The GTM Tech Index Brief

Two weeks in one line

The assistants got the keys and an allowance. Planhat let Pi send the email, Close kept Chloe off the send button, and Notta gave admins a credit cap per member, which is the Halloween candy rule applied to software. Clay and Boomerang made access follow the identity provider, HubSpot let AI connectors read sensitive fields on a Super Admin switch, and prices moved to the minute, the domain found and the scan that found nobody.

This issue covers October 18 to 30, with 44 entries across 44 vendors, one each. Every one of them was verified at the vendor’s own source, which no earlier issue of this Brief has managed. Nine were rated high impact, twelve were filed against the autonomy and oversight axis, and seven landed on October 20 alone.

It is Halloween, and the assistants came to the door in costume: dressed as colleagues, asking before they touch anything, and carrying a credit limit.

The agent got an allowance

The last issue watched autonomy turn into a dial with named positions. This period the dial grew a meter.

Planhat’s Pi can now search the public web and draft and send emails from inside a conversation, so a user can research an account before a call and have Pi send the follow up. Admins choose tool by tool whether web search and sending are available. Pi has moved from drafting the email to sending it, and the decision about whether it may sits in an admin panel rather than a prompt.

Close went the other way and made it policy. Chloe Chat left beta for every customer on every plan. It answers questions about leads and pipeline, briefs a user on the day’s meetings, drafts emails and can create or update records. It never sends the emails it drafts, and every record change waits for approval. Past a free weekly allowance per user, Chloe draws on the organization’s AI credits where those are switched on.

Autonomy and budget arrived in the same release note. A chaperone with an allowance is a very specific product design, and it is a sensible one.

Vitally launched Quin in beta, included in current plans for existing customers. Quin answers questions across a whole book of business and acts on them, updating properties on Customers, Organizations, Tasks, Notes and Projects, suggesting Key Role assignments that wait on a review card, and searching the web. Admins can switch off web search and page fetching in security settings.

Gainsight released Agent Studio in beta, where teams describe workflows in plain language and reuse them as skills, including scheduled runs such as a nightly briefing on accounts at risk. It can close a CTA, update a success plan and build dashboards and page layouts. Users approve every action before it runs, existing permissions still apply, and admins manage MCP connections, access and token usage per user.

Notta made the budget explicit. Workspace owners can turn Notta Brain on or off for everyone, set a shared credit cap per member, grant exceptions by member or by CSV, and control slide generation, image generation, shared database upload, smart input and scheduled tasks separately. New members inherit the settings. It is the Halloween candy rule applied to software: you may have some, not all of it, and the grown ups decide.

Otter.ai launched Work Mode for all paid users. It builds briefs before meetings from past conversations, open action items and connected apps, and after calls drafts follow up emails, documents and CRM record updates for review. Live Assist, private coaching prompts in real time inside Zoom, Teams or Google Meet, reaches enterprise customers now and the rest of the paid plans over the next few weeks, with custom agents built on a team’s own playbooks.

Demandbase launched three agents: an AEO Agent that compares how the major AI assistants describe and recommend a brand to each member of a buying group, a Campaign Agent that ties campaigns to revenue goals and recommends adjustments, and an Audience Agent that builds and maintains audiences from ICPs and buying groups. They run on a headless architecture that opens Demandbase data to outside assistants over APIs and MCP. Demandbase says marketers stay in control, and the announcement does not describe an approval step, which is the sentence a buyer will want filled in.

Smaller agents learned manners. Calendly’s Callie can now be emailed directly or sent a forwarded thread with private scheduling instructions the invitee never sees, and books over an existing event only when told it may. FirstTouch added a Slack agent that builds lists, campaigns and flows for LinkedIn outreach, proposes the audience and the steps, and shows a preview before anything runs. SPOTIO lets admins write up to 15 custom one tap prompts for its field copilot, test each before it reaches reps, and changed the default prompts to perform the task, such as logging an activity, rather than explain it.

Demodesk’s assistants and user built agents can now create battlecard decks and write, update and delete cards, as can a connected outside AI tool, within the user’s own permissions. Qwilr released AI Context, shared instructions such as brand voice or a sales playbook that apply automatically to AI image generation, page creation and, on the Scale plan, variable prefill. Set once by an admin, applied to every rep’s output, no prompt discipline required.

Our read

The permission slip from last issue has grown a spending limit. Close draws credits after a weekly allowance, Notta caps credits per member, Gainsight meters tokens per user, and Vitally leaves web access to an admin switch. Autonomy and budget are turning into one setting, held by one admin, which means the console where someone decides what the agent may do is now also where someone decides what it may cost. Finance will want a login.

Who is allowed in the house

Trick or treat ran on a guest list this period. Several releases were about who gets through the door, what they may read once inside, and where the data sleeps.

HubSpot let Super Admins allow supported AI connectors to read Sensitive Data properties and engagements in accounts that have Sensitive Data turned on, in public beta. ChatGPT, Claude and Copilot can read both, Gemini and the Remote MCP server can read engagements only, access is read only, Highly Sensitive Data stays restricted, and users still need their CRM object and property permissions.

That replaces a blanket block with a bounded one, on a switch only a Super Admin can flip. It is the right shape, and it is also the moment the question moves from whether an assistant may see the sensitive field to which assistant, which is a harder meeting.

Clay added SCIM provisioning through the WorkOS Admin Portal, so Okta, Azure AD or another identity provider adds and removes workspace members automatically, with provider groups mapped to the Admin, Editor, Viewer and Sales rep roles. SSO is not a prerequisite. The same release lets the Write to Audiences node upsert by external record ID instead of creating duplicates, which parallel workflow runs will appreciate.

Boomerang AI shipped the whole enterprise access set at once: SAML SSO, domain verification through a DNS record, SCIM provisioning and deprovisioning, and API keys that can be restricted to up to 50 IP addresses or ranges. For a tool that reads CRM and calendar data, that is the list a security review opens with.

Aircall extended regional data storage to its AI products. AI Assist Pro data for accounts created on or after October 1, and AI Voice Agent call logs, extracted data and post call actions for accounts created on or after October 6, now rest in AWS Frankfurt for the EU or AWS Sydney for Australia, with US Oregon as the default. Existing accounts are not covered yet, with migration targeted for the end of 2026, and account, metadata, billing and SMS data stay mostly in the United States.

An EU or Australian buyer can now give legal a named data center for the AI conversation data, which is the question that has stalled more than one voice agent purchase. The two date thresholds are the detail to check before assuming an existing account is covered.

GetResponse added open tracking consent. Accounts can ask subscribers before tracking email opens, choose whether to keep, anonymize or stop tracking for contacts who have not decided, put an editable consent checkbox on forms, landing pages, websites and webinar registrations, and let subscribers change their mind on the preference page. Consent status and history show on the contact card and can be set by import or API and used in segments.

It is built for accounts in France and Italy, or sending there, where CNIL and Garante guidance now asks for that consent. The practical consequence is in the automations: anything built on a Message opened condition sees fewer opens once contacts decline. A tool asking permission to know whether you opened the email is more courtesy than most of us extend before reading over a colleague’s shoulder.

GetAccept drew a line inside the contract. Teams can now comment and suggest tracked changes internally, as a draft and after sending, with comments labeled Internal or Visible to recipients and accepted changes on a sent contract held until a new version is published. The setting is off until an admin turns it on.

Our read

A year ago the security questionnaire asked whether the tool had SSO. This period the tools answered before being asked, then went further: who may read the sensitive field, which country the AI transcript sleeps in, whether the subscriber agreed to be watched opening the mail. Each of those is a setting a buyer can point to in a review, and most of them ship cautious: GetAccept’s internal suggestions stay off until an admin turns them on, Aircall’s regional storage covers new accounts first, and the HubSpot door opens only when a Super Admin opens it, which is where that door belongs.

The record filled up

Everything agreed to talk to everything else. In a horror film that is the moment before the lights go out. In GTM software it is a Tuesday, and this period the connections were unusually clear about what travels and what does not.

Vector released Personas, Audiences and CRM sync. Personas define buyer roles once for ad audiences, activity monitoring and CRM sending, audiences recheck membership weekly and add or remove people as they qualify, and CRM sync writes job changes, LinkedIn engagement and visits to high intent pages to Salesforce and HubSpot contacts and accounts, with a stage and a plain language explanation. New customers get it first, with existing customers moved over by their Vector team.

Salesforge’s Salesforce integration now syncs both ways for Leads and Contacts, updates, custom variables and optional deletions included, and logs sequence emails, replies, LinkedIn messages, InMails and connection requests on the matching record, which brings it level with what its HubSpot integration already did. Attio shipped Activities, which pull support tickets, social DMs and product events onto a record as structured data with a custom schema, so an activity can trigger a workflow, such as opening a deal after a call. Reporting on them is in alpha.

Octave now imports recorded Zoom Phone calls through the existing Zoom connection, so phone conversations feed buyer needs and objections into its library suggestions. The quieter part is a new Ingestion Overview showing Received, Filtered, Pending and Processed totals per provider and comparing Zoom and Gong totals with the records those accounts still make available, so an owner can trace a missing call without a support ticket.

Showpad’s Fall release added MCP connectors for Gong, Map My Customer, Badger and Databricks, so its Genie Agents recommend content against the deal in front of the seller rather than static content alone, with availability varying by tier. Demandbase opened its intelligence to outside assistants the same way, as above.

The data vendors opened agent doors with the reveal held back. Lusha’s new Recommendations API returns a ranked queue of contacts and companies built from a team’s Target Profiles, each with a lead score, an ICP fit score and a signal score, recalculated daily and again within minutes when a profile changes. It carries no emails or phone numbers. Those come only when a returned id is passed to Enrich, so contact data and credit spend follow an explicit step.

folk put contact enrichment on its API and MCP, up to 500 people a request and automatically for new leads, so a team’s own agents can fill in emails and phone numbers without the interface. Generect added a request that turns a company name into its domain, alone or in batches of 50, with the source and confidence behind each match, at $0.005 per domain found and nothing for a miss.

Coresignal’s Agentic Search can now take auto as its entity and decide from the prompt whether to search employee, company or job data. When it cannot decide with enough confidence it returns an error rather than a guess, which is a rare and welcome personality trait in an API.

Swarmhit released an official JavaScript and TypeScript SDK in beta that covers every endpoint, types the results and webhook events, retries safe calls and verifies webhooks. And Positive Group acquired Humanlinker, making its AI prospecting available inside Positive noCRM as a first integration, reaching about 8,000 noCRM users. Humanlinker keeps its name, product and support, and its founders now lead Positive’s Sales division.

Our read

The plumbing is getting modest about what it knows, and that is the right temperament for something feeding an agent. Lusha’s queue carries scores and no phone numbers until someone pays to see one. Octave’s ingestion page shows what it filtered. Coresignal would rather fail than guess. Vector writes a plain language reason beside every signal it drops into the CRM. A record that fills itself up is only useful if a rep can tell where each line came from, and this period most of them could.

The meter moved to the request

The last issue watched the seat give way to the result. This period the unit got smaller.

Pipeline CRM launched Pipeline Phone, a dialer inside the CRM for calls in the browser or on a rep’s cell. Every call lands on the record with its recording and transcript, reports track volume, talk time and outcomes, and a call can trigger an automation. It is a paid add on billed per minute from a prepaid balance with no per user fee: $0.025 a minute incoming and $0.03 outgoing to US and Canadian numbers, $2.00 a month per number, and switching on recording plays an announcement to callers.

Signado moved every scheduled lead source to one scan price. A scan that finds nobody costs 1 credit, a scan that finds people costs 5 credits per new lead and 1 per person already on the list, keyword sources now pay the scan fee too, and first scans and sources that run once pay nothing. A keyword source on a daily schedule now spends a credit even on the days it finds no one, so the bill tracks how often a source runs as well as what it finds.

Breakcold changed how its public API and MCP requests are charged. Each route now carries its own usage fee against plan usage instead of a token balance deduction, a breaking change for anything built on the old logic. Since October 24 a request past the limit returns HTTP 402 with the code usage_limit and the reset time, and the read and write fees were lowered on October 27.

402 Payment Required has sat in the HTTP specification, reserved for future use, since the 1990s. It turns out the future it was waiting for was an AI agent that will not stop asking.

The allowances from the first section belong here too: Close’s free weekly allowance per user before AI credits apply, Notta’s cap per member, Gainsight’s token usage per user, Generect’s $0.005 per domain with misses free, and Vitally’s Quin included in current plans while it is in beta.

Our read

A minute, a domain, a scan, a request. Each is a price finance can put in a cell, and none compares to another vendor’s until someone asks what the unit is. The newer part is the stop: a 402 with a reset time, a weekly allowance that rolls into credits, a cap per member. A price tells a budget what one unit costs. The stop tells it how the bill ends, which is the half of the question budgets have been missing.

The phone rang from a local number

Klenty launched Dynamic Presence for its dialers. It picks a local number matched to each prospect’s area code from a pool of more than 350 numbers covering all 335 US area codes, rotates by prospect location rather than call volume, falls back to the closest area code, and reuses the number a prospect previously answered. It works in both the parallel and power dialers. The claim of twice the connect rate is Klenty’s own figure.

Rotating caller IDs is also the pattern carrier spam filters watch for, which is why the detail about reusing the number a prospect already answered is the one that matters. A number that keeps its history keeps its reputation.

Aloware fixed the handoff. Calls transferred from its AloAi voice agent now arrive with the agent name, the reason for the transfer and a short summary, so the caller does not start over. A Leave Voicemail tool routes callers to a chosen line or user’s voicemail, Eleven V4 is a selectable voice model, and lines get separate missed call handling for business hours and closed hours, including forwarding after close.

Deliverability got two more dials. Snov.io now verifies emails inside a prospect list and can check them again, with Bouncer and MillionVerifier available alongside its own verifier, a last verified date on each email, a refresh for anything older than 30 days, and bulk verification on token packages without a Sales Suite plan.

FoxReach can match the sender to the recipient’s provider: the first email to a lead on Google goes out from a Google mailbox, and to a lead on Microsoft from a Microsoft mailbox, if one has daily capacity left. It is off by default, follow ups stay with the first mailbox, and it can be switched through the API, the MCP server, the SDKs and the command line. It is the outreach version of arriving at the door in a costume that matches the house.

SuperSend lifted the June cutoff that had kept LinkedIn away from organizations created after that date, so LinkedIn is available to every organization again, new signups included. The change follows a move onto a new connection provider, and admins can still turn LinkedIn on or off per organization or user. The release says the provider changed. It does not say what changed about the exposure to LinkedIn’s terms, which is the question a buyer should ask before switching it back on.

Our read

Deliverability features and evasion features look alike from a distance, and the honest ones tell them apart with a default and a disclosure. FoxReach ships its matching off. Klenty labels its connect rate as its own. SuperSend kept the admin switch and said the provider changed. The question is the one this Brief keeps asking: who holds the default, and do they know what it is for.

Market notes

Evidence got a method attached. Reevo now reviews the emails, calls and notes of any deal marked Closed Lost that had at least one meeting and writes a brief with a main reason, a confidence level and up to three secondary reasons, each linked to the source email or transcript line, sorted into one of six categories. The reason a deal died is no longer the word pricing typed by the person who lost it.

Salestools introduced Attribution, a journey view per account that joins paid and organic visits, outbound email, LinkedIn, calls, meetings, chat and CRM activity, with first touch, last touch, linear, U shaped, W shaped, time decay and data driven models for replies, meetings, opportunities and won stages. Ad impressions without a tracked visit are left out, and the data driven model falls back to linear until it has enough data, which is the kind of caveat attribution tools rarely volunteer.

Fullcast moved compensation plans, draw plans, crediting rules and the commissions ledger onto a new engine, history included. Every ledger entry now carries a step by step explanation from the commissionable amount to the tier that paid it, team payee splits credit every role holder, and plans can run without targets. For anyone who has argued with a comp statement in December, the explanation is the feature.

MaxIQ lets admins hide forecasting hierarchy levels, with deals rolling up to the next visible one and vacant levels substituting the level above, so a forecast can follow an org chart with gaps. Its Rooms gained join request approval, a Mutual Action Plan tab and a details drawer. Storylane added Ask AI over demo analytics, so leads captured, hottest accounts and visitor trends come back as answers rather than dashboard filters.

In enablement, Outdoo AI announced closed loop coaching, where one scorecard grades both practice and real calls and each gap it finds becomes feedback, a short practice session or a refresher, with AI Trainers built from a company’s playbooks and kept current by Knowledge Sync. ActiveCampaign added if steps with multiple branches to its Automation Canvas, routing by field values, tags, list status and segment membership, so one step replaces a stack of conditions.

What two weeks say about the space

Forty four entries, one per vendor, and the theme is the allowance. The assistant got the keys to the record and, in the same release, a limit on what it may spend, what it may read, and where its data may rest.

Nothing in this issue jumped out from behind a door. The scariest thing in GTM software this period was an HTTP 402, and even that came with a reset time. The vendors spent two weeks writing down the rules of the house: who gets in, which fields they may read, how many credits they may burn, which mailbox they may send from, and what happens when they hit the limit.

A limit is only useful to a buyer who knows the price of the unit underneath it. The count below shows how many vendors publish one.

Index Answer

Which GTM vendors publish a price a buyer can budget from?

Fewer than a third. Of the 407 vendors the GTM Tech Index grades on Commercial Transparency, 123 publish real prices, meaning plans, seat or usage economics and the shape of enterprise pricing at a standard a buyer can budget from without a call. That is 30 percent of the index.

Another 152 publish partial pricing, usually real entry tiers with enterprise left opaque, which brings the share with anything checkable to 68 percent. 66 publish a pricing page that describes the structure without a number, and 66 offer a demo request as their entire commercial disclosure.

Among the 35 vendors filed primarily as AI SDR and outbound agents, the products this period’s metering is spreading across, 5 publish real prices: AiSDR, Cognlay, Evergrowth, OptaReach and RingEmAll. Another 16 publish partial pricing, and 7 treat a demo request as the whole answer. Even the published ones meter on different units, contacts researched, emails sent, credits or meetings booked, so two rate cards rarely compare without first asking what the unit is.

This period added more units to the list: a minute of a call, a domain found, a scan that found nobody, an API request at a fee set per route, a weekly allowance of assistant questions before the credits start. Each is a price a finance team can put in a cell. None is a price it can compare to another vendor’s without knowing what one unit of work is.

A vendor that publishes the unit price has said what one piece of work costs. A vendor that also publishes what happens when the agent runs out, a 402 with a reset time, a paused run, a cap per member, has said how the bill stops. This period several did the second, which is newer and, for anyone who has opened a surprise invoice, more useful.

Every vendor graded A, category by category, with the artifact each grade was read from, is on the pricing transparency page.