87% of sales teams use AI. Silos slow half.

Sales was one of the first functions to put AI into daily work, and the adoption numbers now look settled. Salesforce's State of Sales 2026 survey reports that 87% of sales organisations use AI across the sales cycle. It is a vendor survey, so read it as a direction, not a census. The more useful number sits next to it: 51% of sales leaders who use AI say disconnected systems slow their AI initiatives. The bottleneck has moved from the model to the CRM.

ETHORITY reviewed 25 sales AI offerings for the Good AI Prism in October 2026, reading each provider's own pricing, security and legal pages. The aim was simple: find the facts a buyer needs before signing, and see how many providers put them on the record.

Five jobs along one pipeline

The Sales & CRM area map groups the offerings by the job they do. Prospecting and data tools such as ZoomInfo Copilot, Claygent, Apollo.io and Leadfeeder AI find and enrich accounts. AI sales development agents such as Alice from 11x, Ava from Artisan and RegieGO write and send outreach. Conversation intelligence tools such as Gong, Modjo, Avoma, Sybill and Einstein Conversation Insights record and analyse calls. Enablement platforms such as Showpad Genie, Highspot Nexus and Seismic Aura AI put content in front of sellers. CRM copilots and forecasting tools such as HubSpot Breeze, Dynamics 365 Sales Copilot, Zia, Pipedrive AI, Attio, Aviso, Outreach and Salesloft sit inside the system of record.

Each job touches different data. A prospecting tool works on third-party contact data. A call recorder holds the voices of your customers. A CRM copilot reads your whole pipeline. The due diligence should follow the data, not the feature list.

Four pricing units, one budget

Of the 25 offerings, 12 publish a price. The units differ more than the amounts. Pipedrive lists plans from US$14 per seat per month, billed annually. Microsoft includes Copilot in Dynamics 365 Sales Enterprise at US$105 per user per month. HubSpot charges overage for Breeze at US$0.01 per credit. 11x prices its Alice agent from US$3,750 per month for a set number of new prospects.

Seat, credit, bundle and digital-worker pricing produce different cost curves. Seats scale with headcount. Credits scale with usage and can surprise a finance team at quarter end. A digital-worker subscription is priced like a junior hire and should be judged like one: by meetings booked and pipeline created, not by emails sent. Put candidates into Compare side by side and convert each to cost per qualified meeting or per opportunity.

12 of 25 sales AI offerings publish a price on their own pages. The other 13 quote on request.

Scenario. For 20 sellers, Pipedrive's entry price of US$14 per seat comes to US$280 a month; Dynamics 365 Sales Enterprise at US$105 per user comes to US$2,100. One Alice subscription from 11x starts at US$3,750 a month. The comparison becomes fair only when each is divided by qualified meetings or opportunities created.

Where the customer data goes

Hosting and training terms separate the field more than features do. Gong lets customers choose US or EU storage, with the US as the default, which makes the choice a setup task rather than a given. Apollo.io states that it hosts with US-based data centre providers.

On training, some state that customer data can train or improve their models, at least by default. Lavender says its terms allow training on de-identified customer data. HubSpot and Apollo.io offer opt-outs for their own models. For a sales team, that data includes emails, call transcripts and deal notes about named people at your customers. Check the default before the first import.

Agents raise the stakes

Most vendors now sell agents that act, not only assistants that suggest. Salesforce reports that nearly 90% of respondents plan to adopt AI agents by 2027. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, mostly on cost, unclear value or weak risk controls. Both can be true. An AI agent that books meetings creates value only if the meetings convert, and risk if it writes to the wrong person in the wrong tone.

Agents that talk to prospects also bring rules. Under the AI Act, as amended by Regulation (EU) 2026/1744, AI systems that interact directly with people must say so, a duty that applies since 2 Aug 2026. Our piece on why tool access is not authority sets out how to bound what an agent may do in your name.

What good looks like after 90 days

A sales AI rollout that works leaves a clear trail after one quarter. The team can name the one metric it set out to move and show the before and after values from the CRM, not from the vendor's dashboard. Every AI-generated email, call summary or forecast has an owner who can explain how it was produced and who corrected it when it was wrong. The data fields the tool depends on have improved, because the tool exposed the gaps. And finance can reconcile the invoice with usage: seats in use, credits consumed, meetings booked.

The failures look different. Licences are bought for the whole team but used by a few. Outreach volume rises while reply rates fall. Forecasts look more precise but miss by the same margin as before. Call recordings pile up without anyone reading the analysis. None of these failures is a technology problem. Each is a management decision that was skipped at the start: which metric, which data, which owner, which stop rule.

51% of AI-using sales leaders say disconnected systems slow their AI initiatives (Salesforce State of Sales 2026, via CX Today).

Why the data plumbing decides

The Global Market Study's advice for this function is short: check CRM and data integration before buying, then measure selling time, win rate and pipeline per rep. That order matters. A copilot on top of an incomplete CRM produces confident summaries of wrong data. A forecasting tool trained on stale stages forecasts the past.

The practical test is a data inventory before the tool shortlist. Which fields are complete, who owns them, and which systems hold the rest? The answers usually narrow the shortlist faster than any demo. They also tell you whether the tool you need is a sales AI at all, or a data job that has to happen first.

What to do Monday

1. Pick one pipeline metric. Selling time per rep, win rate or pipeline per rep. Record today's value.
2. Map the data before the tool. List the CRM fields the AI will read and write, their owners and their completeness.
3. Shortlist three by job. Use the area map and keep each candidate in its job group; do not compare a call recorder with an SDR agent.
4. Ask the training question in writing. Default setting, opt-out route and whether call recordings are included.
5. Fund skills with the licences. In Germany 57% of companies with 20 or more staff now use AI (Bitkom, Sep 2026), yet skills lag. Our AI literacy programme covers sales teams.

For the hand-over from sales to service, see the Customer Service topic area; for campaigns feeding your pipeline, the Marketing & Content topic area.