How to Find Your Next Profitable Customer Segment

A business with loyal customers and flat growth usually hears the same advice: spend more. So the budget goes up, the platforms absorb it, and the cost per new customer climbs while the number of genuinely new customers barely moves. In an illustrative quarter, finance asks why acquisition costs rose 30% for the same output, and nobody has a good answer.

The problem is rarely only the budget. It is that the extra money bought more of the same audience. Growth stalls when a business keeps optimising for the customers it already reaches. The way out is not louder spend; it is a better question: who else, specifically, should be buying from us, and what evidence do we have?

This is a method for answering that question with your own customer economics, testing the answer cheaply, and scaling only what holds up. It is the segment-level twin of our guide to testing a new channel before scaling, and it borrows the same rule: model first, test small, scale on evidence.

Two forces push mature accounts towards the customers they already reach.

Optimisation gravity. Smart Bidding and lookalike systems learn from the conversion signals you give them. If those signals mostly come from the easiest converters, the machine can keep finding people close to the customers you already have. Your audiences become a denser circle around the current buyer base, and each incremental pound risks buying a slightly more expensive copy of the same person. As we argued in PPC in 2026 vs 2016, the auction now prices your business economics; when your targeting stops widening, rising CPCs become much harder to absorb.

Volume-led segmentation. Many businesses, when they segment at all, segment by order count. The "best customers" are the ones who buy most often, and the platforms are fed audiences built from them. But the highest-volume segment is not always the highest-value one once you account for margin, discounts, returns, and service cost. Optimising acquisition towards your highest-volume buyers can mean recruiting less profitable customers at increasing cost.

Both forces are invisible in platform reporting, because platform reporting has no idea what a customer is worth to you after the click. Which is where the method starts.

Before touching any platform, answer one question from your own records: which customers were actually worth acquiring?

Pull 12 to 24 months of customer data from your CRM or order system and rank customers by lifetime value with margin included, not by revenue alone and certainly not by order count. Depending on your business, that means repeat rate, average margin per order, retention or churn, refund behaviour, and service cost. You are looking for the top slice of customers whose economics you would happily replicate. The full mechanics, including how to turn the ranking into targetable seeds, are in our guide to LTV segmentation in paid media.

Then profile that slice, and here is the discipline: profile them on what was true before they bought. What did they first purchase? What did they search? Where did they arrive from? What size or type of business are they? Which occasion or problem triggered the first order? The attributes that existed before conversion are the only ones you can target on; everything after conversion is outcome, not signal.

A deliberately invented example makes this concrete. A coffee-equipment retailer ranks two years of customers by margin-inclusive LTV and finds the top slice is not the espresso hobbyists who order most often. It is small cafes and offices: fewer orders each, but higher basket values, fewer returns, and consumables on repeat. Profiled before first purchase, they searched supply-and-service terms rather than product-review terms, and their first order was a bundle rather than a single machine. That is a targetable difference. It was not visible in the ad platform, because the ad platform ranked segments by conversion volume, and the hobbyists won that ranking.

With the high-value profile in hand, generate expansion hypotheses by asking what near-neighbour groups share the same underlying economics. Adjacency usually runs along a small set of dimensions:

• Use case: same product solving a different problem (the office kitchen instead of the home kitchen).
• Buyer type: same problem, different buyer (the operations manager instead of the enthusiast).
• Vertical: the industries next to the ones your best customers sit in.
• Occasion: the trigger events that create first purchases (opening, upgrading, replacing, gifting).
• Geography: markets where your economics should transfer, checked against acquisition cost benchmarks rather than assumed.

Write each hypothesis as a sentence a finance lead can challenge: "Segment X resembles our high-LTV customers in A and B, should convert at economics no worse than C, and we can reach them through D." If you cannot fill in those blanks, it is not a hypothesis yet; it is a hunch.

Before money goes anywhere, put a number range on each segment, with the assumptions shown. Search volume for the segment's intent terms, audience sizes on the relevant platforms, and category data give you a reachable-demand estimate. Multiply through your expected conversion economics and the question becomes concrete: if this segment performs at our target economics, how many customers a month is it worth, and does that move the growth number we care about?

Two honesty rules. First, show the assumptions next to the answer; a sizing whose assumptions are hidden is a pitch, not a plan. Second, size the segment you can actually reach with the intent signals that exist, not the theoretical population. The gap between "all UK cafes" and "UK cafes currently searching for equipment suppliers" can be large, and only the second number funds a forecast. Invented arithmetic for shape: 4,000 monthly searches across the segment's intent terms, with 3% clicking through at your economics, is 120 visits, not a growth strategy on its own. The same segment with three more reachable intent themes and a partnerships route might be. The point of sizing is to kill weak hypotheses cheaply, before the ad budget does it expensively.

Segments earn budget in stages, and the pass marks are unit economics, not platform metrics.

Stage one: message evidence. Before building campaigns, test whether the segment responds to the proposition at all: a dedicated landing page variant, segment-specific copy against a small paid audience, or direct outreach in B2B. Cost: small. Answer: does anyone in this segment raise a hand?

Stage two: controlled paid test. One segment, one offer, a fixed budget, and a pre-agreed pass mark written down before launch: target cost per first customer, expected payback window against the segment's projected LTV, and a decision date. Judge the test on blended results and, where volume allows, the methods in our incrementality guide. Do not judge it on platform ROAS alone. As the platform versus incremental ROAS distinction shows, mature retargeting-heavy segments often look better on the dashboard than a genuinely new audience. That comparison can kill good expansions in week two.

Stage three: scale behind evidence. A segment that passes gets budget in steps, with the same review at each step. This staging discipline is the core of how we run customer acquisition growth engagements: value evidence first, sized hypotheses, cheap tests, and scale gates that finance signs off on, because they helped set them.

The gate at each budget step uses numbers that are yours to set. In this illustration, cost per new customer stays within the payback target, first-order margin recovers acquisition cost within your chosen number of months, margin holds as volume grows, and refund and service costs stay in line with the source segment. When a scaling segment starts missing the gate, the honest reading is often that the reachable slice is exhausted at current economics, not that the team failed. Cap it, harvest it, and move to the next hypothesis.

One platform-side action improves everything above: feed value back. If your conversion data tells the platforms every customer is worth the same, the machine has little reason to prefer value over volume. Import value-weighted conversions, margin-adjusted where you can, so automated bidding has a better signal for the customers your evidence says are worth having. Value-based bidding is the bridge between the analysis in this article and what the algorithm actually does with your money.

• Do not buy lookalikes of your biggest audience segment without checking whose economics you are replicating. Lookalikes start from the seed; a volume-led seed can replicate volume-led economics. The full diagnosis is in why lookalike audiences stop working.
• Do not judge a new segment against a mature segment's CPA in its first weeks. New segments carry learning costs and lack retargeting pools; compare against the pass mark you set, not against your best legacy number.
• Do not test five segments at once. Budget spread across many small tests produces many unreadable answers. One or two segments, properly funded and properly measured, beat a portfolio of noise.
• Do not skip the waste check. Expansion funded on top of a leaky account scales the leaks alongside the tests. Clean first, then grow: the pre-scale waste audit is the place to start.

• Rank 12 to 24 months of customers by margin-inclusive LTV. Identify the top slice you would happily replicate.
• Profile that slice on pre-purchase attributes only: first product, search terms, arrival path, trigger occasion.
• Write three adjacent-segment hypotheses in the challengeable format: resembles our best customers in A and B, expected economics C, reachable through D.
• Size each honestly: reachable demand times expected economics, assumptions shown. Kill the weak ones on paper.
• Run stage-one message tests on the survivors. Cheap, fast, binary.
• Give the best survivor a controlled paid test with a written pass mark and decision date.
• Scale in steps behind the unit-economics gate, and feed value-weighted conversions back to the platforms as you go.

If steps one and two are impossible because the customer data is not connected to the marketing data, that is finding one, and it explains more about the growth stall than any audience setting.