Most AI pilots in GTM stall because of the operating model, not the model. Agents amplify what they are fed: incomplete CRM data produces bad recommendations, brittle workflows break, and tools that sit outside a rep’s day get ignored. The pilots that scale pick one narrow workflow, run on clean data, live inside the CRM, Slack or inbox, learn from corrections, and move one board-level metric within 30 days.
Executives are investing
PwC’s AI agent survey of 300 senior executives, run in May 2025, found that:
- 88% said their team or function plans to increase AI-related budgets in the next 12 months because of agentic AI.
- Of those adopting AI agents, 66% reported increased productivity.
- 46% were concerned their company may be falling behind competitors in adopting AI agents.
- 28% ranked lack of trust in AI agents as a top-three challenge.
So leaders want results now, and early adopters are seeing them. The problem is getting from a promising pilot to something the whole team uses.
But most pilots stall
The MIT NANDA report, The GenAI Divide: State of AI in Business 2025, reviewed more than 300 public AI initiatives, interviewed 52 organisations and surveyed 153 senior leaders between January and June 2025. Its conclusion is blunt: lots of pilots, big budgets and few scaled wins. On enterprise-grade AI tools, the report says:
Most fail due to brittle workflows, lack of contextual learning, and misalignment with day-to-day operations.
- Brittle workflows: the automation follows a recipe that breaks when one ingredient is missing.
- No contextual learning: the system does not learn from outcomes, so it repeats its mistakes.
- Misalignment with daily work: the tool does not match how people actually work, so they ignore it.
All roads lead to data
Clari Labs made the same point from the revenue side. Its research found that 78% of enterprises are still in the early stages of AI adoption, and 67% do not trust the revenue data that AI depends on. Agents are not magic. In GTM, bad data looks like this:
- Incomplete or inconsistent CRM fields, so the agent cannot work out the deal stage or next step.
- Duplicates and poor enrichment, so it recommends outreach to the wrong person.
- No call transcript coverage, so meeting prep and sentiment have nothing to work from.
- Stale activity and signals, so it suggests next steps that are already out of date.
That is why we fix the CRM and data foundation before building agents on top.
What the teams that scale do differently
- Start narrow. One tight, high-value workflow solved end to end before expanding.
- Embed in real work. The output appears where decisions are made: the CRM, Slack or the inbox, not a separate app.
- Design for learning. Keep context, capture corrections and feed outcomes back in.
- Fit the workflow, not the interface. Knowing the sales motion beats a polished screen.
- Measure early and visibly. Move a metric the board cares about, such as meetings, pipeline speed or acquisition cost, quickly.
Three GTM pilots that scale
Pick one play, one ICP, one owner and one KPI, and run it for 30 days.
Pilot A: Enrich and qualify
How: automatic enrichment, then ICP scoring. Only high-confidence leads go to reps. Low-confidence leads go to an SDR to validate.
KPI: share of enriched leads that become qualified against your baseline, and time to qualify.
Pilot B: Meeting prep and win plan
How: a one-page brief delivered to Slack or the CRM before each call, with three signals, a one-line opener, two likely objections with suggested responses and a recommended next step.
KPI: lift in meeting-to-opportunity conversion.
Pilot C: Signal-driven outbound
How: agents watch funding, hires, product launches and intent signals, score accounts, draft a one-sentence opener and run a light three-touch sequence. The AE approves the first handoffs.
KPI: meetings booked per outreach and touches per booked meeting.
This is close to what we built for Boundless: Clay ICP scoring, Slack alerts and automated briefs on top of rebuilt sending. Outbound replies went from 0 to 148 in six weeks, with 2 hours back per AE per day. Read the case study.
Do not forget the people
PwC’s respondents ranked organisational change and employee adoption among the real barriers. Even a good pilot stalls if reps do not know how to use it on their own accounts, which is why we pair builds with hands-on AI training for sales teams.