Most companies we work with have a high degree of AI adoption. But there’s enormous variation between teams and individual employees. Many have settled into a few use cases, and it’s useful to provide contained learning sessions to help them improve on the familiar and engage with the possible. Here are some light-touch examples that, mostly, would fit into a discrete lunch-and-learn time slot.
Sales is the function where AI output meets a human with an incentive to detect it and discount what it says. We’re all looking for a good reason to say no.
It’s also the function with the most sensitive relational data. Deal terms, competitive intelligence, and named-individual notes about how a buyer’s internal politics work, which is highly valuable, but would be embarrassing if it surfaced.
The data-tier rule. Train the team to understand AI’s relationship with information by having them sort it into three buckets. Public research (a prospect’s published accounts, their website, news coverage) is safe anywhere. Internal commercial data (pipeline values, discount history) belongs in enterprise tools only. Relational notes (what a rep wrote about a specific person’s behaviour or organisational politics) generally don’t belong.
The buyer-detection exercise. Have reps run AI-drafted outreach past a colleague playing the recipient, who flags anything that reads as machine-generated. The feedback is immediate and slightly humiliating, which can produce behaviour change fast.
For a bit of context, Gartner surveyed 646 B2B buyers (published March 2026) and found 67% prefer a rep-free buying experience and 45% used AI during a recent purchase. The buyer is running their own AI-assisted evaluation. Training that only increases outbound volume is optimising for a model the buyer is already killing.
Marketing is where AI is most immediately capable and most likely to produce measurable damage. Generation is easy, but consistency and originality aren’t. Marketers are always on the hook for more content than they can reasonably produce, so AI is the siren’s song.
A deeper problem is that most mid-market marketing teams have never explicitly defined their voice. It usually lives in one or two people’s heads, and it’s fine so long as they read everything. AI breaks that model by pushing volume.
You have to understand your ‘voice’ before AI can. Have the team independently score the same AI draft against criteria they’ve agreed on. The scoring disagreements are the training content, and they’ll reveal that “our voice” wasn’t fully shared knowledge.
Build a context block. That’s a standard chunk of background, who we are, who we sell to, how we sound, what we never claim, prepended to every prompt so output stays consistent regardless of who’s writing. It’s reusable, it encodes the voice work, and it means a junior hire produces on-brand first drafts.
And for what it’s worth: if you are tired of AI-drafted content on LinkedIn or in your inbox, so is everyone else. AI drafts reproduce common industry phrasing, which creates a differentiation problem and occasionally a legal one. If you’re going to use AI to 3x your content or more, you still have to have humans top and tail and title new pieces, or AI-wary readers will shut off.
Customer Support is the function under the strongest pressure for volume metrics, which is exactly the pressure that erodes review discipline. Human agents gradually approve drafts without reading them because the queue is long and the drafts are usually fine.
Run an escalation-boundary workshop. Give human agents 30 real tickets and have them sort into auto-draft, draft-with-review, and human-only. Where they disagree is where your policy needs to be explicit.
Hard-carve the exclusions. Complaints, vulnerable customers, safety issues, and anything with legal content bypass AI drafting entirely. Enforce these in the system where you can, not in policy alone, because policy erodes under queue pressure. Involve the team in setting these, and you’ll learn something about both the issues and the team.
Next week we’ll look at similar activities for finance, HR, and operations.
Richard
Andy Haley is a self-described recovering enterprise architect, and now CEO of Sullivan and Stanley. He argues that AI should never be looked at in...
Watch video
"We don't necessarily talk about AI in isolation. In fact, we never talk about AI in isolation," says Andy Haley,...
Read more
Most companies we work with have a high degree of AI adoption. But there's enormous variation between teams and...
Read moreGet ahead with the most actionable insights, playbooks and real-world AI use cases you can adopt right now, in your inbox every week
