Table of Contents
Introduction
This is an end of year review on the fast-moving transformation of organisations being enabled by Agentic AI.
It is based on an analysis of major reports and research that emerged during 2025. A reference list is provided at the end if you want greater detail. This article focusses on customer contact. An organisation wide version of the review will appear in a follow up article for those with broader interests.
For now, the headline is that Agentic AI is already reshaping the way organisations sell to, serve and communicate with customers. Even though in 2025, most contact leaders are still stuck in a gap between impressive demos and reliable, scaled impact.
The research base on agentic AI this year is clear: customer contact is one of the earliest and richest arenas for value. Yet it is also where organisational, data and trust challenges show up most sharply.
Even though we now getting used to the proposition that AI can both generate and act, it is the degree to which organisational readiness is able to support degrees of autonomy and decision making that remains challenging and will likely remain so for the rest of the decade.
The fact that we continue to witness accelerated timelines on the technology required to fulfil the promise at the heart of Agentic AI should not be read as a deployment green light.
For the immediate future, it’s going to be organisational readiness that determines velocity of change and ROI rather than just ‘better AI’. Something all the research points out as crucial.
While new exciting functionality such as Nano Banana Pro is both great fun and transformative within individual hands, it requires workflow, governance, reskilling and all the other dimensions of organisational readiness to make it safe and scalable in an organisational context.
Bearing these points in mind, what’s been happening in customer contact in 2025?
From Chatbots to Digital Colleagues
For customer contact, the crucial shift is not from “no AI” to “AI”, but from static tools to agents that can plan and act on a customer’s behalf across multiple steps. Traditional rule‑based chatbots answered only what they had been explicitly scripted for, and retrieval‑augmented (RAG) assistants made those answers more natural but still fundamentally reactive.
The research on agentic AI describes systems that can interpret intent, decide which tools and systems to call, execute a sequence of actions and then decide whether to resolve or escalate. All within a single customer interaction.
In contact centre and customer service contexts, this means moving from “FAQ bots” to genuine case‑handling agents that can gather information, validate identity, check entitlements, trigger refunds or bookings and update back‑office systems before coming back to the customer with a final outcome.
In sales and marketing, agents are described as orchestrators of the end‑to‑end journey: spotting intent, assembling tailored propositions, scheduling follow‑ups, nudging customers at the right moment and updating CRMs. Without any need for human involvement beyond oversight on quality and significant decisions.
One of the many reports in the review captures this emerging dynamic in AI-Human partnering: agents combine predictive and generative AI “in the flow of work” across the full value chain, rather than sitting on the side as a clever tool.
Also what makes this evolution particularly important for customer leaders is that it aligns with how customers already experience brands.
Consumers do not think in terms of channels or functions. Instead, they experience a fragmented set of promises, processes and people. Agentic AI, done well, offers the possibility of stitching those fragments into something that feels like a coherent relationship rather than a series of disconnected transactions.
Where Agents Are Actually Deployed in Customer Contact
The broad AI adoption data shows that customer‑facing functions are no longer on the sidelines.
In one global survey of almost 2,000 organisations, service operations and marketing and sales both appear among the top ten functions where AI agents are already being piloted or scaled, even though penetration still lags behind IT and manufacturing.
Only a few percent of respondents report agentic AI “scaled or fully scaled” in service operations and in marketing and sales, but the direction of travel is clear.
A separate enterprise‑value study shows that in 2025, around 70% of all AI value is being realised in core business activities, with digital marketing, customer journeys and core customer service together accounting for a meaningful share of that pie.
Within that research, “core customer service” alone contributes close to a tenth of total AI value, with digital marketing and customer‑journey initiatives adding another significant tranche.
Crucially, the same study finds that agents already account for about 17% of total AI value today. This is expected to approach 30% by 2028 with customer‑facing workflows highlighted as a major driver of that growth. This is the current sweet spot in terms of ROI and organisational readiness. A fuller explanation can be found in this previous Brainfood article.
The review also brings together a set of concrete deployments that show how far beyond simple chatbots the leading organisations have already moved in service, sales and marketing contexts.
For example, a global beauty company has built a virtual beauty assistant that guides customers across more than 20 markets and multiple brands, combining consultation, product education and personalised recommendations inside a unified digital journey.
Deployed in under a year on a unified data and cloud platform, this agentic experience is expected to generate around $100m in incremental revenue and to double the ROI of traditional e‑commerce pathways by turning static product pages into conversational journeys.
In the utilities sector, a European provider has rolled out a multimodal AI assistant to millions of customers, handling routine queries, account changes and simple troubleshooting across channels.
The reported outcomes include significantly reduced average handling times, faster responses and a higher proportion of enquiries resolved without human intervention while boosting satisfaction scores.
And in banking, one well‑known virtual assistant now handles more than a million customer interactions daily, providing spending insights, flagging recurring charges and supporting simple service requests, with measured benefits in both cost‑to‑serve and cross‑sell uplift.
Behind these headline examples sit thousands of less glamorous but no less important micro‑workflows where agents are quietly orchestrating customer contact: generating first‑draft replies, summarising long histories, checking policies and proposing next actions for human agents to approve.
The pattern is consistent: where organisations have moved past experimentation, they are embedding agents not as standalone channels, but as co‑workers inside every step of the customer lifecycle.
How Agents Change Service, Sales and Marketing
The collective research is helpful in distinguishing three distinct but overlapping zones in customer contact.
Service: from tickets to outcomes
In service, agentic AI changes the unit of work from “interactions” to “resolved outcomes”.
Instead of counting how many chats or calls a bot can deflect, leading organisations are designing agents that own an entire service goal: restart failed billing, rebook a journey, diagnose a product issue, update a contract.
One automotive manufacturer deployed an internal agentic ticketing solution that engages colleagues, collects diagnostic details then automatically creates and routes IT tickets for issues such as password resets and access problems. Metrics such as cost per ticket, automation rate, first‑contact resolution and policy‑compliant escalation are tracked to show both efficiency gains and improved experience.
A government case study in the same research shows how a citizen‑facing agent can take on the burden of navigating complex information on behalf of the user.
The agent continuously ingests legislation and service documentation in multiple formats, then returns clear, consistent, privacy‑compliant answers to questions that would previously have required manual research or long contact‑centre calls.
Here, business value is measured in cost‑per‑query, reduction in manual content curation and reduced time for citizens to access critical information, but the bigger story is trust: citizens get a single, dependable “front door” into the state.
Sales: from leads to live pipelines
On the sales side, the review highlights agents that act as persistent pipeline managers rather than occasional helpers.
In some deployments, agents automatically extract and update opportunity data from emails, meeting transcripts and documents. In others, they qualify inbound leads, schedule follow‑ups and nudge salespeople when deals are at risk or when a competitor move demands a response.
A major CRM provider, for example, has introduced agentic capabilities that can draft outreach, schedule meetings, send follow‑up messages and even perform lead qualification steps, with early adopters reporting double‑digit uplifts in sales and shorter cycle times.
The broader AI value research supports this direction. It shows that marketing and sales functions are among those most likely to report revenue gains from AI, with around two‑thirds of respondents in some surveys citing increased revenue and improved pricing or product‑mix outcomes where AI is embedded deeply into customer‑facing workflows.
Although much of that value still comes from analytics and recommendation engines rather than full agents, the trajectory is towards more autonomy. Virtual colleagues that keep the pipeline clean, keep the team honest and keep the customer experience coherent even when human attention is fragmented.
Marketing: from campaigns to continuous conversations
In marketing, the shift is away from campaign calendars and towards persistent, personalised dialogue. Major consumer brands are now using multimodal agents to watch browsing behaviour, purchase history and engagement signals in order to dynamically assemble recommendations, offers and content across channels.
Amazon’s long‑standing personalisation engines have been re‑framed as a form of orchestrated agentic workflow: analysing behaviour, predicting demand, nudging customers and coordinating fulfilment in a largely autonomous loop, contributing to substantial uplifts in sales and loyalty.
Another example features a large media and subscription player that uses agents to recommend content and orchestrate notifications across platforms, increasing retention through more relevant and timely suggestions.
In each case, the agent does not just generate copy; it manages a multi‑step engagement: select segment, choose the next best content, time the contact, send via the right channel, observe the response, update the profile and adjust the plan.
This is classic agentic behaviour. It changes the nature of marketing work from deciding every campaign manually to designing and supervising a system that is always learning and responding.
What The High Performers In Customer Contact Are Doing Differently
A central theme across the research is that a small group of “high performers” or “future‑built” organisations are pulling away from the pack, and customer contact is one of the clearest arenas where this divergence shows.
These organisations are not necessarily the ones with the biggest technology budgets. But they are the ones treating agentic AI as a lever to redesign how customers experience the brand and how humans and digital colleagues share the work.
First, they are more ambitious.
High‑performer segments in the surveys are several times more likely than others to say they are using AI to fundamentally reshape customer journeys, not just to shave a few seconds off average handling time.
They are also more likely to set growth and innovation objectives for AI initiatives alongside efficiency targets, rather than viewing AI purely as a cost‑reduction opportunity.
In customer contact, that translates into agents that are designed to improve satisfaction, loyalty and revenue as well as cost, right from the pilot stage.
Second, they rewire workflows around agents instead of inserting agents into old processes.
For instance, redesigning end‑to‑end flows where agents are accountable for outcomes and human employees become supervisors, exception‑handlers and designers of better work, not just extra hands when the technology fails.
In contact centres that might mean moving from flat skill‑based routing to an architecture where a triage agent handles simple queries, a specialist agent prepares case context and drafts, and a human agent steps in only when judgement, negotiation or deep empathy is truly required.
Third, they invest heavily in measurement and governance from the start.
Future‑built companies are several times more likely to track AI value systematically, to have clear guardrails and to operate with shared ownership of AI between business and IT.
In the customer domain, that means monitoring not only cost per contact and containment, but also error rates, escalation quality, customer trust and the distribution of work between human and digital labour. Back to an initial point on readiness, autonomy is increased only when evidence shows that the agent is safe, reliable and beneficial for both customers and colleagues.
Underpinning all this is a people‑first approach.
The research is blunt that most roadblocks to AI value are about people, organisation and process not algorithms. In particular, it suggests that customer‑facing functions already under pressure from burnout and attrition are particularly sensitive to mis‑steps.
Designing An Agentic Operating Model for Customer Contact
The review material offers practical guidance for contact leaders who want to move beyond pilots and proofs‑of‑concept. It points to four interlocking design questions: where to start, how to build, how to measure and how to bring people with you.
Where to start: use cases that matter
Organisations should map potential agentic applications across two axes: business value and complexity.
For customer contact, low‑to‑medium complexity, high‑value candidates often include billing enquiries, simple product questions, order‑status updates, basic troubleshooting and routine account changes. All areas with clear rules, good data and high volumes.
On the sales and marketing side, common early wins include lead‑qualification agents, follow‑up orchestration, personalised content assembly and campaign‑insight agents that summarise performance and suggest optimisations.
The key is to choose workflows that are meaningful enough to matter but constrained enough that you can manage the risks and gather evidence.
The research warns against both extremes: trivial proofs‑of‑concept that never translate into scaled value, and over‑ambitious, highly complex journeys that over‑reach the organisation’s current data, governance and change‑management capabilities.
How to build: patterns, not one‑offs
On the technical side, treat agentic AI as an ecosystem rather than a string of isolated bots. This mirrors the operational CX goal of designing for end-to-end journeys rather than isolated touchpoints. AI is most effective turning data into actionable insights, decisions and execution when operating in a unified space/ecosystem.
For customer contact, that means investing in a small set of reusable patterns: a triage agent template, a case‑summarisation agent, a recommendation agent, a knowledge‑research agent, an outbound‑engagement agent and so on, all connected to a shared data and identity backbone.
The same research highlights three architectural elements that matter disproportionately for contact scenarios:
- Multi‑agent collaboration so that different agents can specialise and coordinate
- Human‑in‑the‑loop checkpoints(so that risky or ambiguous actions are escalated
- Robust memory management so that conversations and journeys feel continuous across channels and time
In the context of building out ecosystems, new(ish) interoperability standards such as MCP and A2A become important enablers by reducing the need for brittle bespoke integrations between agents, CRMs, ticketing tools and knowledge bases.
How to measure: beyond cost‑per‑contact
Measuring agentic enabled outcomes matters. One recommendation is to build an “agent scorecard” across nine operational dimensions, many of which map directly to familiar CX metrics but applied at the workflow level: outcome quality, cost and efficiency, autonomy, accuracy, safety and compliance, adaptivity, flexibility, scalability and adoption and trust.
For a customer‑service agent handling billing disputes, for example, that might mean tracking not only time‑to‑resolve and cost‑per‑case, but also error rates, policy‑violation rates, the proportion of cases escalated appropriately, the agent’s ability to cope with changes in billing rules and the degree to which human agents rely on or override its suggestions.
Strategic value is then rolled up across four buckets (direct, indirect, opportunity and efficiency) so that boards can see how a portfolio of customer agents is changing cost‑to‑serve, revenue, risk and customer outcomes over time.
How to bring people with you: teammates, not tools
Finally, the organisational sections of the review stress the importance of communication and co‑creation in customer‑facing teams.
Colleagues need to see concrete examples of how their roles will change. What tasks will be automated, what new responsibilities will emerge, how will success be measured? All require honest discussions about both opportunities and trade‑offs.
The most effective narratives position agents as partners that remove drudgery and expand human impact, not as silent threats waiting in the wings But of course this must reflect the truth of your own AI-Human partnering strategy.
If so, then co-design is the way forward involving experienced advisors, team leaders and marketers in the design of workflows, testing early versions and establishing guardrails.
It also means giving them the skills and time to work effectively with digital colleagues. This means understanding capabilities and limitations. Knowing when to override or escalate, and learning how to interpret and improve agent behaviour through feedback.
The research is clear that organisations which under‑invest in this human side of change see adoption stall. Even when the technology performs well in the lab. A point made earlier.
The 2026 Agenda for Customer Contact Leaders
Looking ahead from this 2025 evidence base, agentic AI in customer contact feels less like a fringe experiment and more like the next operating model for sales, service and marketing.
The technology trends analysis positions agentic AI as one of the fastest‑growing frontiers, with rapid innovation in multi‑step reasoning, deep‑research agents and agent‑to‑agent communication, even though most enterprises are still officially in an “experimentation” adoption phase.
At the same time, the value gap research shows that a small minority of pioneering organisations are already industrialising agentic workflows and pulling ahead on both CX and economics.
For contact leaders, the agenda is therefore less about whether agents will matter, and more about how to make deliberate choices in four areas that the research surfaces as decisive.
Scope and ambition: Will agents be confined to cost‑cutting in narrow slices of service, or will they be used to rethink the experience across marketing, sales and service as a connected system?
Operating model: Will they sit at the periphery of contact operations, or will they be integrated into team structures, metrics and career paths as standard co‑workers?
Governance and trust: Will organisations wait for regulators and incidents to force clarity on autonomy, escalation and accountability? Or will they invest early in clear, human understandable guardrails?
Capability building: Will upskilling be limited to a handful of specialists, or will front‑line teams be given the literacy, tools and time to become active shapers of how agentic AI is used in their work?
The research is unambiguous on one final point
Most of what differentiates leaders from laggards in agentic AI has very little to do with the cleverness of the models.
In customer contact especially, success is being written in the design of journeys, in the quality of collaboration between humans and machines and in the willingness of leaders to treat agents not as a gadget but as a catalyst to rethink how customers and organisations meet.
Final Word
If you found this a useful, evidence-based summary of what’s been happening, why not ask Brainfood for a customised discussion we can facilitate with you and your colleagues. This would built from a fuller analysis of this year’s commercial and academic research surfacing the latest insights and best practice on Agentic AI that matter in your context.
Index Of Referenced Reports – Agentic AI’s Impact on Customer Contact in 2025: end of year review
- The State of AI in 2025: Agents, Innovation, and Transformation – McKinsey & Company / QuantumBlack, AI by McKinsey; global survey of 1,993 participants on AI and agentic adoption, value and organisational practices.
- The State of Agentic AI in 2025 (PwC‑based synthesis) – Comprehensive review of agentic AI capabilities, cross‑sector use cases and the “service‑as‑a‑software” model including customer‑facing examples such as Amazon, Bank of America, Netflix and others.
- The Widening AI Value Gap – Build for the Future 2025 – Boston Consulting Group; empirical study of AI maturity and value realisation across 1,250 CxOs, including quantification of agentic AI’s share of AI value and distribution across functions such as digital marketing, customer journeys and core customer service.
- Technology Trends Outlook 2025 – Agentic AI Section – McKinsey Global Institute; places agentic AI among the fastest‑growing technology trends, with data on job‑posting growth, equity investment and early enterprise applications.
- Unlocking the Right Agentic AI Use Cases – Deloitte India; introduces value‑versus‑complexity and three‑tier complexity models for selecting and sequencing agentic AI use cases, including customer‑facing scenarios.
- Evaluating Agentic AI Technology Readiness – The Business Imperative for Agentic AI – Deloitte India; details ecosystem architecture, organisational readiness profiles and communication standards (MCP, A2A, ACP) relevant to enterprise‑grade, customer‑facing deployments.
- Defining and Measuring the Success of Agentic AI – Deloitte India; proposes a dual‑lens (operational and strategic) measurement framework with nine operational dimensions and four value buckets, illustrated with customer‑service and citizen‑service case studies.
- The State of Generative AI in the Enterprise – Deloitte; provides statistics on enterprise exploration of agentic AI, multi‑agent systems and autonomous agents, with particular reference to automation in service and contact functions.
- Sector and Use‑Case Compendium within the PwC and Deloitte syntheses – Consolidated case studies of agentic AI across sectors, including virtual beauty assistants, multimodal customer‑service agents, subscription‑media recommender agents and CRM‑embedded sales agents.
- Function‑Level AI Value Distribution Analysis – Detailed tables from BCG’s value‑gap report quantifying AI’s contribution across R&D, digital marketing, consumer journeys, sales, core customer service and support functions.
