The five ways to get results from AI in marketing: hypothesis testing, brand voice scoring, MMM upgrades, personalisation flows and creative iteration.
A CMO came up to me after a conference last month. Tablet in hand, 100 LinkedIn posts on the screen. "AI produced these. We publish 100 posts a week. What do we do now?" My short answer was blunt: do nothing. Those posts were producing no value at all.
The biggest illusion about AI in marketing is this: teams mistake output volume for productivity. The real question is different. Which decision did AI get you to faster? On which campaign did you produce a better hypothesis? On which message did you spend less money?
Here is a truth distilled from the field: marketing operates as a decision chain far more than a process. Used well, AI halves the number of links in that chain. Used badly, it adds links and wears the team out.
In the old world, a marketing team forms a claim: young female users convert better with emotional messaging. Testing that claim takes 4 weeks of briefing, 6 weeks of creative and 4 weeks of testing.
The result arrives 14 weeks later. The campaign either works or it doesn't.
In the AI world, you open the same hypothesis into 12 different variations in 30 minutes. You show Claude or GPT your past campaign data and ask: which 12 message frames have never been tested on this segment?
Once the answer lands, the team takes the first 4 frames to market within 1 week.
I see the same pattern across most of the e-commerce companies I work with. A campaign that ran on a single scenario in 2020 can now enter A/B testing with 8 different message frames in 3 days.
The cost stays the same while the speed rises to 5 times. AI multiplies the number of tests and puts decision material in front of the team quickly.
One note matters here: AI builds the hypothesis, you run the test, the customer decides. Do not break that order. Do not take an AI's "this is the best message" at face value. AI predicts; the real click-through rate confirms.
Brands have lived with a nightmare for years. 4 different agencies, 6 different content creators, 12 different channels. Does the output belong to the same brand? Usually no. The brand book exists; nobody reads it.
AI offers a surgical fix for this problem. You feed it your brand book, the golden example content from the last 12 months and 20 pages of examples you never want to see again.
From then on, every new piece of content receives a brand voice score between 0 and 100; anything below 70 goes back for revision.
In multi-brand group structures, agency output rises by an average of 12 to 15 points once this scoring layer is in place. Same brief, same budget; brand consistency climbs in measurable terms.
According to McKinsey's 2025 report, companies that measure brand voice with AI see customer recognition metrics that are 23 per cent higher. The reason is simple: consistency feeds trust, and trust drives recall.
MMM is known as the marketer's most tedious, most expensive and slowest piece of work. It is usually run once a year, takes 12 weeks, gets handed to an external consultancy and ends its life in a report.
The AI-driven approach turns MMM into a continuously flowing decision-support tool. The model is fed the past 36 months of media spend, sales, conversion and market share data. AI then runs a live weekly simulation of one question: what happens to sales if I pull one unit of spend out of this channel?
Brands are now doing what classical regression used to require at half the cost and 10 times the speed. Coupa's published figure of 276 per cent ROI comes out of this category. The reason is clear: marketing budgets are large, misallocation is expensive, and even a small improvement produces a large return.
A brand should plan a budget of 8 to 12 thousand dollars a month to run MMM with AI.
In return, it gains 8 to 15 per cent in annual media efficiency through the right channel combination.
The figure in Salesforce's 2026 sales report reads as follows: AI-supported personalisation lifts sales productivity by 67 per cent. Used raw, that figure misleads. In marketing, personalisation only produces an impact as large as it does in sales when the right segmentation architecture is in place.
A correct personalisation flow is built in order. First, customer segments are re-derived with AI. The old demographic segment logic no longer works. Segments are drawn from behaviour, purchase frequency, channel preference and lifecycle.
Then 3 message variations are prepared for each segment. AI produces them; the content passes through the human team's quality control. Most personalisation projects sink at this point. When AI output goes live unchecked, brand erosion begins.
The third layer is channel automation. Across CRM, email, push and in-app channels, AI suggests which customer should be contacted through which channel and at what time.
One B2C brand that managed to build this structure runs separate contact flows across 6 segments; AI's weekly improvement suggestions go live with team approval. After 6 weeks, the weekly active user metric can climb by double-digit percentages.
The hardest admission for a CMO: 70 per cent of creative output is average, 20 per cent is weak and 10 per cent is brilliant. AI cannot shake that ratio, but it shortens the time needed to find the brilliant one.
In the old iteration, the agency brings 3 concepts, the marketing director decides which one runs, and usually picks the middle one.
AI iteration works differently. AI produces 12 to 15 concept variations, the team narrows down to a top 3 in a short working session, then the idea is polished by human hands.
The distinction that matters: AI enriches combinations rather than producing creativity. The brilliant idea does not come from AI. But AI will lay out 8 visual combinations, 6 headline melodies and 4 story arcs of that brilliant idea in a matter of minutes.
This is where the "agent manager" concept HBR published in early 2026 earns its meaning. The marketing leader now works with 1 agency and a handful of AI agents instead of 12 agencies. The management style changes with it.
Back to the CMO at the conference. AI produced 100 posts a week for him. That output amounted to a pile of waste far more than value, because every post spoke with the same voice, the same structure, to the same target audience. It had quantity without quality.
Correct use produces 10 times the return. If the same CMO ran AI differently, the picture would change: 5 posts produced a week, each speaking to a different buyer persona.
Every post is held above a brand voice score of 80, modelled on the best-performing post structures of the last 12 months.
Those 5 posts generate 10 times the conversion of 100 generic ones.
MIT Sloan's 2025 research points at the same spot: 95 per cent of marketing teams that fail to produce value from AI are using it without a quality layer.
Play | Time Saved | Decision Quality | Brand Risk |
Hypothesis Generation | High | Medium-high | Low |
Brand Voice Scoring | Medium | High | Very low |
MMM | High | High | Low |
Personalisation | High | High | Medium |
Creative Iteration | Very high | Medium | Medium-high |
Decide your first step against this matrix. Brand voice scoring carries the least risk and delivers the fastest result. It is the best place to start in the first 90 days.
Once the marketing engine is in order, the next link in the chain is sales. Empowering your sales team with AI in 6 steps carries the same discipline into the sales line. For the customer service side, see the Klarna case study analysis.
The five ways to get results from AI in marketing run through hypothesis testing, brand voice scoring, MMM upgrades, personalisation flows and creative iteration. Decision speed matters far more than output count.
Producing 100 LinkedIn posts does not count as value. Correct use produces 10 times the return with fewer, sharper pieces. Start with brand voice scoring in the first 90 days.
If you would like these five plays brought to your leadership team as a live case story on stage, browse Speaker Agency's artificial intelligence speakers and get in touch. Our guide on how to choose an AI keynote speaker will help you match the right name to the brief.
Brand voice scoring is the easiest entry point. The team already holds its brand book and the golden examples from the last 12 months. These are fed into the AI system and a score is produced. Within 6 weeks, the internal team's view of AI changes. Hypothesis generation should be the next step.
A brand voice layer is mandatory. AI output receives a score first; anything below 70 goes back. A human editor runs quality control in parallel. In multi-brand structures, this layer alone brings a 12 to 15 point improvement. Unchecked output produces brand erosion.
Classical MMM takes 12 weeks; continuous AI-driven MMM updates weekly. Initial setup takes 6 to 8 weeks, after which the flow runs automatically. For brands with large budgets it is the highest-return play; the media efficiency gain is measured in concrete annual terms.
For a mid-sized B2C brand, 8 to 15 thousand dollars a month is the starting point. That covers the content production API, the brand voice layer, the MMM platform and team training. ROI is measured between three and six months in; do not expect a return in the first 90 days.
They are transforming rather than ending. The CMO who worked with 12 agencies is moving to a structure of 1 agency plus several AI agents. The agency's value is shifting from production speed to strategic perspective. Production goes to AI, strategy stays with humans. HBR calls this role the agent manager.
After Dinner Speakers
Contents
