Your AI problem is political, not technical

Last week’s growthLIVE roundtables produced some of the most candid thinking we’ve heard from mid-market leaders on AI adoption. Sessions ran under Chatham House Rule, so observations are anonymised.

The data problem is more often political than technical

Across the roundtables, leaders described legacy databases in a state of advanced decay. Marketing teams working from client lists with an 18-month decay rate, resulting in records that are nearly obsolete almost before a campaign launches. One participant cited a database still carrying contacts who would be 130 years old.

Just as damaging is the power structure around the data. Centralised revenue operations teams trying to consolidate data are running into active resistance from departments unwilling to surrender administrative control. Some technical teams have found workarounds, using open-source frameworks like Model Context Protocol to connect isolated systems directly to AI engines without API coding. Board and IT sign-off is still required before anything connects.

One marketing leader described her engineers building repositories and AI workflows freely. Her own team manually re-uploaded PDFs every time something changed on the front end, because they hadn’t been trained on developer tools. She didn’t dress it up: her commercial team were dinosaurs, watching their engineers drive flying cars.

Tracking workers doesn’t build the trust AI needs to function

Reps welcomed the tool when it helped them prioritise their own workflows, flagging intent signals and surfacing which deals deserved attention. What they resisted was the pipeline review, where the same data stripped away their ability to manage upwards. One participant described hooking Avoma call recordings into HubSpot and feeding the data through to Claude, scoring deal health automatically. Salespeople love a spin, and the AI removes it.

In a separate session, when an organisation switched on call recording for performance monitoring, staff didn’t raise a formal objection. They told management directly: ‘it’s like you don’t trust us.’

Demonstrating that the tool could draft follow-up emails, surface a prospect’s personal details ahead of a call, or flag a contact’s recent holiday was enough. Once staff saw personal value, adoption followed, though it took nine months.

Synthetic seniority is hollowing out the talent pipeline

Several participants brought up synthetic seniority: junior people using AI to produce work that reads as authoritative without the judgement to back it up. One described interviewing what appeared to be a highly experienced candidate, only to discover post-hire that the person ‘has no clue.’

A second leader described junior colleagues submitting AI-generated copy because it reads well, without the baseline to interrogate whether it was accurate or on-brand. When leadership pushed back, they couldn’t defend it. They hadn’t built the logic themselves.

Entry-level roles have traditionally been where professional judgement forms. You do foundational work, make cheap mistakes, build the instincts that underpin harder decisions later. If those tasks are automated before the instincts develop, the next generation of senior practitioners is being built on a hollowed-out base.

As the foundational work gets automated and mechanical execution becomes commoditised, the premium shifts back towards critical thinking and judgement. Psychology, philosophy, the things a prompt can’t replicate.

The tools are multiplying faster than the governance

One VP of Marketing wanted to burn her database down and start over. Her team had been given generative AI tools without guardrails and used them for documentation and briefs. The AI began self-referencing those outputs, and the result, in her words, was ‘an exponential growth in bullshit’ inside their own systems, internal hallucinations compounding until the database was closer to noise than record.

Across the sessions, organisations were dealing with the same problem: shadow AI. Employees reach for whatever tool works, often in parallel, often without oversight, and the outputs land in shared systems.

Where licensed tools are seen as inferior to whatever’s freely available, employees bypass IT governance entirely. One leader solved this by locking the entire organisation into a single approved tool, giving her visibility over every prompt and letting her correct bad behaviour before it compounds. A recently merged business tried the opposite, banning all unapproved tools, only to find the entire workforce using them anyway through workarounds. One CRO took a different route: his technical team had been building internal agents, but their job was building for clients and internal tooling pulled focus from that. They moved to buying off the shelf.

The organisations spending most on outreach are the least likely to respond to it

Several executives described the outreach infrastructure they’d built. Automated SDR voice diallers. WhatsApp pre-qualification flows processing thousands of leads a month. One had built internal tooling to automate push notifications, transactional emails, and market commentary at scale.

Asked about incoming automated outreach, the same executives gave the same answer: they ignore all of it. One B2B partner described receiving emails that were relevant, well-scripted, and correctly targeted. He still didn’t engage. The volume had made it meaningless. Customisation used to signal effort. AI has driven that cost to zero. As he put it, the sender has to ‘mean what you’re saying.’

Across the mid-market, aibl sees this consistently: the outbound investment is real, but the return is harder to find. The ones getting traction have stopped treating it as a volume problem. The differentiator is judgement.