Every list of AI use cases is written as though the prerequisites do not exist. They do, and they decide the outcome. Each entry below states what must already be true before the automation will work — because that, not the model, is what separates the ones that pay from the ones that get quietly switched off.
How to read this list
Three groups, in the order most SMEs should attack them: revenue first (the fastest measurable return), then cost and time, then control and risk. Every entry carries a prerequisite — if you cannot satisfy it, skip the entry rather than force it. Time savings assume a 40–200 person firm and are deliberately conservative.
Group A — Revenue: get to the customer faster and more often
1. Quotation drafting from an enquiry
An emailed or WhatsApp enquiry becomes a draft quote against your price book, ready for a human to check and send.
Prerequisite: a current price book in one file, with unit prices your sales team actually uses. Realistic gain: 60–75% off time-to-first-draft; quote turnaround from a day to under two hours. Fails when: pricing is negotiated per customer with no recorded logic — the model cannot infer what was never written down.
2. Lead follow-up ladders
Every quote or enquiry gets a scheduled sequence — day 2, day 5, day 12, day 30 — with a context-aware message drafted for the salesperson to approve.
Prerequisite: quotes recorded in one place with a date and a status. Realistic gain: follow-up coverage from roughly 40% of quotes to over 90%. This is usually the single largest revenue item on the list. Fails when: messages send without review and read as spam.
3. Dormant-customer reactivation
Identify customers who bought regularly and then stopped, rank by historic value, draft a specific re-approach referencing what they used to buy.
Prerequisite: 18+ months of sales history with a reliable customer identifier. Realistic gain: reactivating 5–8% of a dormant list is a normal result and costs almost nothing. Fails when: "Acme Ltd", "ACME LIMITED" and "Acme (Nairobi)" are three records — you will contact the same person three times.
4. RFQ and tender qualification
Read an incoming tender or RFQ, extract scope, deadlines, mandatory documents and disqualifying clauses, and produce a go/no-go summary.
Prerequisite: someone who can define your qualification criteria in writing. Realistic gain: two to four hours of senior time per document, and fewer bids submitted for work you could never have won. Fails when: treated as the decision rather than the brief for the decision.
5. Proposal and scope-of-work drafting
A structured brief becomes a first-draft proposal in your house format, with your standard terms and previously used scope language.
Prerequisite: three or four good past proposals to serve as the pattern. Realistic gain: 50–70% off drafting time. Fails when: nobody edits it — clients recognise generic prose immediately, and it costs you credibility, not just the deal.
Group B — Cost and time: remove the queue, not the person
6. Supplier-invoice and delivery-note extraction
Scanned or emailed documents become structured line items matched against the purchase order.
Prerequisite: purchase orders that exist as data, not just as printed paper. Realistic gain: 85–95% straight-through processing with an exception queue. Fails when: you design the happy path and leave the exception queue for later. Design the exception path first.
7. Accounts-receivable follow-up
An ageing report drives a reminder ladder, escalating in tone and routing disputes to a person.
Prerequisite: an ageing report you trust and a named owner for disputes. Realistic gain: 8–15 days off DSO in the first quarter, which is a working-capital release, not just saved time. Fails when: reminders go to customers whose invoices are already disputed — one wrong chase costs more goodwill than ten right ones earn.
8. Customer-complaint triage
Incoming complaints are classified, given severity, routed to the right owner, and acknowledged immediately.
Prerequisite: categories your own staff agree on. Realistic gain: acknowledgement inside minutes rather than a day; the queue disappears, the role does not. Fails when: your categories are ambiguous internally — a model will not resolve what humans dispute.
9. Management report generation
Scheduled operational and financial reports built from the source tables, with a written commentary on what changed and why.
Prerequisite: one agreed definition per metric. Realistic gain: two to four days of month-end effort, and reports that arrive on day 3 instead of day 15. Fails when: two departments define "revenue" differently and the report is disbelieved by both.
10. Meeting actions and task tracking
A recorded or transcribed meeting produces a decision log and an owner-dated action list pushed into your task tool.
Prerequisite: the willingness to record meetings, and a consistent place for tasks to land. Realistic gain: the cheapest item here, and often the most visible to leadership. Fails when: actions land somewhere nobody looks.
11. Internal knowledge answering
Staff query your own material — policies, specifications, maintenance manuals, historic job records, contract terms — and get sourced answers.
Prerequisite: documentation that is actually good. Realistic gain: real, where the corpus is strong. Fails when: the documentation is thin. This pattern amplifies a knowledge base; it does not create one.
Group C — Control and risk: catch what people miss
12. Price and margin-leak detection
Flag every invoice line sold below its floor price, every unapproved discount, every product whose realised margin has drifted.
Prerequisite: a recorded floor price per item. Realistic gain: usually the highest-value item in this group and the least glamorous. Margin leaks in SMEs are quiet and continuous. Fails when: there is no agreed floor price to compare against.
13. Inventory exception monitoring
Rather than another stock report, surface only the exceptions: negative balances, items below reorder, slow-movers tying up cash, counts that drifted from the system.
Prerequisite: stock movements recorded at the time they happen. Realistic gain: fewer stockouts and a visible reduction in dead stock within two quarters. Fails when: issues are recorded weekly from paper — the exceptions are then just noise.
14. Duplicate and anomaly detection in payments
Flag duplicate supplier invoices, near-identical payment requests, and amounts materially outside a supplier's normal range.
Prerequisite: supplier records with consistent identifiers. Realistic gain: small in frequency, large in individual value. Fails when: flags go to someone with no authority to stop a payment.
15. Compliance and document-expiry tracking
Track licences, certificates, insurance, contracts and statutory filings, with escalating reminders before expiry.
Prerequisite: a list of what you hold and when each expires — assembling it is most of the work. Realistic gain: avoided penalties and avoided disqualification from tenders. Fails when: reminders route to an individual instead of a role, and that individual leaves.
Where to start
Do not start with the most impressive. Start with the one where all three of these are true: it runs at least 20 times a month, you can already see the data it needs, and you can check its output cheaply. In most SMEs that is lead follow-up (largest revenue effect), quotation drafting (most visible to staff) or meeting actions (cheapest, fastest credibility).
Ship one. Measure it against the baseline you recorded beforehand. Then use that number to fund the next one. Firms that run this loop three times end the year with automations that are used; firms that buy a platform end the year with a licence.
