Most writing about AI and revenue describes content production, which is a cost activity wearing a revenue costume. Money in an operating business comes from four mechanisms only: winning more of what you quote, charging what you intended to charge, keeping customers you already have, and converting sales into cash sooner. Everything below attaches to one of those four.
The four mechanisms
- Conversion — a higher share of quotes becomes orders, usually through speed and follow-through rather than persuasion.
- Price realisation — the margin you designed is the margin you actually get, because leaks are visible.
- Retention and reactivation — existing customers buy again, because someone noticed they had stopped.
- Cash velocity — the same revenue arrives sooner, which funds the business without borrowing.
If a proposed AI project does not attach to one of those four, it is a cost project. That is not a criticism — cost projects are often the right call — but it should be approved on cost grounds, not sold as growth.
The ten workflows, costed
Assumptions: a 40–200 person firm doing roughly KES 9,000,000 a month at 28% gross margin, 60 quotes a month averaging KES 180,000; ~KES 130 = USD 1. Benefit is stated as gross profit, not revenue — a distinction most vendor cases quietly skip. First-year ROI = (12 × net monthly benefit − build cost) ÷ build cost.
| Workflow | Mechanism | Build (KES) | Run / mo | Net benefit / mo | Payback | Year-1 ROI |
|---|---|---|---|---|---|---|
| 1. Price-realisation monitoring | Price | 180,000 | 7,000 | 111,000 | 1.6 mo | 640% |
| 2. Proposal & scope drafting | Conversion | 130,000 | 9,000 | 69,000 | 1.9 mo | 537% |
| 3. Dormant-customer reactivation | Retention | 120,000 | 6,000 | 62,000 | 1.9 mo | 520% |
| 4. Lead follow-up ladder | Conversion | 200,000 | 10,000 | 86,000 | 2.3 mo | 416% |
| 5. Churn early-warning | Retention | 170,000 | 7,000 | 75,000 | 2.3 mo | 429% |
| 6. Cross-sell trigger | Retention | 160,000 | 8,000 | 56,000 | 2.9 mo | 320% |
| 7. Collections acceleration | Cash | 180,000 | 8,000 | 63,000 | 2.9 mo | 320% |
| 8. Tender qualification | Conversion | 240,000 | 14,000 | 80,000 | 3.0 mo | 300% |
| 9. Quote win/loss analysis | Conversion | 150,000 | 5,000 | 47,000 | 3.2 mo | 276% |
| 10. Speed-to-quote | Conversion | 220,000 | 12,000 | 63,000 | 3.5 mo | 244% |
Before you react to the ROI column: these percentages look extraordinary because the denominators are small, not because the returns are magical. A KES 180,000 build returning KES 111,000 a month is a 640% first-year return and also a rounding error against most firms' annual costs. The risk in these projects is never the ROI percentage. It is whether the benefit is real and attributable. The rest of this article is about establishing that.
1. Price-realisation monitoring — the one nobody asks for
Mechanism: price. Every invoice line is checked against its floor price. Unapproved discounts, stale price-list items and drifting product margins are surfaced weekly to a person who can act.
The arithmetic. On KES 9,000,000 monthly revenue, price leakage of 1.4% is a typical finding where discounting is discretionary and unmonitored — that is KES 126,000 of revenue at close to 100% margin impact, since the cost was already incurred. Recovering two-thirds of it gives roughly KES 84,000. Add KES 34,000 a month from catching stale price-list items during input-cost inflation. Gross ≈ KES 118,000; net KES 111,000.
Prerequisite: a recorded floor price per item. If you do not have one, that is the project — and it is a spreadsheet, not an AI project.
2. Proposal and scope drafting — 1.9 months
Mechanism: conversion, via volume and response time. A structured brief becomes a first draft in your house format using your own past language.
The arithmetic. 18 proposals a month; drafting falls from 4 hours to 1.3 hours = 48 hours saved at KES 1,000 blended = KES 48,000. The conversion effect is separate: capacity to respond to 4 additional opportunities a month that were previously declined for lack of time, at a 20% win rate and KES 180,000 average value and 28% margin = 4 × 0.20 × 180,000 × 0.28 = KES 40,000. Apply a 25% haircut for optimism = KES 30,000. Gross ≈ KES 78,000; net KES 69,000.
3. Dormant-customer reactivation — the cheapest revenue you own
Mechanism: retention. Identify customers who bought regularly and stopped, rank by historic gross profit, draft a specific re-approach referencing what they used to buy and when.
The arithmetic. A firm at this scale typically carries 180–260 dormant accounts. Take 200, of which 120 are contactable and creditworthy. A reactivation rate of 6% is a normal result for a specific, non-generic approach = 7 customers. Average annual gross profit per customer KES 168,000 → 7 × 168,000 ÷ 12 = KES 98,000 a month at full run rate. Haircut heavily for the fact that reactivation is front-loaded and decays: take one-third = KES 33,000 sustained, plus KES 35,000 in the first months amortised. Gross ≈ KES 68,000; net KES 62,000.
Prerequisite: one identifier per customer. Duplicate records mean contacting the same person three times, which converts your cheapest revenue into your most embarrassing outreach.
4. Lead follow-up ladder — the largest single item
Mechanism: conversion, via coverage. Every open quote gets a scheduled sequence with drafted, context-aware messages the salesperson approves in one click.
The arithmetic. Coverage rises from roughly 40% of quotes to over 90%. The recovered half converts below the attended half — assume 3 points lower — giving a net uplift near 1.8 points across the book: 60 × 180,000 × 1.8% × 28% = KES 54,400. Add 14 hours of manual chasing at KES 560 = KES 7,800, and KES 34,000 of repeat business from customers not previously re-contacted. Gross ≈ KES 96,000; net KES 86,000.
5. Churn early-warning — 2.3 months
Mechanism: retention. Detect the pattern that precedes loss — order frequency falling, basket narrowing, complaints rising, payment slowing — and route the account to an owner while it is still recoverable.
The arithmetic. 240 active accounts, annual churn 11% = 26 accounts a year. Early warning plus intervention saves a third = 8.7 accounts. At KES 168,000 annual gross profit each → 8.7 × 168,000 ÷ 12 = KES 122,000 a month. That is optimistic; halve it to KES 61,000 and add KES 21,000 from earlier detection of margin decline on at-risk accounts. Gross ≈ KES 82,000; net KES 75,000.
The honest limit: a warning without a named owner and a budget to act changes nothing. Most churn models fail on the intervention, not the prediction.
6. Cross-sell trigger — 2.9 months
Mechanism: retention and basket size. When a customer buys A, flag that similar customers buy B within 60 days, and put that in front of the salesperson at the moment of contact.
The arithmetic. 300 orders a month; a 4% attach rate on a KES 60,000 average add-on line at 28% margin = 12 × 60,000 × 0.28 = KES 202,000. That figure assumes perfect execution; salespeople act on perhaps a third of prompts, giving KES 67,000. Haircut to KES 50,000, add KES 14,000 from reduced stockouts on commonly paired items. Gross ≈ KES 64,000; net KES 56,000.
7. Collections acceleration — cash, not revenue
Mechanism: cash velocity. Stated plainly: this does not increase revenue. It increases the cash you hold against the same revenue, which for most SMEs is the binding constraint.
The arithmetic. DSO 68 → 56 days on KES 9,000,000 monthly revenue releases KES 3,600,000 once. At a 16% overdraft rate that is KES 48,000 a month of avoided finance cost, plus 22 clerk-hours at KES 400 = KES 8,800 and KES 14,000 of reduced write-offs. Gross ≈ KES 71,000; net KES 63,000.
Note the shape of this benefit: the working-capital release is one-off, the interest saving is recurring. Do not model the KES 3.6m as monthly income — a mistake that appears in a remarkable number of vendor business cases.
8–10. Tender qualification, win/loss analysis, speed-to-quote
Tender qualification (3.0 mo). 12 tenders a month, senior review from 3.5 hours to 1 = 30 hours at KES 1,560 = KES 46,800; plus 18 hours of avoided unwinnable bid preparation = KES 28,000; plus KES 19,000 of avoided compliance disqualifications. Net KES 80,000.
Quote win/loss analysis (3.2 mo). Structured reasons for every lost quote, aggregated monthly. The benefit is not the analysis; it is the two or three pricing and scoping corrections it produces each quarter. Valued at a 0.8-point win-rate improvement: 60 × 180,000 × 0.8% × 28% = KES 24,200, plus KES 28,000 from discontinuing pursuit of segments that never convert. Net KES 47,000.
Speed-to-quote (3.5 mo). Turnaround from a day to under two hours. On contested quotes a 2-point win-rate uplift gives roughly KES 60,000 of gross profit, plus KES 15,000 of estimator time. Net KES 63,000. Slowest payback on the list and still a 244% first-year return — which tells you how low the bar actually is.
The rules that keep these numbers honest
- Name the account line. Every claimed benefit must land somewhere a bookkeeper can point at. "Improved efficiency" is not a line.
- Gross profit, never revenue. A 2-point win-rate uplift on KES 10.8m of quotes is KES 216,000 of revenue and KES 60,000 of profit. Only one of those is yours.
- Halve it, then approve it. If it survives a 50% haircut it is a real project. If it only works at full assumptions, it is a proposal, not a plan.
- Hold a control where you can. One region, one product line, one sales team. Without it you cannot separate your automation from your market.
- Re-measure at 90 days against a baseline you recorded first. The baseline is the part everyone forgets, and without it the debate becomes a matter of opinion — which the vendor will win.
What to do first
Run price-realisation monitoring. It has the shortest payback on this list, it requires no change in customer behaviour, it is invisible to your competitors, and the leak it finds is almost always larger than the finance team expects. Fund workflow two from what it recovers.
