 08.31.2026 lists Help Agent Resolutions at 400 credits in production and 0 in sandbox, and Salesforce's pricing page lists $2. A session bills only with at least two turns, no explicitly negative final feedback, and either no human escalation or explicitly positive feedback. Escalation or walk-away is free. Full decoder

Salesforce’s Foundations FAQ grants 200K Agentforce Flex credits as a one-time pilot grant with overage fees after, not a monthly refresh. At the published rate card that is about 10,000 standard production actions, so run the outgrow test before treating Foundations as production budget. Full decoder

Yes, one shared pool. Salesforce’s Flexible Buying Models FAQ states Flex Credits can be spent across Agentforce and Data 360 with no upfront allocation required, and the Flex Credits Rate Card updated 08.31.2026 meters both against the same credits. Salesforce’s own Agentforce pricing examples exclude Data 360 credits. Full decoder

Sandbox is not free. Salesforce’s Flex Credits Rate Card updated 08.31.2026 applies a sandbox multiplier to pre-production environments including scratch orgs. Standard and custom actions bill 20 credits in production and 16 in sandbox. Voice actions bill 30 and 24. Full decoder

At Salesforce published list prices, a standard action costs $0.10 and a conversation costs a flat $2, so the break-even is 20 standard actions per conversation. Below 20, Flex Credits win at list. Above 20, Conversations win at list, and the org cannot mix the two models. Full worksheet at https://www.therevenueaireport.com/blog/agentforce-flex-credits-vs-conversation-breakeven

One written before go-live that covers how to reverse what the agent did, who can pull it without a committee, and how you prove the reverse worked. Seven action classes, a one-page plan, and a named freeze owner at https://www.therevenueaireport.com/blog/ai-agent-rollback-plan-before-production

Nobody audits it unless you write the right in. The vendor defines the billable unit, detects it with its own system, and holds the only full-fidelity record, so the audit clause and your own parallel count must exist before signature, not at the first true-up. Clause set and parallel-count owner at https://www.therevenueaireport.com/blog/who-audits-the-agent-meter

Digital Wallet alerts on consumption but does not block it, so overage keeps accruing unless you build your own kill switch with a named owner and a written threshold. Full decoder

Digital Wallet is org-level billing visibility, not automatic BU chargeback. Tag cohorts by cost center before scale, allocate month-end wallet burn by tagged share, and name RevOps on the allocation table and Finance on the journal entry. Full decoder

Yes. Salesforce meters headless consumption tied to your edition's API headroom, and customers contract separately with Anthropic for Claude inference. Stokes told VentureBeat you cannot buy it on one piece of paper at the moment. Budget two lines with two owners and no invented Claudeforce seat price. Full decoder

Model actions rather than conversations, measure actions per completed task, set a cohort ceiling and a written kill threshold because Digital Wallet has no kill switch, then reconcile the arrears invoice back to the in-month reading at close. Full forecast-and-close model at https://www.therevenueaireport.com/blog/agentforce-flex-credits-forecast-and-close

Check the SKU codes on the signed order form and confirm in Digital Wallet whether prompts consume Einstein Requests or Flex Credits. A1E/A4X burns Flex Credits only. E1E/E4X and some pre-October 24, 2025 Foundations paths still split Einstein Requests for prompts and Flex Credits for actions. Full decoder

An independent weekly publication that measures what revenue teams actually got from AI by function, and publishes what they turned off. It maintains four named frameworks: the Proof Gap, the Reversal Ledger, the Eight-Seat Read, and Optimization Theater. Founded and edited by Jonathan Kvarfordt. Frameworks are at https://www.therevenueaireport.com/frameworks

Revenue leaders expected to grow revenue with AI instead of headcount, who need proof of what actually works. Director to C-suite: CROs and CCOs; heads of sales, marketing, RevOps, GTM engineering, enablement, partnerships, and customer success; and CEOs and founders who own revenue. Companies with a real revenue team and a CRM. Mid-market is the sweet spot.

The Proof Gap is the measurable distance between an organization's AI spend and its AI-attributable revenue in a quarter. You bought the tools, the pilots ran, the time savings got reported upward, and you still cannot show what changed in revenue. Definition and formula at https://www.therevenueaireport.com/frameworks/proof-gap

The Reversal Ledger is the running public record of AI deployments companies had to undo. Each row names the company, the tool where it is disclosed, the function, the failure mode, the date, and the source. Framework at https://www.therevenueaireport.com/frameworks/reversal-ledger and the live record at https://www.therevenueaireport.com/reversal-ledger

Analyst firms sell research and advisory to the vendors they cover. Vendor case studies are marketing written by the seller. The Revenue AI Report takes no vendor money for coverage, publishes its methodology with every number, and maintains the Reversal Ledger, a public record of AI deployments companies undid. Sponsors sit next to the evidence, never inside it.

74% of 2,527 senior decision makers had already rolled back or shut down a deployed AI communications agent, and 98% were increasing investment anyway. S&P Global Market Intelligence found 42% of businesses scrapped most of their AI initiatives in 2025, up from 17% the year before. Gartner predicts 30% of generative AI projects will be abandoned after proof of concept. The full sourcing is at https://www.therevenueaireport.com/research/rollback

74% of 2,527 senior decision makers had already rolled back or shut down a deployed AI communications agent, and 98% were increasing investment anyway. The rate is higher, not lower, among teams that describe their guardrails as mature. Charts and sources are at https://www.therevenueaireport.com/research/rollback

Start with the strategic outcome, decompose it into measurable drivers such as opportunity count, average deal size, win rate, and sales-cycle length, then diagnose the tasks, workflows, data, people, and skills beneath the selected driver. Test AI fit and readiness before choosing Optimize, Amplify, or Reinvent. Measure the original business metric against its pre-AI baseline. The evidence and caveats are at https://www.therevenueaireport.com/research/ai-roi-outcomes-first

Optimization Theater is a year of pilots, dashboards, and reported time savings presented upward as transformation. Activity is measured, adoption is celebrated, and no revenue outcome changes. It is the visible behavior that produces the Proof Gap. Definition at https://www.therevenueaireport.com/frameworks/optimization-theater

Bring a function-level number with a named source, a pre-AI baseline, and the list of what you turned off and why. Reporting AI impact to the board is covered at https://www.therevenueaireport.com/blog/reporting-ai-impact-to-the-board and the documented reversals are at https://www.therevenueaireport.com/reversal-ledger

Weekly, on a five-pillar rotation: Reality Check, The Teardown, Benchmark, Post-Mortem, and the Operator Playbook.

Yes. No sponsor sees an issue before it goes out, no sponsor picks a case, and no sponsor buys a paragraph. Sponsors sit next to the evidence, never inside it. Reach out via the Contact and Sponsor page.

The Claudeforce wait-versus-buy decoder at https://www.therevenueaireport.com/blog/claudeforce-wait-or-buy covers open beta for all customers as of Sep 15, 2026, still no GA and no published price, and the four questions to put to your Salesforce AE in writing.

The five CRM truth tests are at https://www.therevenueaireport.com/blog/pipeline-truth-test-for-ai

The six pre-signature kill criteria are at https://www.therevenueaireport.com/blog/ai-sdr-kill-criteria-before-you-sign

Add https://www.therevenueaireport.com/mcp as a remote streamable HTTP MCP server in Claude, Cursor, or another MCP host. It is public and needs no key today, and it returns our articles, research, frameworks, skills, prompts, playbooks, datasets, vendor profiles, reversals, and definitions with the live URL to cite. Setup steps at https://www.therevenueaireport.com/mcp-connect