The campaign for one: marketing personalisation stopped being a cost problem

Personalisation is the promise marketing budgets have never let leaders keep.

The reason is worth stating precisely, because it is not the reason most vendors sell against. Plenty of teams already run account-based marketing (ABM) and use personalisation tools for segmentation and targeted delivery. The gap has never been targeting. It has been content. Producing genuinely differentiated content for each audience or persona stayed too slow and too expensive to do at scale, so teams personalised the envelope and standardised the letter inside it.

That constraint is what has shifted, and the shift is recent enough that most marketing business cases have not caught up with it. What is at stake is a category of programme that was assessed, costed, and rejected on production grounds, sometimes only a year or two ago. If the production cost falls by orders of magnitude, those rejections were correct on the evidence available and are now simply out of date.

This is written for marketing leaders who own that business case. It matters just as much to the data protection, legal, and marketing operations colleagues who will inherit the consequences, because everything that made personalisation expensive to produce also made it self-limiting. Remove the cost ceiling, and the governance questions arrive much faster than they used to.

In our testing at Aicadium, that constraint no longer holds. Our SPECTRUM campaign engine generates persona-differentiated campaign packages at under US$0.025 per run in internal benchmarks. That figure changes what is affordable.

Affordability is easier to argue from arithmetic than from adjectives, so here is the arithmetic.

In our internal benchmarks, a complete persona-differentiated campaign run costs under US$0.025 to generate, excluding individual-level graphics. At that rate, a distinct variant for every one of a thousand personas comes to about US$25 in total. The production constraint that justified segment-level compromise has effectively disappeared.

We reached that number through SPECTRUM, our AI campaign engine. It generates three to five differentiated personas from a brief, or even a meeting transcript. It then produces distinct copy per persona across blog, email, social and landing-page formats.

Persona-level differentiation is what that US$0.025 buys today. The question a leader will ask next is how much further the same economics reach.

SPECTRUM today creates persona-level campaigns with differentiated copy for each audience segment you define. Two further levels are within reach with the right inputs. Account-based campaigns are the next step, because you can name the target company and SPECTRUM builds around it. True one-to-one personalisation means a distinct campaign for a single named individual. It is possible when you supply a contact list, with the additional cost of a generated or purchased list. The generation itself stays cheap. The list, and the data behind it, is the added cost.

Reaching for one-to-one only makes sense if persona-level output is genuinely differentiated in the first place. That is the claim most worth testing, and the easiest to fake.

Real personalisation changes the argument, not just the name field. The find-and-replace approach is familiar from marketing automation. Think of the personalisation tokens in mass email platforms that swap in a first name or company while the body stays identical. That is not what we mean. In our runs, persona variants differed in structure and substance. A thought-leadership email read as educational and problem-led, with no sales pressure. A product-launch email for another audience led with outcomes and a direct call to action. The system switches its approach based on campaign goal and persona, producing materially different copy.

That distinction matters for a chief marketing officer (CMO) evaluating vendors. Ask to see two persona variants of the same asset side by side. If the differences are cosmetic, the personalisation is cosmetic.

Being able to differentiate every asset does not mean every asset should be differentiated. Deciding where not to was one of our more useful findings.

Personalisation is a tool, not a default. Organic social posts address a public audience rather than an individual, so we generate them per theme, with the themes derived from an analysis of all personas. Email drip campaigns sit at the other extreme. An email costs effectively nothing to write and deliver, so one drip campaign per persona is affordable. The granularity compounds the relevance of every delivered email.

The practical rule we derived is short. Personalise where the reader is addressed as an individual. Aggregate where they are addressed as a public.

When does a segment become an individual?

A “campaign for one” is a campaign generated for a single, specific recipient. It is built on their industry and role, on the pain points that go with both, and possibly on their interests and search history, rather than a segment they approximately fit. Account-based campaigns are the step before it: the target is a named company. AI generation makes both affordable, and the data you feed it determines whether they are accurate.

What still limits one-to-one personalisation?

Data, judgment, and timing. Generating a thousand variants is cheap. Knowing enough about a thousand individuals to make each variant true, welcome, and compliant is the real work. Doing it in real time, as a prospect acts, is harder still. Privacy regulation and brand risk also scale with specificity, which is why human review gates matter more at higher personalisation levels, not less.

Those limits are not obstacles on the way to one-to-one personalisation. They are the thing that has to be built alongside it.

The closer a campaign gets to an individual, the higher the stakes of getting them wrong. A segment-level error is a wasted impression. An individual-level error is a reputational incident with a name attached, whether the mistake is a wrong role, a stale employer, or a sensitive inference. Personal data used for individual targeting also falls squarely under privacy law. The EU’s General Data Protection Regulation (GDPR) and Singapore’s Personal Data Protection Act (PDPA) both govern exactly this use. Consent and provenance, therefore, need auditing before generation starts, not only after.

Our working practice holds three controls constant as specificity rises. Every variant passes an automated evaluation for brand voice, factual sourcing and bias before human review. Every data field used for personalisation must have a known, defensible source. And a human approves each campaign before anything ships. Generation costs fall. The governance bar does not.

If governance is what scales rather than cost, then the sequencing question changes shape. It is no longer where to find budget. It is where to start.

Rebuild the business case, because the old one is obsolete. Personalisation programmes rejected two years ago on production-cost grounds deserve re-evaluation under generation costs measured in cents. The sequencing that has worked for us is deliberate. Start with a few high-value audiences, prove the persona variants are meaningfully different, and instrument the quality gates. Then extend towards account-based and individual campaigns as your data and lists allow.

The economics are the easy part now. The organisations that win the next cycle will be the ones whose data, governance, and review processes can keep pace with what AI content generation makes possible. How we keep humans in that loop, deliberately, is the subject of the next piece in this series.