2027 Retail Leaders’ Series · The martech ceiling
Martech has hit its ceiling.
Retailers keep paying more for customer engagement tools that are used less, prove less, and still need a team to run them. The value has moved to a different layer. This report shows the numbers, why the ceiling exists, and what replaces it.
- 33%
of martech capability actually used
Gartner, 2023
- 78%
of marketing leaders say martech fails to deliver ROI
eClerx, 2026
- 38%
retail leaders who say their AI investment is working
MarTech, 2025→2026
- 45%
say vendor AI agents miss promised business performance
Gartner, 2025
The evidence
For a decade, the answer to a customer engagement problem was another tool or another module. The data now says that answer has stopped working.
Martech is the second largest line in the marketing budget at 22.4%, and marketing labor takes another 21.9%. Close to half of every marketing euro goes to software and the people who plan, build and run the programs on it. Over the same period, usage of that software fell by almost half, and confidence in the AI layered on top of it is falling year on year.
Exhibit 1
Share of martech capability in use
Gartner Marketing Technology Surveys (2023: 405 marketing leaders). Two thirds of what is licensed sits idle.
Exhibit 2
Retail leaders who say their AI investment is working
−16 pts in one year
MarTech, “The truth about martech in 2026”. Adoption held steady; confidence did not.
Exhibit 3
The ceiling, in six numbers
- 44%
of marketing budget goes to martech (22.4%) plus marketing labor (21.9%)
Gartner CMO Spend Survey 2025, n=402
0 of 50+
senior marketing leaders could clearly explain the ROI of their martech
McKinsey, Rewiring Martech, 2025
- 53%
of customers say personalized marketing gave them a negative experience
Gartner, 2025, n=1,464
- >40%
of agentic AI projects expected to be cancelled by end of 2027
Gartner, June 2025
- ~130
of thousands of “agentic” vendors judged to be real agents; the rest is relabelled assistants
Gartner, June 2025
- 15,384
martech tools on the market, up 9% in a year. More choice, same operating model
chiefmartec, 2025
The broken ROI
Most engagement revenue comes from simple sends: newsletters, bulk campaigns and transactional triggers. Most of the bill is for the lifecycle automation layer on top.
Exhibit 4
Where engagement revenue comes from vs. what the platform bills for
Where the revenue comes from
- 20%Lifecycle automations
- 55%Segmented / bulk campaigns
- 25%Triggered notifications
What you are billed for
- 90%Lifecycle automation capability
- 10%Bulk, segmented and triggered delivery
90% of the bill for 20% of the revenue
Replenit analysis of typical enterprise retail engagement programs. Illustrative.
The revenue that keeps the program alive comes from work any reliable, API-based delivery service can do. Sending bulk email through a cloud email API costs in the order of $0.10 to $0.23 per 1,000 emails. The premium is for the lifecycle layer, and that layer is exactly where output is capped by how many journeys a team can design, build and maintain.
Why the ceiling exists
Retention is simple to define. “Keep this customer buying” fits in one sentence. It is expensive to operate because every tool turns that sentence into hand-built rules, segments, journeys and caps.
Customer engagement platforms were born in the software era. Their unit of work is the campaign or the journey, and their capacity is the size of the team that operates them. Three structural limits follow, and no AI feature added on top removes them.
Three structural limits
What each limit means in practice, and what it costs
Structural limit
What it means in practice
The cost
The objective is traded to fit the tool
A business objective becomes a campaign brief shaped by what the platform can configure. Whatever the tool cannot express is dropped before launch.
Part of every objective never reaches a customer.
Humans are the engine
Segments, filters, nodes, triggers and frequency caps are built and maintained by hand. Output grows only with headcount.
Marketing labor is 21.9% of budget, as large as the software line. The operating team is a major part of the true cost of ownership.
Priced on usage, not outcomes
Licenses are charged on profiles, seats or messages whether capability is used or not. The vendor is not accountable for the commercial result.
33% utilization at full price. Renewal is the vendor’s goal; your outcome is yours to deliver.
Exhibit 5
What martech AI optimizes vs. what the business needs decided
Question
Martech AI (co-pilot)
What the objective requires
Starting point
A campaign your team already built and launched
The business objective itself, at full scope
What it tunes
Send time, subject line, variant, channel pick
Whether to act at all, which workflow, which product, why now, in what tone
How it learns
Reinforcement on clicks and opens inside human-defined experiments
From what each customer bought, needs and ignored, tied to the commercial outcome
Who directs it
A marketer, every time
Your team sets objectives and guardrails once; the work runs
Who it serves
Makes the tool operate better, which protects the renewal
Manages your customers better, which grows the P&L
Gartner (Oct 2025, n=413): 45% of martech leaders using vendor AI agents say they miss expected business performance.
The new operating model
Three operating models, three eras. Each starts from the same business objectives. What changes is how much of the objective reaches the customer.
Exhibit 6
Only the workforce model carries the full objective through to every customer
1 · The campaign era (2010s–2023)
Inside the automation / personalization tool
Business objectives
Campaign brief
Rules, segments, journeys
Built by hand
Channels / campaigns
Objective narrowed to fit the tool
2 · Legacy + martech AI (2025–now)
Same tool, AI added on top
Business objectives
Campaign brief
Rules, segments, journeys
AI tunes sends and variants
Channels / campaigns
Same model. The objective lost upstream stays lost.
3 · The workforce era (now)
Pre-built and your own skills, instructions, guardrails
Business objectives
Full scope
AI workforce
Skills · Memory · sub-decisions
Execution-ready outputs
Product, timing, content, reason
Every surface
CRM, web, app, store
↺ Learns from every outcome
Exhibit 7
Where value is created, before and after
Objectives & guardrails
Your team
Set the commercial goals, brand rules, offer limits, and teach skills in plain language.
Intelligence layer
AI workforce · new value layer
Reads customer, product and brand memory. Applies skills. Makes the micro and macro sub-decisions behind every action: who, which product, why now, in what tone, toward which outcome. Owns the result.
Execution layer
Your existing stack
CEP, MAP, ESP, app, web, store. Kept as is. Becomes a reliable delivery pipe instead of the place where work is designed.
Value moves from the tool you operate to the layer that operates for you.
What we believe
The business objective should reach the customer whole.
Nothing traded away to fit a tool you paid for.
Every customer deserves reasoning of their own.
Who, what, when and why, worked out 1:1, every day, on every surface.
Teams should lead, and AI workforces should operate.
Your team sets objectives, teaches skills and writes instructions. Agents do the operational work around the clock.
Intelligence should compound.
Every outcome makes the next decision sharper.
What Replenit does differently
Replenit builds AI workforces. Maestro, our AI CRM Manager, runs each business objective through skills, so the objective is managed end to end instead of rebuilt by hand as campaigns.
Exhibit 8
One objective, managed by skills
Objective
Lift repeat revenue from active customers
Skills Maestro applies
- 1st to 2nd purchase
- Replenishment timing per product
- Product-level churn catch: an active customer quietly drops one recurring item
- Routine completion (e.g. skincare)
Sub-decisions per customer
- Act now, wait, or suppress
- Which product and why
- Tone and channel
- Output sent to your CRM / MAP
Skills belong to the retailer. Their data, guardrails and logic are never used to train models for, or serve, anyone else.
Side by side
Traditional martech vs Replenit · Maestro
Dimension
Traditional martech
Replenit · Maestro
Unit of work
Campaign, segment, journey
Skill, reasoned per customer
Who operates
Your team, by hand
Maestro; your team directs
Business objective
Narrowed to fit the tool
Run at full scope through skills
Role of AI
Co-pilot tuning sends
Makes and explains the decisions behind each action
Accountability
License paid regardless of result
Success-based; owns the outcome
Your stack
Where work is designed
Kept; becomes the execution pipe
Data needed to start
Long CDP/data projects
Order history and product feed
Exhibit 9
Built on Google Cloud, proven in retail
- 90%
lower AI pipeline cost: from over $200 to $20 per run
Google Cloud case study
- 10x
faster pipeline execution; new pipelines shipped in 4 hours, not 3 days
Google Cloud case study
- +235%
post-purchase revenue at L’Occitane
Replenit customer result
- 9.1x
revenue per decision at Escentual in 9 months
Replenit customer result
Maestro runs hundreds of specialized agents in parallel on BigQuery, fine-tuned Gemma models on Cloud TPUs and Gemini, with SOC 2 controls for tenant isolation and auditable decisions. The economics matter: an intelligence layer that is cheap to run is what makes 1:1 reasoning viable for every customer, not just a top segment.
Before your next renewal
What share of the capability we license do we actually use?
The market average is one third.
What share of engagement revenue comes from bulk sends a delivery API could handle?
If it is most of it, the premium is paying for something else.
How many people does it take to run our lifecycle program, and what does that cost per year?
Labor is as large a budget line as the software itself.
What does our vendor’s AI decide, and what does it only optimize?
Send time and subject lines are not a retention strategy.
Who owns the outcome if the numbers do not move?
If the answer is “we do”, the tool is a cost, not a capability.
Martech made retailers better at running campaigns. The next gain comes from not having to run them.
Keep the stack as the execution layer. Put the objectives in the hands of a workforce that owns them.