Program report · filed November 17, 2026

AI in SalesMeasured report

SnapLogic's Agent Burned 30 Minutes on a Prospect Who Left 18 Months Ago

SnapLogic, OpenAI and Cognism panellists in London agreed: agents execute workflows, but data quality decides the outcome. Bad records scale at machine speed.

By Nathan Brooks4 min read833 words

Program notes

  1. A SnapLogic agent spent 30 minutes on a campaign targeting a prospect who had left the company 18 months earlier.
  2. Cognism's framework names four requirements for agent-ready data: access, context, accuracy and control.
  3. Every OpenAI account now has its own AI agent fed by a semantic layer shared with forecasting.
  4. SnapLogic needed a year or more to gain a few percentage points of accuracy on early knowledge agents before MCP simplified system connections.
  5. Every earlier version of OpenAI's 'GTM brain' has been retired, replaced through continuous experimentation.

An AI agent ran for 30 minutes, built a campaign, drafted messaging — then the team discovered the target had left the business 18 months earlier. That story, from Jocelyn Arcega, Head of Solutions Engineering at SnapLogic, anchored a panel called Data Before Agents, hosted by Cognism at The Shard in London. He shared the stage with Orian English, GTM Leader at OpenAI, and Viktoria Ruubel, CPTO at Cognism.

The theme was blunt. Agents execute; data decides. Ruubel put it in one line: "You have a team of agents. You're just multiplying the noise. You are not getting to the outcome you want to get to."

What is agentic GTM?

In agentic go-to-market, AI agents handle multi-step work: account research, outreach, CRM updates. People set direction and review output. A chatbot answers. An agent acts.

Arcega traced how fast this moved. Two years ago his team built knowledge agents that answered questions and often failed; a year of work bought a few percentage points of accuracy. Then the Model Context Protocol (MCP) arrived, and agents could connect directly to the CRM and other systems. Progress accelerated. The builders changed too — salespeople, SDRs and customer success managers now build their own agents, because sales work is too varied for fixed workflows.

His benchmark example: a rep with 30 minutes before a customer meeting. The agent pulls interaction history and public information, then builds a presentation in the account's branding. Unthinkable two years ago.

What does it look like at OpenAI?

Every account has its own agent. English's GTM innovation team built a semantic layer connecting the CRM, data sources and meeting transcripts — the same layer feeds forecasting, so everyone works from one source of truth. Reps get proactive nudges via Slack or Dots, OpenAI's always-on agent, which might flag something from yesterday's call and ask whether to escalate, update the CRM, or book a follow-up. Leaders see deal status without chasing anyone.

The tool evolved from a Q&A chatbot in Slack to a proactive system, and leaders now vibe-code dashboards reading the same data RevOps and Finance use. English calls the setup the "GTM brain" — informal, not official. The team chose the account as the unit each agent is built around; an organisation-level agent and always-on individual agents come next.

Why does data quality matter more now?

Because agents don't hesitate. A rep with a bad list might notice something wrong and stop. An agent carries on, then repeats the mistake for the next 500 accounts. Arcega's fix: verify CRM records are current, cross-check what the agent finds online, and confirm the right profile — "mission critical" in his words.

Ruubel raised compliance. A swarm of agents working from non-compliant data multiplies the error by every customer reached, including people on Do Not Call lists. Outdated data wastes effort; uncontrolled data creates risk that is harder to undo.

Her four-part framework for AI-ready data:

  • Access — a route to the CRM, data sources and transcripts; without it, agents guess.
  • Context — definitions of what each field means; without it, "you can ask a question of the data and get completely different answers depending on the context," as English said.
  • Accuracy — continuous verification; a quarterly clean-up cannot keep pace with an agent acting hourly.
  • Control — rules on who and what can use which data. OpenAI uses role-based access; governance must also cover what agents are allowed to do.

When should agents act?

On signals, not speed. Arcega called smart GTM "about being precise more than having a quantity of activities." A filtered list can still hold 500 people; an agent works through them faster than a rep, but only a signal — a hiring push, a company event, a job move — produces the outcome. Most teams will end up buying signal data rather than scraping job adverts and LinkedIn themselves.

Ownership is shifting. Business users build; RevOps becomes, in Ruubel's phrase, "orchestrators of all the agents that have been built within the company"; IT builds connectors. English agreed with a caveat — OpenAI still depends on RevOps to set metrics and systems of record, or every team builds something that answers differently.

Where should teams start?

The panel split. Arcega argued for picking high-impact use cases tied to KPIs like pipeline. English took the opposite line: "The most important thing is just to get started." Every earlier version of OpenAI's GTM brain is retired, and she doubts the current one would exist without those discarded steps. Her advice: budget tokens for experimentation and push daily use. Ruubel said both — name an outcome, measure spend against return, and keep room to experiment as models improve.

Twelve months out, all three predicted the same thing: strong teams will talk about outcomes and data quality, not agent counts, and nobody will chase a rep for a CRM update. Every prediction rests on one condition — the data underneath stays accurate.

via Cognism Blog (Source)

Filed under

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering companies, deals and strategy in Sales technique and methodology, sales enablement at Quota Brief.

141 articles

Also rated

  1. Aurasell's Prospecting Agent Mines Your CRM, Not Vendor Lists
  2. Xactly and ServiceNow Connect AI Agents for Revenue Operations
  3. ThoughtSpot Puts AI Agents at the Center of Revenue Operations
  4. SaaStr's AI Agents Built Their Own Channel — and a Safe Word
  5. AI Agents Can Become Large Liability Models Without Controls

« Previous articleNext article »