Speciality coffee roasted in Cornwall and sold online, by subscription and to cafés. 28 staff.
Verdict: 7 flows (2 running; 4 new; 1 to stop). Those kept running bring £2,312 a year of truly extra gross profit, but £1,617 a year of discount goes to people who would have ordered anyway. The new flows should bring about 372 extra orders and £4,358 a year, most from “Abandoned basket”; stopping the loss-making ones saves £383. After £960 a year to look after them, and 22 hours of set-up (£880), the first-year net is £1,958 (payback 5 months), or £250 if the new flows bring half as much. 1 flow needs data that isn’t captured yet.
| Flow | Status | Extra orders | Gross profit | Discount given away | Sending | Net a year |
|---|---|---|---|---|---|---|
| Welcome series for new sign-ups Email · 9,600 a year · 6 orders per 100 (platform figures) · 10% off | Running | 288 | £2,448 | −£979 | – | £1,469 |
| Win-back after 120 days without an order Email · 3,100 a year · 4 orders per 100 (held-back test) · 15% off | Running | 124 | £843 | – | – | £843 |
| Birthday discount Email · 6,000 a year · 2.5 orders per 100 (platform figures) · 25% off | Stop | 75 | £255 | −£638 | – | −£383 |
| Abandoned basket Email · 7,200 a year · 8 orders per 100 (estimate, half counted) · 70% truly extra | New | 202 | £2,399 | – | – | £2,399 |
| Running-low reminder 25 days after a 250 g bag Email · 5,200 a year · 14 orders per 100 (estimate, half counted) · 30% truly extra · data not captured yet | New | 109 | £1,299 | – | – | £1,299 |
| Brewing guide and review request after a first order Email · 5,200 a year · 3 orders per 100 (estimate, half counted) | New | 39 | £464 | – | – | £464 |
| Abandoned basket reminder by text Text · 1,800 a year · 5 orders per 100 (estimate, half counted) | New | 22 | £268 | – | −£72 | £196 |
The win-back flow was tested against a held-back group, so all its orders count as extra and none of its discount is give-away. The birthday discount goes mostly to people who would have ordered anyway: it costs more than it brings. The abandoned-basket email is credited with 70% truly extra orders (people who left a basket mostly don’t come back unprompted); the running-low reminder only 30%, because many of those customers would reorder anyway or are moving to a subscription. It also needs bag size tagged on each order before it can run. Margin is after postage, packaging and card fees.
People entering each flow and its orders come from the business’s platforms; new flows’ order rates are estimates, of which 50% is counted. Only 50% of a flow’s orders are treated as truly extra unless it was measured against a held-back group or given its own share. Discounts cost margin on every flow order, and are pure give-away on orders that would have happened anyway. Gross profit at 35% margin; texts at £0.04 each; new flows earn nothing for the 2 months it takes to build them (4 where data must be captured first); looking after them takes 2 hours a month. Customers’ later orders, and the cost of people unsubscribing, are not counted.
Hold back 10% of the people entering each new flow for its first two months: that turns estimates into measurements, and shows whether the welcome discount is needed.
Fictional, continuing the example. Tidemoor chose the guided route and, in the final session, set up an agent in an AI assistant the business already uses, connected read-only to the email platform's and the text provider's reports (its IT confirmed this before it committed). On the first working day of each month it reads every flow's month (people entering, messages, orders, and the held-back group where there is one), values it with the plan's own formulas, and sends the managing director a short note. It flags a flow whose orders per 100 fall 30% or more below the plan, one that turns loss-making, a flow marked Stop that is still sending, and any text flow without separate consent; months with fewer than 100 people entering a flow are listed, not judged. It never edits, pauses or sends a flow. Nothing left the business, and no decision changed, without one of its own people approving it.
| Date | What the agent flagged | What was decided |
|---|---|---|
| 13 Oct Flag | Test run at set-up on September: the birthday discount, marked Stop in the plan, was still sending (500 people, 13 orders at 25% off). | Switch it off by the end of October, once the birthdays already promised the code had it. |
| 2 Nov No flag | The birthday discount stopped on 27 October. | Nothing needed. |
| 1 Dec No flag | The abandoned basket email went live on 9 November: 8.3 orders per 100 people against the 8 planned. | Nothing needed; a held-back group of 10% runs with it. |
| 1 Dec Flag | The welcome series fell to 3.1 orders per 100 in November, against 6 in the plan. | The ecommerce manager found a shop link broken by a template change and fixed it on 2 December. |
| 4 Jan Flag | The win-back flow turned loss-making in December: the held-back group ordered for Christmas too, so the 15% discount went mostly to people who would have ordered anyway. | Keep it, without the discount from 1 to 24 December next year. |
| 4 Jan No flag | The welcome series was back to 5.9 orders per 100. | Nothing needed. |
At the check-in on 27 November, after the first monthly run, Geoff and the managing director read the log together; no rule needed changing. What the agent added: the broken link in the welcome series was found within a month of the template change, and the win-back flow's Christmas loss was seen in its own figures, against its held-back group, rather than in the platform's attributed orders. The flows' value is still the plan's; the agent reads it each month, and the new flows are re-measured against their held-back groups two months after going live. No £ figure is claimed for the agent.