AI · 2026-08-04 · 11 min read · by Juan Carlos Zuloaga
The question is rarely "can AI do this?" It is "is this worth automating, and will anyone use it?"
Below are twelve use cases we see pay back inside a quarter for companies between 50 and 500 people. Each one has a precondition. Where the precondition is missing, the use case fails — not because the technology is weak, but because the input is.
1. Inbound triage. Classify, tag, and route incoming messages. *Precondition:* a defined set of categories that map to real owners.
2. Draft-first replies. Generate a first draft for common requests, always reviewed by a person. *Precondition:* a body of past good replies to learn the tone from.
3. Quote preparation. Assemble a quote from structured inputs. *Precondition:* pricing rules written down rather than negotiated case by case.
The pattern: AI is strongest where a human still presses send.
4. Scheduling assistance. Match demand to capacity across sites, staff, or rooms. *Precondition:* accurate availability data.
5. Exception detection. Flag the 3% of jobs, invoices, or orders that look unusual. *Precondition:* enough history to define normal.
6. Checklist and report generation. Turn field notes into structured records. *Precondition:* someone actually capturing the notes.
In service businesses, exception detection is usually the highest-value item on this list and the least requested.
7. Lead enrichment. Fill in firmographic context before the first call.
8. Content production at volume. Briefs, drafts, and variants — with a human editor as the quality gate. This is how a small team maintains a content cluster without hiring.
9. Personalised follow-up sequences. Sequences that reference what the person actually did, not just their segment.
The honest caveat: unedited AI content ranks and converts poorly. The gain is in speed to first draft, not in removing the writer.
10. Recurring report assembly. The weekly numbers, written up, with the anomalies called out. *Precondition:* a single trusted data source.
11. Document extraction. Invoices, contracts, and forms into structured fields. *Precondition:* a review step, because extraction is 95% accurate and finance needs 100%.
12. Scenario summaries. Turning a model's output into language the board can act on.
For each candidate, score 1–5 on: volume, time saved per occurrence, data readiness, and ownership clarity. Multiply. Start at the top.
Data readiness and ownership carry more weight than they feel like they should. A high-volume process with messy data and no owner will consume a quarter and produce nothing.
This scoring exercise takes an afternoon and prevents the most expensive mistake in the category.
- Internal chatbots over documentation nobody maintained.
- Automations built for a process that was about to change.
- Tools rolled out without naming an owner.
- Anything that removes a human check from a customer-facing message.
The common thread: the failure is organisational, not technical, in every single case.
Run the scoring exercise with your operations lead. Pick one. Give it an owner and a date.
If you want to see how this plays out over a full engagement, our case studies walk through the sequencing for a cleaning company, a dance studio, and a hospitality group: https://inspiralgrowth.com/case-studies