Sales organisations report spending roughly 23% of the working week on actual selling, with the rest absorbed by admin, research and internal coordination — the exact overhead AI is best at compressing. That's the honest starting frame: the win isn't a smarter demo, it's giving yourself back the hours a demo used to cost to prepare.

The four pillars, and where AI actually helps

PreSales Collective's framing of the SE role splits cleanly into four areas, and they don't all take AI the same way:

Start with the admin layer, not the demo

The instinct is to reach for AI on the most visible, highest‑stakes work — the demo, the executive deck. That's backwards. The admin layer is where the risk of a bad output is lowest and the time saved is most immediate:

• Meeting notetakers (Gong, Fireflies, Otter and similar) transcribe and summarise a discovery call, surfacing objections and next steps — feed the output straight into the four‑column capture grid in Module 08 rather than re‑listening to the recording
• CRM hygiene — drafting the MEDDPICC scorecard update from a call transcript instead of reconstructing it from memory an hour later, when the sharpest detail has already faded
• First‑pass drafting — a follow‑up email, a one‑page recap, a battlecard refresh — always reviewed, never sent unread. The rule is the same one Module 12 already states for RFPs: a first pass, not a final answer

Discovery & account research

This is the sharpest end of the toolkit, and also the one with the most exposure if it's done carelessly. Before turning any AI research tool loose on a real prospect, borrow the discipline a well‑built one already enforces on itself: every fact labelled verified, source‑listed, inferred, or not found — never blended; no claimed access to a system that wasn't actually queried; no fabricated source link; nothing pushed from a private CRM record into a public search query; and an honest three‑line brief preferred over a confident full page built on guesses. That standard isn't aspirational — it's exactly what a rigorous prompt or skill for this purpose has to refuse to do, including resisting instructions smuggled in through a scraped web page or a pasted CRM note. Judge any research tool, AI or human intern, against that bar before you trust its output in front of a customer.

The demo

Two real capabilities, distinct from each other: personalised micro‑demos assembled from a template library against a specific buyer's stated priorities (useful for complex, many‑integration products where hand‑building every variant doesn't scale), and environment tailoring — pulling a prospect's own data into a demo instance rather than presenting a canned dataset, which Module 10's glossary already flags as instantly obvious to a technical buyer when it's missing. Neither replaces the situation‑slide discipline in Module 09; both just make it cheaper to execute well on the fourth deal of the week, not only the first.

RFPs, proposals and battlecards

Module 12 already sets the rule for AI on a bid response — a first pass against the compliance matrix, never an asserted capability nobody verified. The same rule extends to competitive battlecards (Template 7): AI is well suited to watching for a competitor's public changes — a pricing page update, a new feature announcement, a review site pattern — and flagging that a battlecard is stale. It is not suited to writing the "why we win" section unsupervised, because that section is only as honest as someone's actual win/loss judgement, which is exactly the part this manual keeps insisting stays human.

What doesn't change as the tools do

The category map, dated on purpose

Naming specific products in a manual like this one goes stale within a year — this table is a snapshot of September 2026, included as a map of the categories rather than a recommendation of any one vendor in them: