Beyond the AI Hype: What Organisations Need to Get Right Before They Build
In my previous article, Just Because AI Can Do Something, Doesn’t Mean It Should, I explored a question I believe more organisations need to be asking as we move further into the world of Agentic AI.
We’ve become incredibly focused on what AI is capable of delivering, from automating customer interactions to orchestrating entire business processes; but capability alone doesn’t guarantee business value, and deploying AI doesn’t automatically mean we’re improving the customer experience. That got me thinking about the conversations happening before these projects even begin.
Are organisations genuinely ready for AI?
Do they understand the processes they’re looking to automate?
Are they considering the long-term operational costs, governance and complexities of running AI at scale?
Or are some simply racing towards implementation because the technology is there?
To explore this further, this week i caught up with Amit Namjosji, Executive Director, TTEC Digital
to share their perspective on what successful AI adoption actually looks like beyond the demonstrations, proof of concepts and initial excitement.
We’re getting into the practical realities of AI delivery, from organisational readiness and common implementation mistakes to token consumption, cost management and the decisions that can make or break an AI investment.
The Expert Q&A …..
Where do you see organisations going wrong when they first start exploring AI?
Kelly : There’s enormous pressure on businesses to adopt AI, but are organisations sometimes jumping straight into technology and use cases before properly understanding the business problem they’re trying to solve? What are the most common mistakes you’re seeing?
Amit : The first mistake usually is when organisations start with an AI use case instead of a business problem. AI hasn’t changed the Objective of what we want. The aim is still very much is the problem. But that first incorrect question becomes, “Where can we deploy an agent?” rather than “What outcome are we trying to change, and why isn't the current process achieving it?”. That distinction matters. If a process is poorly understood, fragmented or simply inefficient, adding AI can automate the fragmentation rather than remove it. An example we came across with one of our projects involved a public-sector organisation interested in AI-driven self-service and intelligent voice routing. But its public content was not considered AI-ready, data quality was a concern, and processes were still evolving. Those are precisely the issues to address before deciding how sophisticated the AI should be.
The first question has to be what can the business do better.
What should organisations have in place before an AI project even begins?
Kelly : We talk a lot about AI readiness, but what does that actually mean in practice? From data quality and process maturity to governance, architecture and stakeholder alignment, what foundations should businesses be assessing before committing to implementation?
Amit : The five pillars: Objective (outcome), Process, Data, Governance and the overall operating model that binds it all together. There is a phrase we've been using for any CCaaS implementation which captures the distinction well: Governance determines whether an agent may operate. Data readiness determines whether it can operate reliably. The operating model determines who owns the outcome when it does.
AI readiness isn’t about technology; it's about enterprise readiness for all organisations, big or small.
How can organisations distinguish between a genuinely valuable AI use case and one that simply looks impressive in a demonstration?
Kelly : Proofs of concept can be incredibly compelling, but translating that into a reliable production solution is another challenge entirely. How should organisations assess feasibility, business value, risk and return on investment before deciding which opportunities to pursue?
Amit : Answer the fundamental questions, and you have the answer: ‘What problem are we solving?’ There are multiple ways in which you can approach this; we often use the RICER framework, but that is not the only one.
REACH - The Reach of users affected by the AI change matters. How many people will indirectly or directly be affected by the change? The first step is identifying which segment of the user base or internal teams will benefit from the AI initiative.
IMPACT -The Impact of the change on a scale. Measuring impact is key. Estimate the potential value or improvement that the AI initiative can bring
CONFIDENCE - is defined by the measure of success and the guarantee of outcome. Analyses previous similar implementations, data accuracy, and team expertise. Higher confidence levels should elevate the project’s priority. This is measured against identified possible scenarios
EFFORT- required to implement the AI business scenario
REVIEW - the consolidated score to prioritise which of the handful of AI initiatives are right for execution
AI doesn't always have to replace something dramatic to create value.
In one of TTEC Digital’s UK public-sector implementations, a modernised Microsoft contact-centre platform recorded 29,000+ AI-handled interactions in its first four months, alongside 24/7 digital accessibility. That's a useful operational result because the technology is attached to a genuine service transformation rather than being a standalone experiment.
Token consumption is something that doesn’t always feature heavily in early AI conversations. How significant can its impact become?
Kelly : As organisations move towards more complex AI agents, multi-step workflows and higher volumes of interactions, how does token consumption influence operational costs, performance and scalability? What should businesses understand about consumption models and the potential for costs to grow before they deploy AI at scale?
Amit : Understanding tokens enables better business cases, model selection, budget controls and solution design. So this isn’t just about ROI. The way to look at token consumption is as an additional variable cost unit.
My recommendation is to measure AI almost like a transaction platform: cost per successful outcome, not simply cost per token. Tokens are the raw material. The business buys outcomes.
A cheaper interaction that fails and creates a second contact isn't necessarily cheaper. Equally, an apparently expensive autonomous interaction may be excellent economics if it safely completes something that previously required several systems and significant human effort.
What are the biggest technical and operational challenges when moving AI from a proof of concept into a live business environment?
Kelly : An AI solution might work brilliantly in a controlled demonstration, but real-world environments introduce complexity. Where do you typically see challenges emerge around integration, security, accuracy, monitoring and ongoing maintenance? And how can organisations avoid creating solutions that become difficult or expensive to manage?
Amit : I think this question is changing. Historically, we talked about the journey from PoC to production. With agentic AI and the Forward Deployed Engineering (FDE) model, I think the real journey is from experiment to a continuously evolving operating capability. FDE puts engineering expertise alongside the customer to co-design, deploy and continuously improve AI systems against measurable business outcomes. The emphasis is explicitly on real-world deployment rather than isolated pilots
From what we're seeing with customers, the difficult part isn't necessarily building the first agent. It's creating the environment in which the 50th or 100th agent can operate safely and economically. Microsoft describes the evolution towards the Frontier Firm as moving from humans using AI assistants to human-agent teams and ultimately humans leading teams of increasingly autonomous agents.
Perhaps the biggest mistake is thinking "production" is the finish line.
With agents, production is where the learning really starts. We are now in the age where we're not just putting AI into existing processes. We're creating organisations capable of continuously redesigning those processes around humans and agents working together.
If you could give one piece of advice to a business about to invest in AI, what would it be?
Kelly : Looking beyond the hype, what should leadership teams be prioritising to ensure their AI investments deliver sustainable business value, rather than simply adding another layer of technology to an already complex organisation?
Amit : AI shouldn't become another technology estate sitting beside CRM, contact centre, data and workflow platforms. The real opportunity is to reconsider how work should happen when humans, deterministic automation and AI agents can each perform different parts of a process.
I would therefore start with one meaningful business journey, give it a measurable baseline, establish the human/AI boundary, identify the data and controls required, and prove the economics end-to-end. Then scale what works.
That is also why, in a complex airline transformation we're working on in EMEA, the recommendation was not to recreate the entire legacy environment in one leap. The proposed approach is an evidence-led assessment and migration blueprint that validates requirements, risks and migration waves before committing to the implementation.
I would recommend developing a business strategy that assumes AI exists that would align people, process, data, and governance.
One of the things I find most interesting about AI is that the technology is advancing faster than many organisations can adapt their processes, operating models and governance around it.
And while the possibilities are genuinely exciting, I think there’s an important distinction between deploying AI successfully and simply deploying AI.
The real opportunity isn’t necessarily in building the most sophisticated agent or automating the greatest number of interactions. It’s in understanding where AI can create meaningful value, ensuring the foundations are right and recognising the long-term responsibility that comes with introducing intelligent systems into business operations.
Because perhaps the question organisations should be asking isn’t just “What can we build with AI?” It’s “Are we actually ready to build it, and will it deliver the value we’re expecting?”
