AI in B2B Sales and GTM: 60 Terms Explained

A seller finishes a discovery call. The transcript is available, the CRM has been updated and an AI assistant can produce a summary in seconds. The harder question remains: what should the seller do next?

To answer that well, the assistant needs more than the conversation. It needs to know how your team qualifies an opportunity, what evidence matters at this stage, which assets are approved and what remains uncertain. Someone must also decide whether its recommendation is appropriate.

That is where terms such as context layer, agent, harness and human-in-the-loop become useful. They describe different parts of the work. Understanding the differences helps you ask better questions of a supplier, specify a useful workflow and avoid buying capabilities you don’t need.

This guide explains 60 terms through B2B sales examples. Some are established technical concepts. Others are emerging labels whose meaning varies by provider. It’s a practical reference, not a ranking of products or a claim that every sales team needs every component.

Why the sales workflow matters

McKinsey's July 2026 report, The future of B2B sales, argues that commercial value depends on connecting AI with workflows, data ownership, human judgement and adoption. Its discussion of sales enablement also retains a central role for manager-led coaching.

For a smaller B2B team, our practical interpretation is to choose a defined workflow and make it work from start to finish. Discovery to an agreed evaluation is one example. The task includes understanding the buyer's problem, identifying missing evidence, preparing the right demonstration and agreeing what the evaluation must establish. Producing a summary is only one step.

You can test that workflow without first rebuilding the whole revenue organisation. Agree the standard, supply the evidence, review the output and observe what the seller does with it.

How the main concepts fit together

A GTM AI harness can bring several of these components together. A sales method layer supplies a narrower part: the company's agreed guidance for assessing deals and choosing actions. Neither label tells you everything about the implementation. Ask what the system actually contains and does.

AI sales and GTM terms, grouped by purpose

Use the groups below to find a term. Each explanation links to a fuller glossary entry with a practical check and a limitation to understand.

• Foundations
• Agents, workflows and execution
• Knowledge, method and context
• Retrieval and connections
• Instructions, controls and evaluation
• Sales applications
• Teams and operating models
• Discovery in AI answers

Foundations

AI in sales

AI in sales is the use of machine learning and generative AI to support sales tasks, such as researching accounts, preparing calls, reviewing deal evidence and forecasting revenue.

Sales example: A seller supplies discovery notes and an approved playbook. An assistant identifies unanswered questions and prepares a call plan for review. Full definition of AI in sales.

Generative AI

Generative AI produces new content, such as text, images or audio, from relationships learned during training and the inputs supplied at use time.

Sales example: An assistant drafts a discovery follow-up from the call transcript and the seller’s instructions. Full definition of Generative AI.

Large language model (LLM)

A large language model is a machine learning model trained on large amounts of data to process and generate language. Applications can combine it with retrieval, tools and instructions.

Sales example: The model interprets a seller’s question and drafts an answer; the surrounding application supplies the playbook and any permitted account information. Full definition of Large language model (LLM).

Predictive AI

Predictive AI uses relationships in data to estimate an outcome, classification or future value, such as a lead’s likelihood of converting or expected revenue.

Sales example: A model uses historical opportunities and current deal characteristics to estimate the probability of a win within a stated period. Full definition of Predictive AI.

Agentic AI

Agentic AI describes systems that use AI to choose and carry out steps towards a goal, often calling tools and adjusting their actions as new information arrives.

Sales example: A research assistant finds an account, notices conflicting information and chooses another source before producing its brief. Full definition of Agentic AI.

Agents, workflows and execution

AI sales agent

An AI sales agent is a system that uses an AI model, instructions and tools to complete a defined sales task through one or more steps, with a defined level of autonomy.

Sales example: An agent gathers permitted account information, checks it against an ICP and prepares an account brief for a seller. Full definition of AI sales agent.

AI copilot

An AI copilot is an assistant that supports a person while they work, for example by drafting content, finding information or suggesting next steps.

Sales example: During deal preparation, a copilot shows missing decision criteria and suggests questions for the seller’s next conversation. Full definition of AI copilot.

AI SDR

An AI SDR is a product or workflow intended to perform parts of a sales development representative’s work, such as prospect research, outreach drafting, follow-up or initial qualification.

Sales example: A system researches approved target accounts and drafts outreach for a human to review before sending. Full definition of AI SDR.

Agentic workflow

An agentic workflow is a process in which AI chooses some of the steps or actions needed to complete a task. The amount of discretion depends on the design.

Sales example: After reviewing discovery notes, an assistant decides whether to retrieve a demo guide, a business-case template or a qualification checklist. Full definition of Agentic workflow.

Agent orchestration

Agent orchestration coordinates the tools, agents, data and human hand-offs required to complete an AI-supported process.

Sales example: A workflow retrieves account evidence, prepares recommendations, requests seller review and records an approved result. Full definition of Agent orchestration.

Multi-agent system

A multi-agent system uses two or more AI agents with distinct or shared responsibilities to complete work together.

Sales example: One agent researches an account, another checks claims against sources, and a coordinating process prepares the final brief. Full definition of Multi-agent system.

GTM AI harness

A GTM AI harness is the execution environment around an AI model that manages context, tools, task state and controls for go-to-market work.

Sales example: For deal preparation, a harness supplies relevant materials, allows permitted retrieval, manages the task and records outputs for review. Full definition of GTM AI harness.

GTM engineering

GTM engineering is the work of connecting data, tools and workflows to support go-to-market execution, often using automation, integrations and AI.

Sales example: A team connects approved account signals to research, owner assignment and a CRM task, then measures whether the workflow creates useful conversations. Full definition of GTM engineering.

Sales automation

Sales automation uses software to carry out defined sales tasks or process steps when specified conditions are met. It may use fixed rules, AI or a combination.

Sales example: A completed meeting creates a follow-up task for the account owner. A separate AI step may draft notes for review. Full definition of Sales automation.

Knowledge, method and context

Context layer

A context layer makes relevant business information, definitions and operating guidance available to an AI application. It can draw on documents, structured records and other approved sources.

Sales example: A sales assistant receives the company’s proposition, stage standards, approved assets and the evidence available for the opportunity. Full definition of Context layer.

Context engineering

Context engineering is the practice of selecting, structuring and maintaining the information, instructions and tool outputs an AI system receives for a task.

Sales example: A demo-preparation task gets the buyer’s workflow, unresolved discovery questions and the relevant demo standard, rather than the full sales document archive. Full definition of Context engineering.

Knowledge graph

A knowledge graph represents entities and the relationships between them in a structured form that can be queried and used by software.

Sales example: A graph connects an account to its contacts, opportunities, meetings and products, helping a system find who was involved in a decision. Full definition of Knowledge graph.

Context graph

Context graph is an emerging label for a graph that connects business entities with interactions, events and, in some implementations, the evidence or reasoning behind decisions.

Sales example: A deal’s close-date change is linked to the meeting, procurement requirement and recorded explanation that prompted it. Full definition of Context graph.

Semantic layer

A semantic layer defines business terms, metrics and relationships consistently over underlying data so that different tools can interpret them in the same way.

Sales example: Sales and finance use an agreed definition of new-business revenue when an AI assistant answers a performance question. Full definition of Semantic layer.

Agent memory

Agent memory is information retained or retrieved across steps or sessions so an AI system can use relevant previous context.

Sales example: An assistant retrieves the last approved account plan and the date it was reviewed before preparing the next meeting brief. Full definition of Agent memory.

Context window

A context window is the amount of input and generated content a model can handle within a particular interaction, measured in tokens.

Sales example: A long transcript, several playbooks and the conversation history all compete for space when a seller requests deal guidance. Full definition of Context window.

Sales method layer

A sales method layer is a practical description for structured guidance that tells an AI system how a company wants sellers to assess opportunities and choose actions.

Sales example: It can contain qualification standards, stage guidance, behaviours, approved assets and rules for handling missing evidence. Full definition of Sales method layer.

Retrieval and connections

Retrieval-augmented generation (RAG)

Retrieval-augmented generation combines information retrieval with a generative model: the system finds relevant source material and supplies it as context for an answer.

Sales example: A seller asks about a security objection. The system retrieves the relevant approved security material before drafting an answer. Full definition of Retrieval-augmented generation (RAG).

Grounding

Grounding connects an AI response to relevant evidence, data or other reference points so the answer can be checked against an external basis.

Sales example: A deal review labels a budget statement as confirmed only when it can point to the buyer’s recorded statement or an approved record. Full definition of Grounding.

Embeddings

Embeddings are numerical representations of content that allow software to compare similarity, often helping retrieve material with related meaning.

Sales example: A search for concerns about implementation effort can retrieve a case study discussing onboarding workload even when the wording differs. Full definition of Embeddings.

Vector search

Vector search retrieves items by comparing numerical representations, usually embeddings, rather than relying only on exact word matches.

Sales example: An assistant finds a relevant objection-handling passage even though the seller describes the objection in different language. Full definition of Vector search.

GraphRAG

GraphRAG combines graph-based information structures with retrieval-augmented generation to help a model use connected information when answering questions.

Sales example: A system follows relationships between stakeholders, meetings and opportunities to assemble evidence about a buying group. Full definition of GraphRAG.

Model Context Protocol (MCP)

Model Context Protocol is an open standard through which AI applications can connect to external tools, resources and prompts exposed by compatible servers.

Sales example: An assistant uses an approved connection to retrieve an account record or invoke a permitted CRM function. Full definition of Model Context Protocol (MCP).

Tool calling

Tool calling allows an AI model to request an operation from an external function or service, with the surrounding application deciding how that request is executed.

Sales example: The model requests an account lookup, receives the returned record and uses it to prepare a briefing. Full definition of Tool calling.

Semantic search

Semantic search retrieves information using the meaning or intent of a query rather than relying solely on exact keyword matches.

Sales example: A seller looking for evidence about a slow rollout finds material described as implementation delay or extended onboarding. Full definition of Semantic search.

Instructions, controls and evaluation

Prompt engineering

Prompt engineering is the design and testing of instructions, examples and output requirements given to an AI model.

Sales example: A prompt asks the assistant to separate confirmed buyer evidence, assumptions and unanswered questions in a deal review. Full definition of Prompt engineering.

Agent skills

Agent skills are reusable packages of instructions and supporting resources that teach a compatible AI agent how to perform particular tasks. Formats and loading behaviour vary by platform.

Sales example: A discovery-preparation skill provides a call-plan structure, evidence standards and worked examples for a defined sales motion. Full definition of Agent skills.

Fine-tuning

Fine-tuning is additional training that changes a model’s parameters using a selected dataset to influence its behaviour or performance on particular tasks.

Sales example: A team may fine-tune a model on approved examples to improve the consistency of a specialist classification task. Full definition of Fine-tuning.

AI hallucination

An AI hallucination is generated information presented as factual even though it’s unsupported or incorrect.

Sales example: A call summary states that the economic buyer approved the budget when the transcript only says that funding will be discussed. Full definition of AI hallucination.

AI guardrails

AI guardrails are controls intended to limit an AI system’s inputs, outputs or actions, including permissions, validation rules and approval requirements.

Sales example: A workflow can draft a discount request but can’t approve a discount or send a commercial commitment to the buyer. Full definition of AI guardrails.

Human-in-the-loop

Human-in-the-loop means a person participates at a defined point in an AI-supported process, such as checking evidence or approving a consequential action.

Sales example: A seller reviews a suggested follow-up and its source evidence before anything is sent to the buyer. Full definition of Human-in-the-loop.

AI evaluations

AI evaluations, often called evals, are structured tests that assess an AI system against defined tasks and success criteria.

Sales example: A team tests whether deal guidance identifies missing decision-process evidence without inventing it, using a bank of representative opportunities. Full definition of AI evaluations.

Agent observability

Agent observability is the ability to inspect an AI system’s operation through records such as tool calls, retrieved sources, outputs, errors, timing and cost.

Sales example: When a briefing contains an incorrect claim, the team can inspect which source was retrieved and which tool result the assistant used. Full definition of Agent observability.

Prompt injection

Prompt injection is an attempt to make an AI system follow unauthorised instructions contained in user input or material it processes, such as a webpage, document or email.

Sales example: An account-research agent reads a webpage containing instructions to ignore its task and disclose internal information. Full definition of Prompt injection.

AI governance

AI governance establishes accountability, policies and review processes for how an organisation develops, uses and maintains AI systems.

Sales example: A sales team names owners for approved guidance, data access, output review and incident handling before introducing an assistant. Full definition of AI governance.

Sales applications

Conversational intelligence

Conversational intelligence analyses customer conversations to help teams understand what was discussed, identify trends and support review or coaching.

Sales example: A manager reviews a transcript and selected call moments to examine how a seller explored the impact of a problem. Full definition of Conversational intelligence.

Revenue intelligence

Revenue intelligence is a broad category of tools that combine commercial data to help teams understand pipeline, deal activity, customer relationships or revenue performance.

Sales example: A platform combines CRM and conversation information to flag opportunities where the expected buyer activity hasn’t occurred. Full definition of Revenue intelligence.

Signal-based selling

Signal-based selling uses relevant events or behaviours to prioritise accounts and shape sales activity. AI can help interpret signals, but the approach doesn’t require AI.

Sales example: A new revenue leader and a stated move into enterprise sales prompt research into whether an account needs different sales standards. Full definition of Signal-based selling.

Intent data

Intent data consists of observed behaviours used to infer interest in a topic, product or purchase. It can come from a company’s own channels or external providers.

Sales example: An account’s repeated visits to implementation and security pages may warrant research or a relevant follow-up. Full definition of Intent data.

AI lead scoring

AI lead scoring uses a model to rank or score leads against a defined outcome, such as becoming an accepted opportunity within a specified period.

Sales example: A team prioritises inbound enquiries using account fit and relevant engagement, then checks the results against subsequent qualification. Full definition of AI lead scoring.

AI sales forecasting

AI sales forecasting uses models to estimate future sales outcomes from historical and current commercial data, sometimes alongside human judgement.

Sales example: A model estimates likely revenue for the quarter, while managers review buyer evidence and explain material exceptions. Full definition of AI sales forecasting.

Next best action

A next best action is a recommended step selected for a particular situation using available evidence, objectives and decision rules. It may be generated by rules, AI or human judgement.

Sales example: If decision criteria remain unclear after discovery, the useful next action may be to agree how the buyer will evaluate rather than send a generic proposal. Full definition of Next best action.

AI data enrichment

AI data enrichment uses AI to help add, classify or interpret information about records such as accounts and contacts, drawing on identified sources.

Sales example: An assistant classifies a company’s sales motion from its public material and marks the classification as an inference with a source and date. Full definition of AI data enrichment.

AI sales coaching

AI sales coaching uses AI to provide feedback, practice or suggested improvements against defined selling standards, using inputs such as role-plays, call recordings or deal evidence.

Sales example: A manager reviews AI feedback on whether a seller explored business impact, then coaches the next conversation using the actual call. Full definition of AI sales coaching.

AI role-play

AI role-play is simulated practice in which an AI system plays a buyer or other participant so a seller can rehearse a conversation and receive feedback.

Sales example: A seller practises discovery with an operations leader who has a real problem but hasn’t agreed its financial impact. Full definition of AI role-play.

Account intelligence

Account intelligence brings together relevant information about an organisation, its people, priorities and commercial situation to support account decisions. AI can assist the research and synthesis.

Sales example: A brief separates a company’s published expansion plans from the seller’s hypothesis about the sales challenges those plans may create. Full definition of Account intelligence.

Next best opportunity

Next best opportunity describes a recommendation about which potential deal, account or expansion opportunity deserves attention, based on stated objectives and available evidence.

Sales example: A team prioritises an existing account where a relevant new initiative aligns with proven capabilities and an accessible sponsor. Full definition of Next best opportunity.

AI personalisation

AI personalisation uses AI to adapt content or interactions to a recipient, account or situation using available information. Hyper-personalisation usually describes a more granular or dynamic version of the same aim.

Sales example: A seller prepares a demo narrative around the buyer’s documented workflow and decision criteria. Full definition of AI personalisation.

AI proposal generation

AI proposal generation uses AI to prepare parts of a commercial proposal from approved source material and the evidence available for an opportunity.

Sales example: An assistant drafts a scope summary and maps requirements to approved capabilities, marking unconfirmed items for review. Full definition of AI proposal generation.

AI pricing

AI pricing uses models to support pricing decisions, such as estimating willingness to pay, recommending discounts or flagging unusual commercial terms.

Sales example: A pricing workflow suggests a range using permitted historical data and current deal limits, then routes exceptions to an authorised person. Full definition of AI pricing.

Teams and operating models

Human–AI sales team

A human–AI sales team is an operating arrangement in which people and AI systems have explicit, complementary responsibilities across sales work.

Sales example: AI prepares evidence and drafts a plan; the seller validates the buyer situation; the manager coaches judgement and owns commercial decisions. Full definition of Human–AI sales team.

Revenue orchestration

Revenue orchestration coordinates activities, information and hand-offs across commercial teams and systems. AI may help decide or execute some steps.

Sales example: A buyer’s question reaches the account owner, triggers the right specialist input and leads to a reviewed response with a clear next step. Full definition of Revenue orchestration.

Commercial impact journey

Commercial impact journey is terminology used by McKinsey for a connected commercial workflow organised around an outcome, bringing together data, decisions, people and AI across related tasks.

Sales example: For a small sales team, the practical unit might be discovery through to an agreed evaluation: preparation, evidence capture, follow-up, demo planning and manager review. Full definition of Commercial impact journey.

Discovery in AI answers

Answer engine optimisation (AEO)

Answer engine optimisation is a broad industry label for making content easy for answer-oriented search systems to find, interpret and use in responses.

Sales example: A sales glossary gives a direct definition, a practical example and clear links to related concepts on accessible pages. Full definition of Answer engine optimisation (AEO).

Generative engine optimisation (GEO)

Generative engine optimisation is an emerging label for improving a brand or source’s visibility in answers produced by generative AI systems, particularly those using web retrieval.

Sales example: A practical article explains AI sales terminology, links to primary sources and provides clear evidence about what a service actually does. Full definition of Generative engine optimisation (GEO).

The distinctions that matter when choosing a tool

AI agent vs AI copilot

A copilot usually describes assistance provided to a person. An agent describes a system that can take steps towards a task. The same application may do both. Ask whether it only prepares a recommendation, can update records or can communicate with a buyer. Those are different permissions with different review requirements.

Context layer vs knowledge graph

A context layer concerns the information and meaning made available to an AI system. A knowledge graph is one way to structure entities and relationships. A context layer can use documents and CRM records without a graph. A graph is worth considering when explicit relationships answer questions that simpler retrieval handles poorly.

Knowledge graph vs context graph

A context graph often emphasises events, interactions or decision history, but definitions vary. Ask what the product actually records. Does it know that an opportunity moved stage, or does it retain the evidence and explanation for the change? An inferred explanation shouldn’t be presented as a recorded fact.

RAG vs fine-tuning

RAG retrieves information for the model to use during a task. Fine-tuning changes model parameters through additional training. Making a playbook available to an assistant isn’t normally fine-tuning. Changing account information needs a reliable data route, not a claim that the model has memorised the business.

Prompt vs sales method layer

A prompt supplies instructions for an interaction. A maintained sales method layer can organise the standards, examples, assets and decision guidance used across many interactions. It still needs an agreed way of reaching the assistant. A repository of excellent material isn’t useful to the model unless the relevant material is actually supplied.

Sales automation vs agentic workflow

Sales automation can execute a fixed rule, such as creating a task after a meeting. An agentic workflow gives AI discretion over some steps. If the right action is already known, a fixed rule may be simpler to maintain. Use model judgement where it adds value and test what happens when the information is incomplete.

Conversation intelligence vs sales coaching

Conversation intelligence can help locate and analyse what happened on a call. Sales coaching concerns how the seller improves. Detecting the word “budget” isn’t evidence of good qualification. The useful question is what the seller established, how they established it and what they should do differently next time.

AEO vs GEO

AEO and GEO describe overlapping efforts to make information discoverable and useful in answers. Neither guarantees visibility. For Google’s AI search features, Google’s guidance retains the usual foundations: accessible pages, reliable content and established search practices. Inclusion in a retrieved answer is also different from incorporation into a model’s training.

What this looks like in a live opportunity

Consider an illustrative discovery call. The buyer describes a reporting problem and requests a demo. The seller hasn’t yet confirmed the cost of the problem, the evaluation criteria or who needs to be involved.

A useful AI-supported workflow would do five things:

1. Show the evidence. Identify what the buyer actually said and where it came from.
2. Name the gaps. Keep unknowns visible rather than guessing budget, authority or urgency.
3. Apply the agreed sales approach. Use the relevant stage guidance and qualification standard.
4. Propose a practical next step. Prepare discovery validation and a demonstration around the buyer's workflow, with the intended outcome made clear.
5. Leave the decision with the seller. The seller reviews the recommendation, checks the context and agrees the next step with the buyer.

The output should help the seller make a better decision. A longer summary or a higher number of completed CRM fields isn’t enough.

Next Best Actions is Closing Foundry’s offer for structuring a customer’s sales approach so an agreed AI environment can apply it. The method content and the current opportunity evidence remain distinct. Access to tools, integrations and any automated actions depend on the implementation agreed with the customer.

Questions to ask before you buy or build

• Which sales task are we improving? Be specific about the user, starting point and intended result.
• What does good look like? Define the evidence and selling standard before assessing generated output.
• Which information is required? Establish sources, ownership, access and update frequency.
• What can the system do? Separate reading, drafting, recommending, changing records and contacting buyers.
• How does it handle uncertainty? Test missing evidence, conflicting records and an unsuitable request.
• Who reviews the work? Give the reviewer enough evidence and authority to reject or correct it.
• How will we maintain it? Assign responsibility for the method, assets, configuration and testing.
• How will we know it helped? Measure the task first, then examine broader commercial results with appropriate caution.

Measure the work before claiming the revenue

Time saved has value, but it doesn’t automatically become more selling time or more revenue. Inspect what people do with the capacity released. Similarly, a higher acceptance rate for AI suggestions can indicate useful guidance or insufficient scrutiny. Review samples as well as dashboards.

Where Closing Foundry fits

If the problem is that AI doesn’t know your sales approach, Next Best Actions provides a way to structure the guidance for an agreed environment.

If the problem also includes inconsistent application, limited coaching capacity or unclear ownership of improvement, Strategic Sales Enablement addresses the broader work around standards, development and adoption.

The practical sequence is to define how you want the team to sell, make the relevant guidance usable in the workflow, and review its application in real opportunities. AI can support each step. The sales leader still needs a clear standard and people who can apply it.

Common questions

Do we need all these AI components?

No. Start with the task and its requirements. A configured assistant with approved guidance and manually supplied deal evidence may be enough for an initial use case. Add retrieval, integrations or more complex agent behaviour when a tested need justifies them.

Can AI qualify deals using MEDDPICC?

AI can help organise evidence against MEDDPICC, identify missing information and suggest questions. It shouldn’t turn an untested statement into proof of a Champion, approved budget or a confirmed decision process. Sellers and managers still need to validate the buyer situation.

Does a call transcript provide enough context?

It provides evidence from one interaction, subject to transcription quality. Useful guidance may also require the sales standard, earlier interactions, approved product information and the current opportunity state. The system should make gaps visible rather than silently fill them.

Can we use our current AI assistant?

Possibly. The answer depends on the task, permitted data, access route and the assistant’s capabilities. Start by verifying how the guidance and deal evidence will reach it, who can use them and what output or action it supports.

Should we build or buy an AI sales tool?

Compare a configured existing tool, a specialist product and a custom build against the same task and acceptance criteria. Include integration, maintenance, review effort and source ownership. A custom interface doesn’t solve an undefined sales standard, and a packaged product still needs a fit with the way the team sells.

Does AI replace sales enablement?

AI can assist preparation, feedback and access to guidance. Enablement also involves deciding what good looks like, developing people, helping managers coach and maintaining the approach. Those responsibilities need owners even when the tools become more capable.

How do we stop AI inventing deal evidence?

Reduce the risk through clear evidence standards, relevant sources, controlled permissions and evaluation on difficult examples. Ask the system to distinguish confirmed facts, interpretations and unknowns. Human review remains important; no prompt or product label establishes zero risk.

Is Next Best Actions a complete GTM AI harness?

It’s a custom sales method layer for an agreed environment. A complete harness may also manage execution, tools, state, permissions and monitoring. The method package can contribute to that environment without being the whole system.

Further reading

• Why MEDDPICC fails and how to make it stick
• How to create a sales playbook that works
• Explore the full sales glossary