# Gong Credits Explained: What the Pricing Backlash Teaches Every AI Vendor > What Gong Credits changed, why one customer burned a third of an annual pool in a week, and what AI buyers and vendors should learn. - Source: https://www.zarifautomates.com/blog/gong-credits-pricing-case-study - Published: 2026-08-11 - Updated: 2026-08-12 - Pillar: Enterprise AI - Tags: Gong Credits, AI pricing, usage-based pricing, AI credits, enterprise software - Author: Zarif --- Gong's introduction of Gong Credits in 2026 is a nearly perfect case study in the tension behind AI pricing. The vendor's argument is economically coherent: selected AI workflows process large volumes of calls and emails or run continuously, so the cost grows with work rather than with the number of people who can log in. Raising every seat price would make light users subsidize heavy autonomous usage. The customer objection is equally coherent: organizations already pay for seats and a platform, adopted AI features because the product encouraged them to, and now face a new meter that can be difficult to forecast and administer. The lesson is not “Gong was right” or “Gong was wrong.” It is that a valid pricing unit can still produce a poor migration experience. A Gong Credit is a shared usage unit for selected AI-powered features that process calls, emails, or other data automatically or at scale. Gong allocates a pool based on seat type and count, then customers can purchase more credits. - Gong did not replace all seat pricing with credits; it added a consumption layer for selected at-scale and agentic features - Gong says core capabilities such as calls, conversation insights, deal intelligence, forecasting, coaching, predictions, and everyday AI remain included in seats - A customer publicly reported consuming one-third of an annual credit allocation in one week and described a difficult process for disabling trackers - Gong's co-founder acknowledged the administration problem while defending consumption pricing for expensive agentic work - The failure mode is broader than one vendor: unclear workload forecasts plus difficult controls turn usage alignment into customer anxiety - Buyers should inventory trackers and autonomous jobs, export usage, map credits to outcomes, and renegotiate guardrails before scaling ## What Gong Actually Changed Gong announced the new usage model to customers in late May 2026 and published its help article on June 9. The change introduced credits for capabilities that can run continuously or process data at scale. Gong's documentation lists these credit-consuming areas: - Question-based AI Trackers - APIs for AI Ask Anything and AI Briefer - Gong's MCP server Its documentation also says many capabilities remain included with seats. Gong's public explanation names calls, conversation insights, deal intelligence, forecasting, coaching, revenue and deal predictions, and AI used manually in the product. This makes the design a **hybrid model**: **Annual platform and seat commitment + included credit pool + optional additional credits** Calling it a complete move from seats to tokens would be inaccurate. The seat is still the access and platform layer. Credits meter selected machine work on top. ## How Gong Credits Are Consumed Gong publishes a workload-oriented rate rather than exposing raw model tokens: - Ten emails consume one credit, so one email consumes 0.1 credit - A call longer than ten minutes consumes one credit - A call up to ten minutes consumes 0.5 credit - Processing new data consumes credits; viewing previously processed results does not Gong also says data that has already been processed may be reused across supported features without consuming credits again. That is an important design choice: the economic unit is processing, not every person viewing an insight. The model is more legible than low-level token billing. A RevOps leader can estimate calls and emails more easily than prompt and completion tokens. But the estimate still depends on tracker scope, historical backfill, filters, and how many autonomous workflows are active. ## Why Gong Says Credits Are Necessary Gong's stated rationale has three parts. ### Variable infrastructure cost At-scale analysis and autonomous agents create costs that grow with processed data. The vendor has to absorb, limit, or pass through those costs. ### Fairness between light and heavy users Instead of increasing seat prices for every customer, Gong applies credits to selected heavier capabilities. In principle, customers running more machine work pay more. ### Existing context Gong argues that processing inside its platform preserves account, deal, and interaction context. Exporting transcripts to a separate model can require customers to reconstruct that context and pay a model provider again. Those are reasonable product and economic arguments. They do not answer the rollout question: **Did customers have enough time, telemetry, and control to change behavior before the new meter became consequential?** ## The Customer Reaction That Made the Risk Concrete In July 2026, Gong customer Robby Halford wrote publicly that his company used one-third of its annual credit allocation in a week. He described trying to deactivate an expensive smart tracker and finding it attached across searches, streams, deal boards, and even a former user's configuration. This is one customer's report, not a representative survey of Gong's customer base. It matters because it demonstrates a specific and plausible failure mode: 1. Customers adopt an AI feature under one economic expectation 2. The feature becomes attached to many workflows 3. A new meter makes that footprint expensive 4. The administrative path to reduce usage is harder than the path that created it 5. The customer experiences the meter as a penalty for adoption Gong co-founder Eilon Reshef replied publicly. He apologized for the administration experience, said the company was working on it, and explained that Gong always knew AI Tracker usage was costly and would eventually require limits or consumption pricing. He also said the included allocation was intended to cover basic use and that additional credits were optimized to be cheaper than customers running equivalent LLM processing themselves. The response is notable because it separates two issues: - **Economic design:** whether heavy agentic work should carry a usage price - **Product operations:** whether customers can understand and control that work easily The first can be defensible while the second still needs fixing. ## The Four Trust Gaps Exposed by the Rollout ### 1. The forecast gap “Credits included” is not a forecast. Customers need to see their actual historical workload translated into the new meter. A useful pre-change report would show: - Credits the prior 30, 60, and 90 days would have consumed - Usage by tracker, feature, business unit, and owner - Expected annual consumption at the current run rate - The drivers behind high and low cases - Which configurations can be changed without losing current results ### 2. The control gap Usage pricing requires usage controls. If enabling a tracker is easy but safely disabling it is a scavenger hunt, the product has an asymmetric control surface. Administrators need feature-level pause, scope editing, bulk ownership transfer, stale-owner cleanup, budget thresholds, and impact previews. ### 3. The value gap Credit burn shows work performed, not value delivered. A tracker can process thousands of calls while few people use its output. For every high-consuming tracker, pair the meter with: - Active viewers or downstream consumers - Decisions or workflows influenced - Time saved versus the prior process - Sales or customer outcome being pursued - Quality, false-positive, and action rates ### 4. The timing gap Pricing migrations consume customer attention. A short notice period near a quarter end makes operational cleanup harder for sales teams and RevOps administrators. Even a generous allowance can feel hostile if the customer cannot analyze and adjust before the effective date. ## What Gong Administrators Should Do Now Gong's own guidance recommends narrowing tracker data, reducing historical backfill, and applying filters. Turn that into a structured audit. ### Step 1: Export the usage data Download credit usage, balance, and purchase reports. Attribute each consuming feature to an owner, team, purpose, and renewal-critical outcome. ### Step 2: Rank by burn and value Create four groups: | | High value | Low or unknown value | |---|---|---| | **High burn** | Optimize and govern | Pause or redesign first | | **Low burn** | Keep and monitor | Consolidate during cleanup | Do not optimize every tracker equally. Start with high-burn, low-evidence work. ### Step 3: Reduce the data surface Filter by relevant team, call type, deal stage, account segment, language, date, and email category. Limit backfills to the period that will actually affect a decision. ### Step 4: Reuse processed outputs Gong says viewing and reusing already processed tracker data does not consume new credits. Design briefs, dashboards, and downstream workflows to use existing processed signals where practical instead of creating near-duplicate analysis. ### Step 5: Forecast the exhaustion date **Forecast days remaining = current credit balance ÷ average daily credits consumed** Calculate trailing seven-, 30-, and 60-day rates. A single average hides growth and cleanup effects. ### Step 6: Set an outcome review Every month, review the top-consuming trackers with RevOps, Sales leadership, Finance, and the business owner. Keep, optimize, or stop each one based on outcome evidence. ## Where Deterministic Automation Fits Not every revenue workflow requires another generative analysis over every call or email. Known logic—routing by CRM stage, filtering calls by duration, deduplicating records, checking whether required fields exist, applying account tiers, scheduling jobs, updating systems, and sending approved notifications—can run deterministically. Use an LLM where language ambiguity is the actual problem. n8n is useful as a control layer because a workflow can filter and enrich Gong data, route only qualified items to OpenAI, Anthropic, or another model, apply budget rules, and write results back. It can also keep non-AI steps deterministic and logged. The point is not to rebuild Gong indiscriminately. It is to avoid paying an AI meter for work that a rule, API call, or existing processed signal can do reliably. Before exporting any revenue data, review Gong's contractual terms, security requirements, data governance, and whether the lost native context makes the alternative genuinely cheaper. Gong is correct that reconstructing context outside the platform has a cost. The [deterministic workflows vs LLM calls framework](/blog/deterministic-workflows-vs-llm-calls) helps identify that boundary. ## What Every AI Buyer Should Learn The Gong story generalizes beyond revenue software. ### Ask for a workload translation, not a credit allowance Require the vendor to map last quarter's real activity to the new rate card. Included credits mean little without a predicted burn distribution. ### Negotiate control before discount Hard caps, alerts, pause behavior, no automatic top-up, feature-level attribution, and usage export can matter more than a lower unit price. ### Separate access value from machine-work value Understand what the seat covers, what the credit covers, and whether one depends on the other. Model the total annual commitment, not the attractive marginal rate. ### Test administration during the pilot Ask an administrator to find the highest-consuming configuration, change its scope, transfer ownership, pause it, and verify the effect. A control that exists only in documentation is not operational control. ### Price the alternative honestly An external workflow needs model calls, orchestration, security, monitoring, maintenance, and context. Compare total cost per accepted outcome, not Gong credits versus raw API tokens. Use the [AI contract negotiation checklist](/blog/negotiate-token-credit-based-ai-contracts) before accepting a credit-based renewal. ## What Every AI Vendor Should Learn ### Do not meter before customers can manage Usage dashboards, bulk controls, alerts, ownership, and clean deactivation are part of the billable product. ### Shadow bill before launch Show customers what the new model would have charged for several prior periods. Let them fix runaway configurations while the invoice is still hypothetical. ### Preserve a meaningful included experience If the marketed core suddenly becomes unusable without overages, customers will interpret the change as a hidden price increase. Explain exactly which value remains inside the subscription. ### Translate work into outcomes Show the relationship between credits, processed data, used insights, and business results. A spend dashboard without value evidence invites cancellation. ### Roll out by cohort and learn Start with customers whose usage is well understood and administration is clean. Measure forecast error, support volume, pause behavior, adoption, and sentiment before expanding. The complete [pricing migration playbook](/blog/move-customers-seat-to-usage-based-pricing) turns those principles into a rollout sequence. The Gong case does not prove that consumption pricing is unfair. It shows that usage alignment, forecastability, and administrative control have to arrive together. ## Frequently Asked Questions **Did Gong replace seat-based pricing with credits?** No. Gong still prices licenses per user and applies a platform fee. Gong Credits add a usage layer for selected AI capabilities that process data automatically or at scale. Many core and manually invoked AI capabilities remain included with seats according to Gong. **What consumes Gong Credits?** Gong lists question-based AI Trackers, APIs for AI Ask Anything and AI Briefer, and its MCP server. Credit consumption depends on the type and amount of newly processed data. Gong's live documentation should be checked because covered features and weights can change. **How many Gong Credits does a call use?** As documented in July 2026, calls longer than ten minutes consume one credit and calls up to ten minutes consume 0.5 credit. Ten emails consume one credit. Verify the current rate card and your contract before forecasting. **Why were customers upset about Gong Credits?** Public criticism focused on rapid credit consumption, limited time to adjust, and difficulty finding and disabling configurations that consumed credits. The most detailed example is one customer's reported experience, so it should not be generalized to every Gong customer. **Can n8n replace Gong AI features?** n8n can orchestrate alternative workflows using Gong data, rules, APIs, and external models, but it does not automatically reproduce Gong's revenue context, product experience, governance, or models. Use it selectively when deterministic control, routing, portability, or custom integration creates a better total cost and operating model. --- ## Sources and Further Reading - [Gong credits: New AI usage model](https://help.gong.io/docs/gong-credits-new-ai-usage-model) - [About Gong credits](https://help.gong.io/docs/about-gong-credits) - [Managing credit usage — Gong](https://help.gong.io/docs/when-credits-run-out) - [Optimize existing AI Trackers for Gong credits](https://help.gong.io/docs/guide-optimize-existing-ai-trackers-for-gong-credits) - [Scaling revenue AI beyond the basics with Gong Credits](https://www.gong.io/blog/scaling-revenue-ai-gong-credits) - [Gong Credits customer experience and Gong response — LinkedIn](https://www.linkedin.com/posts/robbyhalford_im-a-big-fan-of-gong-but-this-credits-activity-7478474550117171200-SM7O) - [Gong pricing](https://www.gong.io/pricing) - [Build Custom AI Agents With Logic and Control — n8n](https://n8n.io/ai-agents/) - [From Seat-Based to Token-Based Pricing](/blog/seat-based-vs-token-based-pricing)