Build, buy or blend:
The key to shaping your AI sales strategy
The expert guide to making the right call.
Your questions we sought to tackle
If you’re weighing up whether to build your own AI sales agent or partner with a provider, here are the strategic considerations that should guide your decision
AI agents require a full-time, cross-functional team, with a headcount of thirty being lean and efficient. You’ll need dedicated roles in machine learning, data engineering, analytics, software development, product design, data science, quality and compliance. These aren’t short-term contracts that can ramp down after launch – they’ll need to be retained long-term, to keep your agent competitive and compliant.
What do you want to own?
Building gives you full ownership over your data, brand experience and integration, but requires deep in-house AI expertise. Buying gives you immediate access to proven, high-performing AI models, but may limit customisation depending on your chosen partner. The question – especially for brands looking to do a mix of both – is what do you want (and have the skills) to own?
How fast do you need to move?
The longer it takes to get your AI sales agent to market, the more ground competitors can gain. Building in-house often means lead times of a year plus; hiring talent, prepping data, developing and testing – while the market keeps moving.
For telcos who want to lead and compete, buying can offer a clear speed advantage.
Do you have the right data and behavioural expertise?
The biggest differentiator isn’t your AI sales agent’s interface, it’s the sales psychology behind it. Building in-house means training on your own data, creating your own sales logic and product knowledge, which will be more targeted than off the shelf models. But telco-specialist vendors bring highly sophisticated models, shaped by years of industry-wide sales journeys and customer behaviour.
The right choice depends on the depth of your own data, and the level of sophistication you want to reach.
Can you guarantee your guardrails?
Building from scratch is a significant undertaking, and effective risk management can’t be an afterthought. AI sales agents need robust, pre-emptive guardrails – from bias detection and escalation handling to real-time monitoring and compliance alignment. Specialist partners offer solutions with stress-tested guardrails, designed to protect your business from day one.
The decision is whether you have the rigour and resources to build and maintain that resilience yourself, or whether you can put full trust in a partner.
Comparative analysis:
Build vs buy
Foundations for a sophisticated AI sales agent
Building a high-performing AI sales agent isn’t just a technical project, it’s a long-term operational commitment. To compete on sophistication, compliance and performance, telcos need a robust internal setup from day one – and ongoing investment to support it.
What it takes to build
A dedicated, cross-functional team
AI agents require a full-time, cross-functional team, with a headcount of thirty being lean and efficient. You’ll need dedicated roles in machine learning, data engineering, analytics, software development, product design, data science, quality and compliance. These aren’t short-term contracts that can ramp down after launch – they’ll need to be retained long-term, to keep your agent competitive and compliant.
A strong data foundation
There’s a significant gap between a chatbot that can answer questions and an AI sales agent that can close a deal – and that difference relies on your data. To build an agent that truly converts, you need deep, nuanced intelligence from a wide range of sources: from customer interactions and sales conversations, to the real-world experiences of your reps. Your agent needs to understand subtle behavioural cues, common objections (and how to combat them) and the motivations that go deeper than costs and features.
Real-time data pipelines
To keep your AI sales agent accurate and responsive, you’ll need robust data pipelines that can ingest product data, customer interactions, and CRM inputs in real time – then clean, transform, and route that data reliably across your systems. You’ll need specialised tooling for ingestion, transformation, and monitoring, as well as the cloud infrastructure needed to handle high-volume, real-time data streams.
A large language model you can build on
Choosing the right large language model (LLM) is a foundational decision, not just for performance, but for how you operate. Before your AI sales agent can be fine-tuned to handle complex sales scenarios, you’ll need to decide whether to build on a commercial or open-source model.
Commercial models (like those accessed via Azure or OpenAI) offer managed infrastructure, faster deployment, and easier scaling – ideal if you want to move quickly without taking on operational overhead. The trade-off is higher ongoing costs, with pay-per-access pricing.
Open source models offer more control and flexibility, and avoid per-request charges. But they require internal capability to self-manage and maintain the supporting infrastructure – from powerful GPU compute to a full machine learning platform. That means higher upfront investment and more hands-on responsibility.
With the current level of investment in commercial models, they tend to outperform most open source models – though that is constantly changing as new models are released. For many teams, commercial is a clear starting point, with the flexibility to switch later.
Model customisation
Regardless of which model you choose, a standard out-of-the-box implementation won’t be enough for telco sales. To move from a general-purpose chatbot to a sophisticated agent that mirrors your best human salespeople, you’ll need to invest in custom implementation. That includes:
Specialised conversation capabilities: Tailoring your agent to follow real-world sales flows, respond to objections, and recognise subtle buying signals.
Deep customisation: Aligning tone, behaviour, and logic with your brand and your commercial strategy.
Fine-tuning: Using your own data to refine the model’s performance and make it more relevant to your customers and use cases.
Robust prompting architecture: Structuring inputs and instructions to guide the model’s behaviour reliably and consistently.
Guardrails and safety layers: Ensuring your agent is accurate, compliant, and safe in high-stakes sales environments.
Without these layers, even the best model won’t be ready for customer-facing conversations – and certainly not for converting them.
A strong analytics layer
Ongoing optimisation relies on deep performance visibility. You’ll need advanced analytics tooling to track metrics like conversion, fallback rates and intent accuracy, plus the ability to feed that insight back into your fine-tuning loop.
Compliance and safety frameworks
AI sales agents need strong guardrails, especially in telco, where every sales interaction must follow strict rules on pricing, eligibility, disclosures, data usage and more. That means proactive compliance testing, content moderation, prompt protection, and security auditing.
Together, these are the foundations you’ll need to lay – and maintain – if you choose to build in-house. They’re not blockers, but they are non-negotiables if your goal is a sophisticated, compliant, high-performing AI sales agent. The real question isn’t just whether you can build the agent, it’s whether you’re prepared to invest in keeping it competitive and compliant long-term.
Foundations in place:
What vendor solutions offer out of the box
When you buy, the most complex and resource-heavy elements of running an AI sales agent are taken care of by the vendor. You gain access to a proven, telco-trained conversational agent, without needing to build and maintain the full ecosystem yourself.
Your internal team plays a focused role. A product owner typically manages the relationship, while your data team ensures clean, real-time pipelines between your systems and the vendor’s platform. Beyond that, the day-to-day orchestration and technical upkeep largely sit with the vendor.
They typically take care of:
Model development and tuning, using industry-wide sales data and behavioural patterns
Infrastructure, including GPU environments, orchestration layers, and the backend architecture needed to run large-scale AI agents efficiently
Analytics and iteration, with dashboards, intent tracking, and conversion metrics built in to support ongoing refinement
System integration can be achieved through either headed deployments, which prioritise speed and control, or headless, API-first architectures, offering greater flexibility to seamlessly embed AI across various touchpoints.
One of the biggest advantages – impossible to replicate with an in-house build – is the scale and diversity of industry data which vendors train their models with. While you have access to your own sales records and behaviour, vendors are fine-tuning models based on millions of interactions, over multiple years, across the whole of the telco space.
However, the convenience of a pre-built solution can come with trade-offs. You rely on the vendor’s architecture and update cycles, which may limit your ability to customise or control every aspect. Your data governance and compliance responsibilities remain, though vendors can be expected to provide built-in guardrails that evolve alongside industry regulations, reducing your internal burden. Crucially, subscription models create ongoing financial commitments that can grow over time, locking you into the vendor’s ecosystem and potentially reducing your flexibility to adapt or switch providers without significant cost or disruption.
“AI is still a space of experimentation, but significant investment is being made to solve real-world problems. The key is deciding between bounded and unbounded problems. Bounded problems are well-defined with clear solutions, like customer service or automated reporting. The unbounded frontier, however, tackles highly complex challenges - digital sales being a prime example. In today’s turbulent world, staying close to the customer is crucial. By understanding their needs, we can personalise interactions and engage more effectively. For telcos, the focus should be on proven solutions that deliver tangible results. Sooner or later, CFOs will start asking, "What’s the ROI?"
Ex Chief Strategy Officer, Verizon Consumer
Speed to market:
The competitive clock is ticking
Timing in an AI-driven market isn’t just about efficiency; it’s about being front of the pack. The first to boost conversion rates online. The first to scale new products through digital channels. The first to engage customers with an experience that feels personalised and human. Speed isn’t just a delivery metric, it’s a strategic lever. Here’s why it matters:
First-mover advantage
The first telco to deploy a personalised, conversational experience sets the customer expectation for everyone else. Being early means owning the benchmark.
Build slow, lose fast
Taking your time comes at a cost. While you’re slowly building, the market is quickly evolving, with first-movers learning from real-time feedback, tuning their experiences, and capturing early gains. What’s advanced when you start building can quickly fall behind in the time it takes to get live.
Opportunity cost
Telco customers often arrive on-site with high intent, but drop off due to friction or lack of guidance. AI sales agents engage that intent in real time, expertly guiding users through their needs, preferences and objections, to find an ideal fit. Every month without that support is a month of missed opportunities, conversions and revenue.
Customer experience gap
Continue Reading
Download the full report to explore the trade-offs between speed, control, compliance, and maintaining a competitive edge.
You’ll also gain access to our decision-making framework, designed to help guide your next steps with confidence.