The hype curve and the boring middle where money is

For a 5–50 person service firm, the money in AI is not at the frontier and not in refusing to play — it is in the boring middle: transcription, enrichment, drafting, routing, summarising. These uses are two or three years past their demo moment, which is exactly why they work: stable tooling, known costs, measurable output. The frontier is where the conference talks are; the middle is where the payback periods are.

What does the hype curve actually describe?

Gartner's hype cycle is the famous version: a new technology spikes to a "peak of inflated expectations", crashes into a "trough of disillusionment", then climbs a "slope of enlightenment" to a "plateau of productivity". You can argue with the methodology — plenty do — but as a description of how attention and utility come apart in time, it matches what I see daily building these systems, and it is the backdrop to everything in AI Automation for B2B: what actually works.

The practical reading: attention peaks years before utility does, and utility arrives quietly, after the press has moved on. Whatever is loudest right now — agents that run your whole company, this quarter's autonomous everything — is, almost by definition, at the point of maximum gap between promise and deliverable. Meanwhile the things that stopped being exciting — speech-to-text, document drafting, data extraction — have crossed into being infrastructure.

Why do small firms keep buying at the peak?

Three mechanisms, none of them stupidity. Fear of missing out is engineered: vendors sell urgency ("your competitors are already doing this") because urgency closes deals. Demos are unrepresentative: a frontier capability demos brilliantly on curated inputs and fails on your actual, messy data — and you cannot tell the difference from the demo. And the peak is when budget is easiest to authorise, because the board has read about it too.

The result is a predictable pattern I meet in audits: a five-figure spend on something impressive that never reached production, sitting alongside a team still manually retyping meeting notes into the CRM — a task the boring middle solved years ago for pennies.

What actually lives in the boring middle?

The workhorse list is unglamorous and consistent:

• Transcription and summarising — calls become notes, decisions, and CRM entries without anyone typing.
• Enrichment and research — filling firmographic and contact fields at a fraction of manual cost.
• Drafting under supervision — first drafts of emails, proposals, and content, with a human holding the pen last.
• Extraction and classification — pulling structure out of documents, invoices, enquiries; routing them to the right place.
• Matching and deduplication — the data hygiene nobody volunteers for.

Nothing on that list will trend anywhere. Everything on it has stable APIs, per-unit costs you can look up, and output you can check — which is precisely what makes the arithmetic in Calculating automation ROI before you build come out positive.

How do you time adoption deliberately?

With rules rather than mood. The mechanism I use: when a capability first appears, then it goes on a watchlist, not a purchase order. When the tooling stabilises — meaning pricing is published, APIs stop churning monthly, and practitioners you trust report production use — then it qualifies for a bounded pilot: one process, one month, one metric agreed in advance. When the pilot clears the ROI threshold, then it gets built properly, with monitoring and an owner. When it misses, then it returns to the watchlist with a revisit date, because costs fall and capabilities rise — a "no" in the trough is often a "yes" two years later.

Deliberate second-mover is the correct posture for a small firm. You carry none of the frontier's breakage, and the middle is deep enough that you will not run out of profitable builds for years.

What keeps you honest on the way?

Two disciplines. First, keep a human in the approval path anywhere output touches a client or a ledger — where that checkpoint belongs, and where it can safely come out, is the subject of Human-in-the-loop: where approval belongs. Second, contain the evaluation habit. I run tool evaluation as a scheduled, bounded slot inside my weekly structure — the shape of which is in How I run my week — because unscheduled fascination with new tools is how operators end up on the peak with everyone else.

The boring middle does not photograph well. It merely compounds — which, for a firm your size, is the entire point.

---

Next step: the Growth System Audit — £450, seven days, credited against any build — maps where your growth system leaks and what to build first.