Best sales AI assistants

A sales AI assistant is a layer over the tools a rep already uses, and its job is to remove typing.

Four tasks make up most of the category: transcribing and summarising calls, writing CRM fields and activity records, drafting follow-up emails and next steps, and answering questions about an account or a deal from the data already captured.

Nothing in that list generates pipeline on its own. The payback is rep hours and CRM completeness, which is why the business case has to start from how much admin time you are buying back.

The category is larger than it looks because three different product types sell into it. Meeting assistants such as Otter.ai record and summarise, and price per seat in the $8 to $40 range.

CRM-native assistants such as HubSpot Breeze and Microsoft Copilot for Sales come bundled into a seat you already pay for and meter the AI work by credits. Revenue-team assistants such as Aviso and Clari Copilot sit on the inspection side and quote per customer.

Measure the problem before you shop. Ask five reps to log, for one week, the minutes they spend on notes, CRM updates and follow-up drafting after each customer call. Most teams land between 20 and 40 minutes.

Multiply by calls per week and headcount, and you have the number any quote has to beat. The sales AI assistants category lists the vendors we track with a fit line and a directory-level price for each. All figures checked September 2026.

The strongest case is a team with high meeting volume and low CRM completeness.

If each rep runs ten customer calls a week and the opportunity records are still missing next steps and economic buyers, an assistant recovers both the time and the data in one purchase.

The weakest case is a team with few, long, high-value meetings, where a rep who writes their own notes is doing part of the thinking rather than wasting time.

Seat caps decide eligibility more often than budget does.

Otter.ai Pro is capped at 5 billed users and Business at 25, and CRM note sync is limited to 1 user on Pro and 5 on Business, so a 10-rep team that wants every rep syncing to Salesforce is in an Enterprise conversation regardless of what the per-seat price says.

Winn.ai sets a 20-seat minimum on its Enterprise plan, which prices out smaller teams that want admin controls.

Two prerequisites matter more than the product. The first is a CRM whose required fields are defined, because an assistant that writes into a schema nobody agreed on produces tidy records of the wrong things.

The second is recording consent, which is a legal question in two-party-consent jurisdictions and a trust question everywhere. Decide both before rollout rather than after the first rep objects.

If what you want is better forecast inspection rather than less typing, you are shopping in the wrong category. Read our guide to evaluating revenue intelligence software instead, because the products there are bought to change what managers see, not what reps type.

Six capabilities separate these products, and only two of them show well in a demo.

Transcription accuracy on your accent mix and your product vocabulary is the first, and the only way to test it is to feed the tool three of your own recorded calls, including one with a strong regional accent and one full of product names.

Second, CRM write-back: does the assistant propose a field update for a human to accept, or does it write directly, and can an admin restrict which fields it touches? Direct write-back without field-level control is how a clean CRM becomes a confidently wrong one.

Third, who needs a paid seat. Managers, enablement and RevOps often only view calls and dashboards, and vendors differ on whether those count. Avoma makes viewers and collaborators free.

Attention does not publish rates at all, so ask whether a 10-rep team becomes 13 or 14 seats once managers are added.

Fourth, retrieval across deals rather than within a call. An assistant that answers questions about one meeting is a transcript tool. One that answers what the buyer said about pricing across six calls and two threads is doing something a rep cannot do by hand.

Fifth, the credit model, covered below. Sixth, export: can you take transcripts and summaries with you, in a usable format, if you do not renew? A call archive you cannot export is a switching cost, not an asset.

If the CRM vendor already has your seats, the AI usually arrives as a credit allowance rather than a new subscription, and the cost shows up at the point you scale usage.

HubSpot no longer sells Breeze as an add-on: every paid Core Seat includes the Breeze Assistant at no extra charge, and Breeze agents consume HubSpot Credits.

Each hub includes an allowance, 500 a month on Starter, 3,000 on Pro and 5,000 on Enterprise, you get the highest single allowance you qualify for rather than the sum, and unused credits expire at month end. Extra credits are $0.01 each.

The Prospecting Agent costs 100 credits per lead, so a Pro team's 3,000 credits cover 30 leads a month; researching 500 leads uses 50,000 credits and costs $470 a month on top of seats.

Salesforce sells its sales AI as Agentforce for Sales, either as an add-on from $125 per user per month on Sales Cloud Core and Advanced, or bundled in the Max edition at $550 per user per month billed annually.

For 10 reps the add-on route on Core is 10 x ($195 + $125) x 12 = $38,400 a year, against $66,000 for Max, so Max only pays off if you would otherwise buy several of the products it bundles.

Agent actions draw down a Flex Credit pool: 500,000 a year on Core, 1 million on Advanced, 2.75 million on Max.

Microsoft no longer sells Copilot for Sales as a standalone line. The capability ships as the Sales agent inside Microsoft 365 Copilot at $30 per user per month paid yearly, with Copilot Business at $21.

Ten sellers is $3,600 a year on top of their Microsoft 365 and CRM licences, and Dynamics 365 Sales Enterprise at $105 per user per month adds $12,600.

This is the cheapest end of the category and the one most teams should start in, because it addresses the measurable problem: rep minutes lost after every call.

Otter.ai publishes the lowest rates here. Basic is free with 300 minutes per user per month and a 30-minute cap per conversation, which cuts off most discovery calls. Pro is $8.33 per user per month billed annually with 1,200 minutes and a 90-minute cap.

Business is $19.99 per user per month annually and removes the monthly minute limit. A 10-rep team on Business pays $2,398.80 a year, but only five of those reps can sync notes to Salesforce or HubSpot.

Minutes do not roll over and HIPAA is an add-on.

Buyers should also know that In re Otter.AI Privacy Litigation, No. 25-cv-06911-EKL, is a consolidated class action in the Northern District of California over recording and training on conversations without participants' consent; on 13 August 2026 Judge Eumi K. Lee let the federal wiretap, California privacy and Illinois biometric claims proceed.

Liability is undecided, and legal should see that before a company-wide rollout.

Avoma prices three base plans per paid seat: Startup at $19, Organization at $24 and Enterprise at $39 per seat per month on annual billing. Conversation Intelligence and Revenue Intelligence are separate add-ons at $29 per seat per month annually each.

Ten seats on Organization with Conversation Intelligence is $6,360 a year; adding Revenue Intelligence takes it to $9,840. The base plans cover notes and transcripts only, so decide which reps need which add-on.

Winn.ai and Attention sit above both and neither publishes a figure. Winn.ai is built around live in-call guidance and automatic CRM updates, with a 20-seat Enterprise minimum.

Attention targets CRM field automation and next-step extraction; its pricing page returned a 404 in September 2026 and it raised a $30M Series B in June 2026, so expect packaging to move.

These products assist the manager more than the rep, and all of them quote per customer, so budget a procurement cycle rather than a credit card.

Aviso is an assistant layer over forecasting and deal intelligence. It publishes no prices, but Vendr reports a median Aviso contract of $73,662 a year, about $6,139 a month; for a 50-user forecast that is roughly $123 per user per month.

Treat it as a negotiating benchmark, not a rate, and ask whether executives who only view the forecast are priced below full users.

Clari Copilot, the call recording and AI summary product formerly called Wingman, is now sold by Salesloft following the merger that closed in December 2025, and is quoted as part of a Clari Forecast or Salesloft engagement bundle.

ZoomInfo Copilot is sold as GTM Workspace and priced inside the wider ZoomInfo subscription by seats and data packages.

Vendr reports a median ZoomInfo contract of $33,500 a year across 1,573 purchases but does not break out the AI layer, so ask for Copilot priced as an increment on your current contract and separately as part of a new package.

Add-ons bought mid-contract usually co-term, which changes the renewal number more than buyers expect.

Four narrower options round out the set, none publishing rates. Revenue.io sells seat-based plans to Salesforce teams of 15 seats and up. Dialpad Sell combines calling with AI coaching. aiOla runs voice agents that let field reps update Salesforce by talking after a visit.

ASPR AI offers either per active user per month or outcome-based pay-as-you-win with unlimited users, which is worth a conversation if your volume is lumpy.

Fit, published price and the dimension that decides each purchase, as of September 2026.

Otter.ai: transcripts and summaries at the lowest published rate. Free tier; Pro $8.33, Business $19.99 per user per month annually. Decider: the CRM sync cap, 1 user on Pro and 5 on Business.

Avoma: notes now, coaching later. Startup $19, Organization $24, Enterprise $39 per seat per month annually, add-ons $29 each. Decider: which reps need which add-on, since viewers are free.

HubSpot Breeze: teams already paying for Sales Hub seats. Included in every paid Core Seat; extra credits $0.01. Decider: credit burn, at 100 credits per researched lead.

Microsoft Copilot for Sales: sellers living in Outlook and Teams. $21 to $30 per user per month paid yearly, on top of CRM licences. Decider: which Microsoft SKU includes the Sales agent.

Salesforce Agentforce for Sales: Salesforce-first enterprises. From $125 per user per month as an add-on, or $550 in Max. Decider: Flex Credit consumption per agent action.

Winn.ai: live in-call guidance. Quote only, 20-seat Enterprise minimum. Decider: whether CRM write-back sits in the lower plan.

Attention: automating CRM fields straight from calls. Quote only, no published rate. Decider: whether managers and RevOps viewers need paid seats.

Aviso: assistance on the forecast. Quote only; Vendr median $73,662 a year. Decider: whether view-only executives cost a full seat.

ZoomInfo Copilot: teams already on a ZoomInfo contract. Quote only, sold as GTM Workspace. Decider: the increment over your current subscription.

Three models operate here and they reward different team shapes. Per seat is the simplest: Otter.ai at $8.33 to $19.99, Avoma at $19 to $39 plus $29 add-ons, Microsoft at $21 to $30.

Per seat punishes growth but is easy to forecast, and the trap is seat caps and feature caps that bite before the price does.

Credit or consumption pricing is how the CRM vendors charge, and it punishes scale instead of headcount.

HubSpot's Prospecting Agent at 100 credits per lead and extra credits at $0.01 means the marginal cost of researching a lead is a dollar, which is fine at 30 leads a month and $470 a month at 500. Salesforce's Flex Credits behave the same way.

Model your intended volume at the published credit rate before signing, because the included allowance is sized for a pilot rather than production.

Quote only covers the forecast-side assistants and the premium note-takers: Aviso, Clari Copilot, ZoomInfo Copilot, Revenue.io, Dialpad Sell, Winn.ai, Attention, aiOla, ASPR AI and Spekit's AI Sidekick.

For each one, ask four things in writing: the per-seat price by module, whether view-only users need paid seats, what AI usage is included in the seat versus metered, and a cap on the renewal uplift.

The add-on structure is the quiet cost. Avoma's base plans exclude coaching analytics and deal intelligence; HubSpot moved quoting out of Sales Hub into a separately priced hub; Salesforce prices Revenue Intelligence from $220 per user per month on top of the edition.

Build a three-year total before comparing anything, and our sales tool ROI calculator is where to hold seat, credit and add-on lines as one number.

An assistant that reps distrust produces worse data than no assistant, because reps start managing what the tool sees. Three things decide adoption, and all three are policy rather than product.

First, tell reps what is recorded, who can see it, and what it is used for. The honest answer is usually that managers can see transcripts and summaries, and that is fine said once, clearly, at rollout.

Discovering it in month three turns a note-taker into a surveillance complaint. Second, separate coaching from inspection: if the only time a rep hears about a transcript is in a performance conversation, the tool becomes a threat.

Third, give reps something back on day one, such as pre-filled next steps and a searchable record of what the buyer said.

Consent is the legal half. In two-party-consent jurisdictions, every participant has to agree before recording, which means an announcement at the top of the call and a documented process for external attendees who decline.

The Otter litigation turns on exactly this question for people who hold no account with the vendor, so your counsel should set the policy rather than the vendor's help centre.

Finally, decide what the assistant writes without approval. The safe default is that it proposes CRM field updates and a human accepts them, with field-level restrictions on anything that feeds the forecast.

Close date, amount and stage should stay human-owned, because a model that nudges them is quietly editing the number leadership commits to.

Work the decision in this order. Measure the admin minutes per call across five reps for one week, so you have a number any quote must beat. Decide which of the four tasks you are buying: transcription, CRM write-back, drafting, or cross-deal retrieval.

Count the real seat requirement including managers, enablement and RevOps. Then test transcription on three of your own recorded calls rather than the vendor's sample.

Start cheap and narrow. For most teams the right first purchase is a per-seat meeting assistant at $8 to $40 a seat, because it is the only part of the category with published rates, a free tier to test on, and no procurement cycle.

Avoma at $19 per seat per month billed annually and Otter.ai Business at $19.99 both clear that bar.

If your CRM vendor already includes an assistant in the seat, as HubSpot does with Breeze, run that first and spend the budget on credits instead of a second subscription.

Move up only when a named job fails. Live in-call guidance, forecast-side inspection and cross-deal retrieval are the three reasons to take a custom quote, and each should be written down before the first demo.

Score vendors on one sheet with our sales software evaluation scorecard, and run a paid pilot rather than a free trial above a few thousand dollars a year.

If the job is finding and researching accounts rather than writing up calls, our guide to AI prospecting software covers that end of the stack.

The best sales AI assistant is the one whose output a rep would have written anyway, arriving before they sit down to write it.