TL;DR Every stage of the marketing value chain is faster and cheaper with AI, and every stage now produces plausible output that nothing in the workflow checks.
- Buyer research, positioning, creative, outbound, and measurement all compressed from weeks or months to days or minutes.
- Most AI marketing workflows depend on two questions: which box of the chain the workflow belongs in, and the specific check that would catch its failure mode.
- The decisions that matter most are made in understanding, strategy and measurement.
- AI gives the same lift to a poor strategy as to a good one, so the lasting advantage is the judgement that tells a good output from a plausible wrong one.
Part 2 of Marketing in the Age of AI. Start with part 1 at /insights/articles/marketing-value-chain/ or the series overview at /insights/articles/marketing-in-the-age-of-ai/.
AI has compressed cost and time in every piece of the marketing value chain. Buyer research that took months now takes a day, a positioning draft that needed a strategy engagement arrives in minutes, creative variants cost almost nothing, and outbound can be personalised across thousands of accounts at once.
As generation gets cheaper, the bottleneck is judging whether the output is right for your market. The output looks finished and reads well. Whether the output is right for your market decides whether it is any use, and the model cannot judge that.
Below is the whole chain, before and after, with the failure mode that shows up in each box.
The map
| Stage | Before AI | After AI | The failure mode (B2B) |
|---|---|---|---|
| 1. Understand | Analyst reports and months of buyer research | Scraped firmographics, intent data, synthetic ICP research | A confident ICP for a buyer that is not yours |
| 2. Strategise | Strategy consultants, several weeks | A positioning draft in minutes | Plausible but undifferentiated: a claim any competitor in the category could also make |
| 3. Create | Product marketing plus agency rounds | Many variants, in minutes | Polished, interchangeable copy that could belong to any competitor |
| 4. Communicate | Hand-built ABM and outbound | Generated and auto-personalised at scale | Reaching the wrong accounts faster |
| 5. Measure | Quarterly attribution and brand trackers | Real-time pipeline dashboards | Optimising MQLs over revenue: precise answers to a question that does not matter |
Every row of the table shows the same pattern. The "after AI" column is better on cost and speed. The failure column shows the result of taking the output at face value and skipping the check.
Each stage in detail
1. Understand
The old way of understanding a market was slow and expensive, so teams did it rarely and the insight went stale. AI changes the cost: you can assemble a buyer picture from firmographics, intent signals and synthetic research in a day.
A synthesised ICP looks clean and complete. It has a job title, a set of pain points, a buying committee. It reads like a real buyer. Whether it matches the accounts that close is a separate question, and the only way to answer it is to compare the profile with closed deals. A confident, well-formatted profile of the wrong buyer sends everything downstream in the wrong direction.
2. Strategise
Positioning used to arrive after weeks of consultant time. Now a coherent draft appears in minutes, and it is rarely differentiated. Ask a model to position a B2B company and the draft tends to be a category-generic claim to innovation and trust that says nothing about how the company differs.
The failure here is subtle because the output is coherent and still undifferentiated. It describes what every company in the category would say about itself. The test is whether a competitor could put their logo on your positioning and have it still fit. If they could, you have a first draft and the strategy work is still ahead.
3. Create
The cost fell furthest at the create stage. Variants are free. You can generate fifty versions of a landing page at almost no cost.
The default output is polished and reads like every competitor's copy. Fifty variants of the same generic claim are still generic. The abundance can hide the problem: shipping this much feels productive, and none of it says anything a buyer could not get from another vendor. The check is whether any of it is specific enough to be true only of you.
4. Communicate
Outbound and ABM used to be hand-built, which put a natural ceiling on volume and forced some selectivity. AI removes that limit. You can generate and personalise across thousands of accounts at once.
If the account selection is off, the failure mode is reaching the wrong accounts faster. The decision that matters is which accounts to select, and that is made at the understand stage.
5. Measure
Measurement went from a quarterly attribution exercise to a live dashboard. The numbers refresh in real time and the charts look authoritative.
That authority is the risk: a real-time dashboard optimising MQLs gives you precise, fast answers to a question that may not matter. If the team chooses MQLs as the metric, the dashboard reports that metric precisely and quickly, and revenue is not on it. A team that optimises to it generates more of the wrong leads, more efficiently, and the chart rises while it does.
Using the map
Two decisions apply to any AI marketing workflow you are building or buying: the box it belongs in, and the check for that box's failure mode.
The box question is which stage of the chain the workflow belongs to. If a tool belongs in box 4, that is fine. However, the decisions with the most effect on results are made in boxes 1, 2 and 5: understand, strategise and measure. Naming the box shows whether the tool adds speed where speed was the limit, or where the limit was the quality of the decision upstream.
The check question is which failure mode that box produces and which test would catch it. The answer is your evaluation function, the specific check for that stage. For the ICP it is whether the profile matches closed accounts; for positioning, whether a competitor could use it unchanged; for creative, whether the claim is true only of you; for outbound, whether the account fits; for measurement, whether the metric moves with revenue. If you cannot state the check, you are shipping unverified output at scale.
The takeaway
AI scales whatever strategy it is given, including a bad one, so generation stopped being an advantage once everyone had it at roughly the same quality, from the same handful of models.
Teams now differ on the judgement to tell good from plausible-wrong at each stage of the chain, and on the discipline to make that judgement an explicit step in the work.
The evaluation function is harder to build than the map, and part 3 and part 4 of this series cover it: the B2B positioning version of the check at /insights/articles/b2b-preferred-partner-positioning/, and the team-level version at /insights/articles/marketing-evaluation-function/.
NavigatorLab maps consumer demand spaces, sizing consumer typologies and consumption occasions in pounds, with growth scoring and brand fit.