The State of Play

A living index of AI adoption across industries — where established practice meets the bleeding edge
UPDATED DAILY
← All domains

💼 Sales & Revenue

AI across the revenue cycle from lead identification to closed deal. The most consistently mature domain: three-quarters of practices are good practice, including lead scoring, pipeline forecasting, and conversation intelligence. CRM copilots are mainstream. The few leading-edge practices involve autonomous prospecting and deal-coaching agents. Momentum is moderate — most gains are incremental rather than transformative.

16 practices: 13 good practice, 2 leading edge, 1 bleeding edge

The Headline

Nearly every revenue team now uses AI, but few can show it wins more business. Two new surveys find only about one in five can connect AI to revenue.

16
practices tracked
3
at the frontier
0
moving this fortnight
3,047
evidence items

The Picture

AI is now an ordinary part of selling. Lead scoring, call recording, forecasting and data enrichment come as standard features in the customer relationship management (CRM) platforms and sales suites most revenue teams already pay for, so nearly everyone uses something. Few are seeing a payoff. In BCG's survey of 398 revenue leaders, 90% reported meaningful AI use but only 8% reported higher quota attainment. The small group pulling ahead has cleaned its customer data and kept prices and calculations rule-based, with AI drafting and people approving. If you use AI but can't point to a change in win rate or quota, you are in the pack. The next wave of agents (software that acts on its own without being prompted) will multiply whatever your data already gets wrong.

This Fortnight

  • ChurnZero surveyed 580 customer success and revenue leaders and found only 21% of customer success leaders connect their measurement of AI to revenue outcomes. Net revenue retention, the share of existing customer revenue kept and grown, has been flat since 2024 despite the spending. If your customer team cannot tie its AI tools to renewals and expansion, expect a harder budget conversation.

  • A TD SYNNEX survey of more than 1,500 channel partners in 45 countries found 66% increasing AI investment but only 21% earning meaningful revenue from it. TD SYNNEX commissioned the survey itself, and partners say they are leaning more on distributor enablement programs. If you sell through partners, assume most cannot yet sell your AI-driven offers without substantial help from you.

  • Only 16% of 220 B2B sellers would let an AI agent take a sales action without reviewing it first, according to a Firmable survey. Half said they would be blamed if an agent acted on bad data. Plan agent rollouts around a person approving each action, not around autonomy, because your sellers will insist on it either way.

  • Sales data is moving into general-purpose assistants: Gong now returns deal risks inside Claude and ChatGPT, and reps at Mosaic draft quotes by asking Claude. Both rely on MCP (a connector standard for AI tools). At Mosaic, the first round of business context covered only about half of quotes, and a rep still sends every one. Ask each vendor what its connector can read and change, and who signs off the output.

  • Rox, an AI-native sales platform pitched as a challenger to Salesforce, closed a $100 million Series B on November 11 at a $1.3 billion valuation. An independent review advised allowing 60 to 90 days of calibration before its recommendations are reliable, and cited no customer deployment. Investors are still funding CRM replacements, but treat the pitch as unproven until it works on your own deals.

Coming Up

  • The RealPage antitrust judgment now reads as a design specification for pricing software that uses competitors' data. It bars data from unaffiliated properties within 180 days and rules out automatically accepted price recommendations unless users have set limits. Legislators in 24 US states introduced more than 40 surveillance-pricing bills in 2026, and Maryland and Connecticut have enacted bans. Before your next pricing-tool renewal, have counsel check what data feeds the tool and whether its recommendations can be accepted automatically.

  • Vendors are moving from agents that read your CRM to agents that write to it, and the controls are behind. Salesloft and Clari's connector is read-only today, with write access planned. An independent review found that ZoomInfo's human approval step for its agent teams is still on the roadmap. LeanData found that 30% of 157 customers had already discovered agent actions on their records with no audit trail. Make approval steps and audit logs a contract condition before you turn on write access.

  • Salesforce has stopped selling its legacy configure-price-quote (CPQ) tool and is steering customers to Agentforce Revenue Management. Salesforce guidance puts a typical migration at three to six months, and no shutoff date has been published. Mergers have also retired the Clari brand and combined Highspot with Seismic. List which of your sales tools have changed owners or products, and plan migrations around renewal dates rather than vendor deadlines.

What's Hard About This

  • The CRM records your AI reads are often wrong, and a better model will not fix that. One audit of a Salesforce instance found that 70% of more than 113,000 account records could not be tied to an identifiable company. Missing data also pushes forecasts too high, because the gaps cluster on the busiest reps and the hardest deals.

  • AI has made sales work cheaper to produce without making it win more often. Loopio's 2026 benchmark shows teams submitting 166 RFPs (requests for proposal) a year, up from 153, while their win rate fell six points to 39%. That suggests cheaper drafting leads teams to bid more and raises the cost of each win.

  • Nearly all the evidence that these tools pay off comes from the vendors selling them. A vendor's audit of 100 AI sales products found only 8 showing a won or lost result changing a future decision. No independent accuracy test of deal-risk scoring has been published. Before signing, insist on a test against your own closed deals.

2012View this domain's timeline →Today

Practices in this Domain (16)

PRACTICETIERTREND
Account intelligence — contact mapping & briefingLEADING EDGE— Steady
Account-based marketing signal identificationGOOD PRACTICE— Steady
Conversation intelligence — transcription & analysisGOOD PRACTICE— Steady
CRM data management & enrichmentGOOD PRACTICE— Steady
Deal intelligence — risk assessment & win/loss analysisGOOD PRACTICE— Steady
Lead scoring & ideal customer profilingGOOD PRACTICE— Steady
Partner & channel sales supportBLEEDING EDGE— Steady
Price optimisation & dynamic pricingGOOD PRACTICE— Steady
Prospecting & outreach personalisationGOOD PRACTICE— Steady
Quote & contract generationGOOD PRACTICE— Steady
Real-time conversation guidanceGOOD PRACTICE— Steady
Revenue intelligence & expansion signalsGOOD PRACTICE— Steady
Sales content — proposals, battle cards & objection handlingGOOD PRACTICE— Steady
Sales enablement content generationLEADING EDGE— Steady
Sales forecasting & pipeline analysisGOOD PRACTICE— Steady
Territory & account planningGOOD PRACTICE— Steady
Read the full technical briefing (1,845 words) →

Where AI Stands in Sales & Revenue

Sales is the domain where AI became ordinary first, and in the form that is easiest to sell. Lead scoring, conversation intelligence, pipeline forecasting and CRM enrichment now ship as standard features of the platforms revenue teams already pay for. Nobody has to make a separate case to buy them. The commercial scale is real. Gong passed $500M ARR in May, Salesforce's Agentforce has crossed $1.5B ARR, and HubSpot says nearly 17,000 customers have activated its Prospecting Agent. The vendor map has collapsed into a few suites: Clari and Salesloft have merged under the Salesloft brand, Seismic has merged with Highspot, and Gartner now publishes a Magic Quadrant for revenue action orchestration. For most of what this domain covers, nobody is asking any more whether the software works.

What remains is whether it pays, and sales is unusually good at answering that, because quota, win rate and forecast error all get counted. On that count the domain looks stuck. BCG surveyed 398 revenue leaders: 90% reported meaningful AI use, but only 8% reported higher quota attainment and 13% better win rates. Salesloft found that every revenue team it surveyed uses AI, yet only 20.6% call it production-ready. Gartner's survey of 318 sales operations leaders puts median forecast accuracy at 70 to 79%, with only 7% reaching 90%, and 69% say forecasting is harder than it was three years ago. Alexander Group recorded AI investment up 90% between 2024 and 2025 while revenue per rep fell 5%.

The structural reason is the same in nearly every practice. The models read CRM records that are incomplete, stale or simply wrong, and they sit on top of selling processes nobody redesigned. One audit of a Salesforce instance found that 70% of more than 113,000 account records could not be tied to an identifiable company. Enrichment vendors' own samples fill industry fields almost completely but find a work email for only a quarter to a third of records.

Momentum is building at two edges. The first is the interface. CRM, CPQ, battlecard, partner-programme and revenue-intelligence data now reach Claude and ChatGPT through MCP connectors from Salesforce, Gong, Conga, Klue, Impartner, ZoomInfo and others, so the general-purpose assistant is becoming where a seller reads an account. The second is narrow, governed automation that keeps the arithmetic deterministic. Quote agents draft but never invent a price, enrichment fills only empty fields and records the source of each write, and territory agents run their calculations in the platform rather than in the model.

Momentum stalls wherever autonomy is the selling point. Teams that cut SDR headcount for autonomous agents found the agents produced sends rather than meetings. In one documented team, reps ignored live in-call prompts for about two weeks. Fewer than 12% of channel partners use their vendor's portal more than once a quarter. Price optimisation is the outlier. It is proven in airlines and hotels, but its spread into consumer retail is now decided by antitrust courts and by more than 100 US surveillance-pricing bills, not by buyers.

What's New, 2026-10-31 to 2026-11-14

The fortnight's evidence mostly measured the gap between using AI and getting a return from it, and put numbers on how reluctant sellers are to let agents act. ChurnZero surveyed 580 customer success and revenue leaders: only 21% of CS leaders connect their measurement of AI to revenue outcomes, and net revenue retention has been flat since 2024. TD SYNNEX commissioned its own survey of more than 1,500 channel partners in 45 countries. It found 66% raising AI investment but only 21% earning meaningful revenue from it. In a Firmable survey of 220 B2B sellers, cited by Scalestack, only 16% would let an agent take a sales action without reviewing it first. An Apollo survey of 328 GTM professionals put full trust in unreviewed AI output at 4.4%. Belkins found refining the ideal customer profile to be the leading strategic use of AI, yet 77% of its 120 respondents could not confirm a positive return. Forecasting drew sharper critiques. Weflow argued that missing activity data biases forecasts upwards, because the gaps cluster on the busiest reps and the hardest deals. It cites Mediafly, whose deal predictions need about 100 opportunities per quarter with full activity capture against about 300 without. Spotlight.ai argued that MEDDPICC fields filled automatically from call transcripts are never re-checked as deals change. Rox closed a $100M Series B on 11 November at a $1.3B valuation. Its reviewer advised a 60 to 90 day calibration window and cited no customer deployment.

Product news continued the move into assistants and agents. ARPEDIO announced eight Agentforce agents for org charts, relationship scoring and missing contacts, each of which waits for a person to accept or reject its suggestions. Introw and Impartner opened partner data to Claude through MCP, and Gong's MCP connectivity now returns deal risks inside Claude and ChatGPT. At Mosaic, reps draft quotes, renewals and amendments by asking Claude, though the first context pass covered only about half of quotes. A consultancy tested Salesforce in Claude on its own production org and found the limit: account research held context well, but skills that read tasks and meetings had almost nothing to work with, because day-to-day contact was never logged. The most striking self-reported result came from LangChain. Its internal GTM agent produced 1,540 customer-facing artifacts in 12 weeks, and deals that received one advanced 47% of the time against 16% for matched deals; LangChain did not disclose how it matched them.

Conversation intelligence evidence settled. A French survey of 13 companies found Modjo in production at six but underused, because managers lack time to coach from it. Pricing evidence this cycle concerned design limits rather than new prohibitions, with legal analysis setting out the RealPage decree's restrictions on competitor data and auto-accepted recommendations. Live guidance and prospecting drew mostly cautionary material. Claap's analysis of its own sales calls over the year to September found 25% of buyer teams asking about live help, against 90% describing post-call coaching needs. Fuzzy AI modelled cold AI SDR sequences exhausting a 15,000-lead list in about four and a half quarters.

Key Tensions

  • Cheaper output, unchanged results. AI has lowered the cost of producing sales work without raising what that work wins. In Content Marketing Institute and MarketingProfs research cited by Fullcast, 87% of marketers using AI for content report productivity gains but only 39% say content performs better. Loopio's 2026 benchmark shows teams submitting 166 RFPs a year, up from 153, while their win rate fell six points to 39%, which suggests cheaper drafting is raising the cost per win.

  • Agents acting on records nobody trusts. Vendors are selling agents that write to the CRM, but only 18% of the 39 respondents to one RevOps survey rate their records clean enough for an agent to act on unreviewed. Claap reports that 57% of buyers on its own sales calls described their CRM as incomplete, wrong or out of date. At HubSpot UNBOUND, Sellestial described a single enrichment error, a Bloomberg profile page taken for a company's website, spreading into four more fields.

  • The assistant as front end, with governance trailing behind. MCP connectors now put deal, quote and partner data in front of Claude and ChatGPT faster than approval and audit controls are being built. An independent review found that the human approval step for ZoomInfo's Agent Teams is still on the roadmap. LeanData found that 30% of 157 customers had discovered agent actions on records with no audit trail. servicePath cites Gartner: 71% of organisations expect to deploy agentic AI within a year, while 21% have governance capabilities for it in place.

  • Proof that is almost entirely self-reported. The largest outcome claims this cycle came without method. Demandbase reported a 30:1 return at Effectual with no period or attribution model, SolarWinds reported $42M won with no baseline, and Balto reported more than 600 million guided calls with no sample or methodology. Go7's audit of 100 AI sales products found only 8 showing a won or lost result changing a future decision, and no independent accuracy test of deal-risk scoring has been published.

  • Pricing's legal perimeter. Algorithmic pricing is being constrained by courts and legislatures rather than by its technical limits. The RealPage final judgment bars data from unaffiliated properties within 180 days and rules out auto-accepted recommendations without user-set parameters. Revived claims allege that Atlantic City casino-hotels charged Rainmaker's suggested rates 90% of the time, and Maryland's HB 895 has barred large food retailers from using personal data to charge more since 1 October. Measured shelf-price behaviour cuts the other way: one study of more than 180 million grocery price records found essentially no change in pricing after digital labels were installed.

Top 10 Evidence Items

  1. The dashboard nobody opens: How AI assistants are moving sales analytics to the moment of decision (news-coverage) — Shows the two-edge momentum claim colliding with itself: the assistant becomes the interface for deal data at the exact moment surveyed sellers say they won't trust its output unreviewed. https://live-kvia.pantheonsite.io/stacker-business-economy/2026/11/06/the-dashboard-nobody-opens-how-ai-assistants-are-moving-sales-analytics-to-the-moment-of-decision/
  2. Salesforce in Claude: What We Learned on Our Own Org (opinion) — A first-hand MCP test exposes the structural flaw the briefing names: assistants read CRM context well only where reps actually logged activity, which most don't. https://www.theskyplanner.com/salesforce-in-claude/
  3. Who Is Accountable When an AI Agent Acts on Your CRM Data? (opinion) — Puts a number on the 'agents acting on records nobody trusts' tension from the seller's own mouth, not a vendor's. https://scalestack.ai/blog/who-is-accountable-when-an-ai-agent-acts-on-your-crm-data
  4. How Missing Activity Data Distorts the Sales Forecast (opinion) — Explains the mechanism behind flat forecast accuracy: missing activity data skews predictions upward precisely on the busiest reps and hardest deals the briefing flags. https://www.weflow.ai/blog/how-missing-activity-data-distorts-the-sales-forecast
  5. Rox AI Sales Tool Review: Can It Challenge Salesforce? (opinion) — A $100M raise with a flagged calibration window and no customer deployment is the clearest instance of the 'proof that is almost entirely self-reported' tension. https://www.crmdaily.co/article/rox-ai-sales-tool-review-can-it-challenge-salesforce
  6. How our GTM Agent builds consistent, high-quality sales artifacts (case-study) — The most-cited positive result this cycle, but its own method for matching comparison deals is undisclosed, so it illustrates how even the best outcome claim here is unaudited. https://www.langchain.com/blog/gtm-agent-high-quality-sales-artifacts
  7. Modjo, Gong or Claap? How French companies record and use their calls (industry-report) — Independent (non-vendor) evidence that conversation intelligence has stalled on adoption, not capture: the bottleneck is manager coaching time, not the tool. https://www.softatlas.io/article/modjo-gong-claap-enregistrement-appels-entreprises-francaises
  8. AI Investment Outpacing Customer Success Measurement: ChurnZero (adoption-metric) — Shows the ROI measurement gap extending beyond sales into customer success, with retention flat despite near-universal AI use. https://www.demandgenreport.com/industry-news/news-brief/ai-investment-outpacing-customer-success-measurement-churnzero/55104/
  9. TD SYNNEX Finds AI Investment Outpacing Partner Revenue (news-coverage) — Channel partners mirror the sales-floor pattern exactly: investment rising sharply while realised revenue from that investment stays rare. https://www.channelinsider.com/ai/td-synnex-ai-investment-partner-revenue-gap/
  10. Algorithmic pricing, collusion and the law in 2026 (industry-report) — The one sub-domain where AI's limits are being set by courts and legislatures rather than by whether the technology works, which the briefing singles out as the exception. https://rillor.com/insights/algorithmic-pricing-collusion-law