Customer Subscription Model: Test a Subscription Offer With AI First

Yes, a small business can add a subscription, and the cheap way to find out is a two-week test before building anything. Work out the margin per customer, guess churn (monthly cancellations) harshly, then ask 50 to 100 existing customers to pay a small deposit. If fewer than about 5% commit at your lowest price, rethink the offer.

Subscriptions are not only boxes of goods. A landscaping firm can sell seasonal care, a yoga studio can sell a membership, a removals firm can sell a yearly storage check, and a tutoring centre can sell a term of weekly sessions. The test below works the same for all of them, and AI helps with the dull parts: the arithmetic, the survey wording and the reading of replies. It cannot tell you what your customers will pay. Only their money can.

Follow me on Instagram@sagnikteaches

The offer we will test: a garden-care plan, step by step

Suppose a landscaping company with three staff wants a monthly plan for homeowners: one visit a month for mowing, edging, weeding and a seasonal job, such as hedge trimming in summer. Right now it earns from one-off jobs and a feast-and-famine winter. The owner thinks $89 a month sounds right, but has no evidence.

Connect on LinkedInSagnik Bhattacharya

All figures here are illustrative, so replace them with your own. The plan is to test three price points, $69, $89 and $119, against a warm list of 150 past customers, with money on the line rather than opinions.

Subscribe on YouTube@sagnikteaches

Step 1: build the margin model before anyone sees the offer

A price that customers accept can still lose money. Do the sum first. For each customer per month, list what the plan costs you to deliver:

• Labour: 1.5 hours at a loaded $24 an hour (wages plus the on-costs you pay) is $36.
• Travel and fuel: $6 per visit.
• Materials and disposal: $4.
• Payment fee: card processing at Stripe's published standard rate for domestic cards, 2.9% plus 30 cents, is about $2.88 on an $89 payment.

Total cost to serve: about $48.88. At $89 the gross margin is about $40.12, or 45%. At $69 the fee is $2.30, so costs are $48.30 and the margin falls to about $20.70, or 30%. At $119, with the same visit plus a larger seasonal job, assume $8 of extra labour and materials and a $3.75 fee: costs of $57.75 and a margin of about $61.25, or 51%.

That already says something: at $69 you would need about twice as many subscribers to earn what $89 earns. Check the table.

Ask an AI assistant to build this, but give it your figures and make it show every line. A prompt that works:

I run a landscaping company. Build a monthly margin table for a garden-care
subscription at three prices: $69, $89 and $119. Costs per customer per month:
labour 1.5 hours at $24, travel $6, materials $4 (add $8 extra labour and
materials at $119). Card fee 2.9% + $0.30 of the price. Show each cost line,
margin in dollars and percent, and how many customers I need for $2,000 a
month of margin. Show the working. Do not add any costs I have not listed.

Check the output with a calculator. Assistants are good at layout and unreliable at long arithmetic, so one or two cells will often be wrong. Fixing those is quicker than building the table by hand, and the check is the job. If you prefer a spreadsheet, the guide to using ChatGPT with Excel shows how to have the formulas written for you. For a product subscription the same model applies, and the margin model for a subscription box adds postage and packaging lines.

Step 2: guess churn harshly and see what it does to the value

Churn is the share of subscribers who cancel each month. You have no data, so guess, and guess pessimistically. If 6% cancel each month, the average customer stays about 1 divided by 0.06, or roughly 16.7 months. Multiply by the monthly margin and you get the lifetime margin per customer.

The "one third" column is a rule of thumb, not a law: spending no more than a third of lifetime margin on winning a customer leaves room for error. Read the table the way an owner would. If your acquisition cost is a $40 flyer drop that wins one customer in five, so $200 per customer, the plan only works if churn is under about 6%. If customers come from your existing list for the price of an email, cost is near zero and even 12% churn is profitable.

Seasonal services churn at the end of the season, so a garden plan might lose a lump of customers every autumn. Model that as an annual drop, or sell the plan as a fixed 12-month term billed monthly, and check with an adviser what notice and cancellation terms apply to you.

A tutoring centre shows how churn varies by offer. A term-long plan for exam students has a natural end, so "churn" is mostly renewals. A yoga studio's membership can lose people early, for example in the first two months after a new-year sign-up rush, so model the first quarter separately. Whatever your business, run 5%, 8% and 12% and see which scenarios still make money.

Step 3: write the offer so one sentence says what they get

A demand test needs a clear offer or it measures confusion. Write it in four lines: what happens each month, what it costs, what is excluded, and how to stop. AI is useful here as an editor. Give it your draft and ask it to find anything a stranger would misunderstand.

Garden Care Plan, $89 a month
One visit a month (up to 90 minutes): mow, edge, weed beds, tidy paths.
Seasonal extra each quarter: hedge trim, leaf clear or bed refresh.
Not included: tree work, new planting, waste over 3 bags.
Cancel any time before the 25th for the following month.

Ask the assistant to list every question a homeowner would have and which of them this text fails to answer. A typical reply (illustrative) points to "what is 'up to 90 minutes' if the garden is large?", "what if it rains?" and "do I need to be home?". Answer those in the offer, because each unanswered question lowers the number who say yes and tells you nothing about demand.

Step 4: choose the test that fits the list you have

The method depends on how many people you can reach, and this is where most online advice is useless, because it assumes thousands of visitors. Work out what your numbers can support.

If you wanted to separate a 2% sign-up rate from a 4% one with reasonable confidence, the standard rule of thumb needs about 16 times p(1-p) divided by the squared difference per group. With p near 3% and a difference of 2 points, that is roughly 1,160 people per price. A small firm with 150 past customers cannot do that, so do not pretend to. Use the test for what it can give:

• Under 100 people reachable: do not split into three prices. Offer one price to everyone and treat the answer as yes, maybe or no. Follow up personally with the ones who hesitated.
• 100 to 500 people: test two prices at most, and read the result as a range, not a verdict.
• Over 1,000 reachable or paid traffic: a true price split becomes possible.

The 150-customer landscaping list sits in the middle. Send $89 to 100 randomly chosen customers, and $119 to the other 50 as a premium option. Then see what the numbers say. Here is how wide the uncertainty really is, using a standard interval for proportions:

Look at the third row: 6 yes answers out of 60 does not mean "10% will buy". It means the true figure could sit anywhere from about 5% to 20%. That is still useful. If the break-even needs 8% of your list to subscribe and the range runs from 5% to 20%, you have enough hope to proceed to a small first launch, not a big investment.

Step 5: the two-week demand test, day by day

This plan uses real money in a low-risk way. Confirm with a qualified adviser how deposits, pre-orders and recurring payments must be described and refunded where you trade.

1. Days 1 to 2: finish the margin table, the churn scenarios and the offer text. Decide in writing the break-even sign-up count, for example "I need at least 15 subscribers from 150 people to make a first month worth running".
2. Day 3: build one page or a simple email with the offer and a payment link for a refundable first-month deposit, say $20 off the first month.
3. Days 4 to 5: send to the list. Randomise who gets which price: sort the customer list by a random column, then take the first 100 for one version and the rest for the other.
4. Days 6 to 12: follow up once. Phone or message the hesitant ones with a single question: "what would make this a yes?" Log every answer verbatim.
5. Days 13 to 14: count deposits and apply the decision rule below. Refund anyone who changed their mind, promptly.

The deposit matters. Asking "would you pay $89 a month?" collects polite agreement. Asking for $20 now collects commitment, and the gap between the two is the number you are trying to find. Customers who say yes to a survey but will not put $20 down tell you something important about real demand.

The decision rule to write down before you start

Write the thresholds before the results arrive, or you will bend them. For the garden plan, the owner writes:

• Go: 15 or more deposits from 150 (10%) and at least 8 at the $89 price.
• Revise: 6 to 14 deposits, or most deposits at $119 and few at $89. Adjust the offer, then rerun with a fresh segment.
• Stop: fewer than 6 deposits, or more than half of the "no" answers give the same reason that cannot be fixed.

Using AI on the answers, with a worked sample

Replies will be messy: emails, texts, scribbled phone notes. Paste them in, with names removed, and ask for themes and counts. An example prompt and a plausible reply, labelled illustrative:

Below are 41 customer replies about a $89 monthly garden plan. Remove nothing.
Group the reasons for saying no or maybe into themes. For each theme give
the count and two short quotes. Then list questions people asked.

The assistant might return: "Price felt high for a small garden (11), prefer paying per visit (9), unsure about winter value (7), want a trial month (5)." Fix its counts by checking a sample of replies yourself. If an assistant says 11 but you find 9, trust yourself. The themes are the useful part. "Prefer paying per visit" suggests a pay-as-you-go option. "Winter value" suggests a lower winter rate. For more on reading large sets of comments, see analysing customer feedback with AI.

Four other businesses, tested differently

A yoga studio: membership against class packs

The studio sells ten-class packs at $120. A $79 monthly unlimited membership tests whether regulars prefer continuity. Cost to serve is nearly zero until the room fills, so the model is about capacity: if 40 members each attend 3 times a week in a 20-person room, you are over capacity. Test with a founding offer limited to 30 places so scarcity is real, not invented.

A removals firm: a yearly storage and move-readiness plan

Moves are rare, so a subscription must offer something between moves: a yearly check of stored items, a box-and-tape delivery and a priority booking slot. The test is smaller and slower, since customers move every few years. Ask past customers, and treat a yes rate under 3% as a sign the plan needs a different promise.

A tutoring centre: a term plan billed monthly

Parents often prefer a plan across a term at a fixed monthly fee to booking week by week. Test by offering the term plan at a small discount to parents of current students and counting how many take it over a fortnight. Churn here is mostly "does the child continue next term".

A product maker: replenishment, not a mystery box

For consumables such as coffee beans or cleaning refills, ask a simpler question: "would you like us to send this every four weeks?" with a skip-or-cancel link. The tutorial on matching items to survey answers covers the curated version.

The test that looked like a success and was not

An example from the sort of trap owners meet. A studio emails 80 members: "Would you join a $79 monthly membership?" Thirty-one reply yes. The owner reads that as 39% demand and starts planning. A week later, with a $20 deposit link sent to those 31, only 4 pay. The true early demand is nearer 5% (4 of 80), not 39%.

Nothing about the 31 was dishonest. Saying yes costs nothing and paying costs something. The fix is a rule: count money, not words. Record both figures in your sheet, "said yes" and "paid", and build the plan on the second.

The costs owners forget until the second month

A subscription creates work a one-off sale does not, and the margin table above leaves it out. Add a line for each of these before you trust the figures.

• Failed payments. Cards expire and bounce. Budget a few minutes per failure to chase, and decide how many days you keep serving before pausing.
• Scheduling. Twelve visits a year per customer is twelve bookings to organise, move for weather and confirm. For 50 customers that is 600 bookings a year.
• Support. Expect questions about skipping, pausing and cancelling. A one-page answer sheet cuts most of them.
• Admin time. Say 20 minutes a month per customer for invoicing, notes and messages. At $24 an hour that is $8 a customer, which would take the $89 margin from $40.12 to about $32.

With that last line the break-even moves from 50 customers to about 63 for $2,000 a month of margin. It is better to meet that figure on paper than in your second month.

Before and after: the same offer, rewritten after the first replies

After ten replies to the garden plan, the owner sees the same confusion twice: "what if my garden is bigger than yours?" and "is winter included?" The rewrite below adds the answers without adding length for its own sake.

BEFORE
Garden Care Plan, $89 a month. One visit a month. Cancel any time.

AFTER
Garden Care Plan, $89 a month for gardens up to 150 square metres
(larger gardens: we quote before you pay). One visit a month, up to
90 minutes. November to February visits cover leaf clearing and
bed protection instead of mowing. Cancel before the 25th for the
following month. First month fully refundable.

The second version prices the large-garden case, answers the winter question and removes the fear of being locked in. None of that is clever; it is simply what the first replies asked for. Run the new text on a fresh group, not the people who saw the old one, or you cannot compare results.

Reading the result and deciding what to build

If the result is a go, build the smallest version that can honestly serve the first batch: a payment link, a simple sign-up form and a calendar. You do not need software yet. Run it for three months, then replace your churn guess with the real cancellation figure. If the result is revise, change one thing at a time, such as the price, the promise or the audience, and rerun. If it is a stop, you have saved the cost of building something nobody would pay for, which is the point of the exercise.

Keep the sheet with your assumptions, results and decision. In six months it becomes a record of what you believed and what proved true. The cancellation reasons and save offers tutorial is the natural next read once subscribers start to leave, and tracking AI KPIs helps you keep the number of figures small. If you want to work through your own margin and test design with someone, that is part of what my AI implementation consultation covers; the contact page is the way in.

Questions owners ask before launching a subscription

Can I take money before the subscription exists?

Many owners take a small refundable deposit or a first-month pre-order, which tests real willingness to pay better than a survey answer. Say clearly what starts when, and how to get the money back. Rules on pre-orders and recurring charges vary, so check them with a qualified adviser before you charge anyone.

How many customers do I need to test a price?

To tell a 2% rate from a 4% rate with confidence you need roughly 1,160 people per price, which most small firms never reach. Instead, ask 50 to 100 warm customers and treat the result as a range. If nobody in 60 says yes at the lowest price, demand is probably under 6%.

What churn rate should I assume?

Assume something harsher than you hope for, because no data exists yet. For planning, run the model at 5%, 8% and 12% monthly cancellations and see which still make money. A product sent each month usually churns faster than a service the customer relies on. Replace the guess with your real figure after three months.

Should the subscription replace one-off sales?

Usually not at first. Offer it beside your current price and see who picks it. Replacing one-off sales removes your control group and can lose customers who dislike commitment. Once the subscription reaches a steady share, you can decide whether to make it the default.

Can AI predict whether customers will pay?

No. AI can draft the offer, build the margin model, write survey questions and summarise replies, but it has never met your customers. Treat its numbers as arithmetic and its suggested prices as hypotheses. Only real people paying, or refusing to pay, tell you whether the offer works.

Further reads

• Is AI Worth It for a Small Subscription Box? Five Jobs First — If the plan is a box, the five jobs AI handles first.
• What Is AI Churn Prediction and Can a Small Gym Use It? — Spot members likely to leave once a membership runs.
• Is AI Dynamic Pricing Worth It for a Small Hotel? — How another small business tests price changes.
• How to Audit Your AI Subscriptions and Cut Wasted Spend — Keep your own tool subscriptions from creeping.
• Is AI Worth It for a Small Business? How to Work Out Your Answer — Check the AI side of the plan pays for itself.
• How Pet Shops Use AI for Stock, Subscriptions and Advice — A shop's version of replenishment subscriptions.
• Tea Shop Marketing Ideas: Tasting Notes and Gift Sets Written With AI — Turn your own cupping notes into shelf cards and product pages, then price subscription tiers and gift sets by margin before AI writes a word of copy.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Stripe published card-processing rate for domestic cards (2.9% + 30c, stripe.com/us/pricing, read 5 November 2026); standard statistical formulas for proportions (Wilson interval, 16p(1-p)/d-squared sample-size rule of thumb).