How agentic work can help companies grow revenue.
The strongest business case for AI agents is not “more automation”. It is a better growth system — one that notices an opportunity, coordinates the work to act on it, and learns from the result, with people still holding the decisions that matter.
An AI agent is software that can interpret a goal, use approved tools and data, take a sequence of actions and return a result. The commercial value appears when that capability is embedded in a complete workflow — qualifying a lead, preparing a proposal, resolving a customer issue, spotting a retention risk — rather than sitting beside the workflow as another place to ask questions.
How to read this
Four kinds of statement, labelled throughout
Most writing about agentic AI blends survey results, vendor projections and opinion into a single confident narrative. This briefing keeps them apart. Every substantive claim below carries one of these labels.
Something a published study measured or reported, with the sample described. Usually self-reported by the organizations surveyed.
A projection or estimate about the future. Useful for direction and scale. Not evidence, and not a commitment.
Our own interpretation, drawn from delivery experience. Argued, not measured — disagree with it freely.
A practical step we would advise a client to take, and the conditions under which it applies.
The evidence
The opportunity is large. Maturity is still rare.
The Capgemini Research Institute surveyed 1,500 senior executives across 14 countries on AI agents. At the time of the survey, only 2% of the organizations surveyed reported deploying agents at scale.
Source: Capgemini Research Institute, Rise of agentic AIThe same research estimated that AI agents could generate up to $450 billion in economic value by 2028 across the markets studied, through a combination of revenue growth and cost savings.
This is a modelled projection about a market, not a figure any individual company should plan against. It tells you the direction of travel and roughly how seriously to take the topic. It does not tell you what an agent will be worth in your organization.
Source: Capgemini Research Institute, Rise of agentic AIPwC’s 2025 survey of 308 US executives found that, among organizations already adopting agents, respondents reported productivity gains (66%), cost savings (57%), faster decision-making (55%) and improved customer experience (54%).
These are self-reported outcomes from adopters. They describe what the organizations most committed to agents believe they are seeing — a useful signal, but one shaped by who chose to adopt early.
Source: PwC, AI Agent Survey- 2%
- of organizations surveyed by Capgemini had deployed agents at scale
- 66%
- of agent adopters in PwC’s survey reported productivity gains
- 54%
- of the same adopters reported improved customer experience
None of the figures above is a revenue outcome. Productivity, cost and decision speed are the machinery that produces revenue — sales capacity, customer loyalty, service quality, time to respond — but the link is indirect and easy to lose.
The gap between a large forecast and a 2% deployment rate is the more interesting number. It says the constraint is not model capability. It is workflow design, data access, controls and the organizational work of changing how a process actually runs.
Growth levers
Where agents can influence the top line
Across the engagements we see, agentic work tends to reach revenue through four routes. Each depends on the agent being wired into a real process with real permissions — not on the quality of its answers alone.
-
01
Convert demand faster
An agent can research an account, enrich a lead, recommend the next action and draft a tailored response before interest cools. People still approve the high-value communication and own the negotiation.
-
02
Make each interaction more relevant
Agents can combine consented customer history, product knowledge and live context to support better recommendations across sales and service channels, using the same governed data your reporting already relies on.
-
03
Protect the revenue you already have
By watching service signals, usage patterns and unresolved issues, an agent can surface a retention risk early and coordinate a response while it still matters. Retained revenue is usually cheaper to win than new revenue.
-
04
Shorten the path to new offers
Agents can help teams synthesize market feedback, assemble analysis and carry the repetitive parts of a launch, which returns time to the people making the product judgment.
“Agentic” does not mean removing people. It means giving people a dependable digital colleague that can carry work across systems, within limits someone has set deliberately.
The revenue system
Five elements turn an agent into business value
When an agentic pilot fails to produce anything measurable, one of these five is almost always missing. They are not sequential stages; they have to hold together at once.
-
01
A measurable commercial goal
Start from conversion, retention, average order value or sales-cycle length — a number someone already owns — rather than from a technology demonstration.
-
02
Trusted context
Connect governed product, customer and operational data, with permissions enforced at the source rather than reimplemented inside the agent.
-
03
A complete workflow
Give the agent approved tools to move from insight to action. An agent that stops at an answer has moved the work, not removed it.
-
04
Clear human decision rights
Define when the agent may act, when it must ask, and who is accountable for the outcome. Write it down before launch, not after the first incident.
-
05
A learning loop
Measure outcomes, review failures honestly and improve the instructions, data and controls. Without this an agent quietly degrades as the business changes around it.
Measurement
Connect operating metrics to revenue metrics
Agentic workflows produce operating improvements first. Revenue follows only if the operating metric is genuinely upstream of a commercial one, and if you were measuring both before the agent arrived.
| Agentic workflow | Leading indicator | Revenue measure |
|---|---|---|
| Lead qualification | Response time, qualified-lead rate | Conversion rate, sales-cycle length |
| Customer service | Resolution time, repeat contacts | Retention, renewal and expansion |
| Personalized recommendations | Engagement, acceptance rate | Average order value, revenue per customer |
| Proposal generation | Preparation time, approval rate | Win rate, pipeline velocity |
Google Cloud’s global study of 2,500 senior leaders reported that 74% of organizations were seeing a return on their generative AI investment. Among the subset already running in production and reporting revenue growth, 86% estimated gains of 6% or more in overall annual revenue.
Read that second figure carefully: it describes a filtered group — organizations that had reached production and already reported growth. It is not the share of all adopters that grew, and it is not attributable to agents specifically.
Source: Google Cloud, The ROI of generative AISurveys of this kind measure association, not causation, and they under-represent the organizations that tried and stopped. Treat them as evidence that production deployment is where value concentrates — and as a reason to establish your own baseline before you build, so you are not left arguing about attribution afterwards.
Practical next step
Start with one revenue journey, not ten agents
Choose a single workflow that has a named owner, enough volume for a result to be visible within a quarter, and a gap someone already complains about. Record the current baseline before any build work begins.
Then build the smallest safe agentic loop that can complete real work end to end. Keep consequential decisions with people, log every action the agent takes, and compare the result against the old process rather than against a projection.
This applies when you can access the underlying data lawfully and can define what a correct outcome looks like. If either is unclear, fix that first — an agent will not resolve it for you.
- 01Pick the bottleneck
- 02Set the baseline
- 03Pilot safely
- 04Measure, then scale
The organizations that get value from agents are rarely the ones that deployed the most of them. They are the ones that picked a process they understood well, instrumented it properly, and were willing to stop if the numbers did not move.
Sources
Research referenced
Each link opens the original publication on the publisher’s own website. We have not reproduced any figure that is not stated in these sources.
- Capgemini Research Institute — Rise of agentic AI capgemini.com/insights/research-library/ai-agents/
- PwC — AI Agent Survey pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html
- Google Cloud — The ROI of generative AI cloud.google.com/resources/roi-of-generative-ai
- OpenAI — The state of enterprise AI openai.com/business/guides-and-resources/the-state-of-enterprise-ai-2025-report/
Figures quoted above are survey results and market projections published by the organizations named. They describe what respondents reported or what analysts estimate. They are not guarantees of financial return, and NovaTechAI makes no claim that any organization will achieve a comparable result. Revenue impact depends on the use case, data quality, controls, adoption and how thoroughly the underlying workflow is redesigned. NovaTechAI is not affiliated with, and this briefing is not endorsed by, any of the organizations cited.
Apply the briefing
Find the one workflow worth starting with.
We run short discovery sessions to identify which of your processes would repay an agentic approach — and which would not. Every solution we deliver includes hands-on training for your team.
experts@novatechai.site