We have been lying to ourselves since, well, basically since forever. We placed customer support agents into a padded room called the “contact center,” handed them a ticketing system, and told them to keep the angry people away from the rest of the business. We tracked average handle times; we cheered when a routing algorithm saved a fraction of a second; and we pretended that managing an interaction was the same thing as solving a problem. Deflecting an issue was the holy grail.

That era is over. The walls of the contact center have been blown wide open, and the debris is currently raining down on the CRM and operations landscapes. The market is shifting from asking the question “who can capture the ticket best?” to “who can actually resolve the problem fastest?” Which is an entirely different category of question. And far more meaningful.

And as Cameron Marsh from Nucleus Research so accurately pointed out in our recent CRMKonvo, that is a much nastier, much more complex place to compete.

TL;DR

If you want to watch the full CRMKonvo, please go ahead here (optimized for smartphones) or here (optimized for tablets/computers).

Else, be my guest and continue to read.

Or feel free to do both …

The Illusion of the “Smart Ticket”

Let’s just be absolutely clear from the start: nobody wants a ticket. A ticket is simply a formalized receipt of failure. It is documented proof that a product broke, a service failed, or a user interface was too clunky to navigate.

For years, many vendors, including specialists like Zendesk, Freshworks, and others have built success around making those tickets prettier, easier to handle and pass around. And companies like Five9, Genesys, Verint and too many more to count happily managed the interaction center. But when Salesforce forcefully entered the CCaaS conversation with new voice and AI capabilities, they fired a warning shot across the bow of every standalone service vendor. Salesforce is betting that because they own the underlying customer data and the core business workflows, they should own the resolution. They are not selling a better queue; they are selling an operations layer.

The response from the service-first vendors is equally bold. They argue that because they capture the initial intent and handle the front-line friction, they are ideally positioned to resolve the issue before it ever touches a core system.

But the time-honored truth is that buyers do not really care about these architectural turf wars. Customers do not care if your platform uses a sophisticated LLM or a massive RAG implementation. They care about one thing: when something breaks, who fixes it? If your AI simply routes a ticket more efficiently to a human who still has to manually process a refund across three legacy systems, your AI is nothing more than expensive window dressing, or as Cameron aptly put it, “automation with a better branding”.

Smelly Data and the LLM Mirage

Which brings us to the most uncomfortable reality of the current enterprise software hype cycle: the data problem. Every vendor is currently parading their AI Agents and “AI Studios” around like magic wands that will instantly get things done while vaporizing operational costs.

But as we discussed on the show, almost every organization suffers from what we fondly call smelly data. They have decades of poorly categorized customer records, duplicate entries, and contradictory knowledge base articles. Pointing a state-of-the-art LLM at a mountain of unwashed data does not give you artificial intelligence; it gives you a very fast, very confident artificial idiot.

Garbage in equals garbage out. It is a cliché because it is true. The organizations that are actually seeing a return on their AI investments are not the ones buying the flashiest tools. They are the ones putting in the hard and unglamorous work of normalizing their data models, cleaning up the data and integrating their core systems. If you do not have a pristine data foundation, your generative AI will simply generate more work for your human employees to clean up while chasing your customers down rabbit holes.

The “Suite vs. Best-of-Breed” Debate Reignited

For the last decade or so, the enterprise software pendulum swung heavily toward “best-of-breed.” We bought point solutions for every micro-problem and relied on brittle APIs to hold the Frankenstein monster together.

The trend slowed a few years and nowadays AI is starting to violently push the pendulum back toward the suite, or at least a platform. When you deploy an AI agent to resolve a customer issue, that agent needs instantaneous, read-and-write access to lots of data, often including billing, shipping, inventory, and customer history. Every single integration point is a potential point of failure. Every seam between different software vendors is a place where context gets lost and the AI starts to hallucinate.

As Cameron noted, the burden of proof has shifted. It is no longer up to the suite vendors to prove they are better. It is up to the best-of-breed vendors to prove that their necessary integrations will not slow down the AI or destroy the seamless execution of a workflow. If an integration creates lag or drops context, the ROI of the entire project collapses. The suite wins not because it has the best individual features, but because it has the fewest broken bridges while being good enough to do the job.

The True Predictor of Success: It Is Not the Tech

Here is the most interesting takeaway from Cameron and Nucleus Research evaluating dozens of successful AI implementations: the specific technology stack rarely predicts the success of the project. Whether a company chose Salesforce, Zendesk, or a bespoke solution, the common denominator of success was the human element.

The companies that succeed have a stellar relationship with their vendor’s Customer Success Management team. They admit what they do not know, they ask for help with workflow design, and they partner with their vendors rather than treating them like mere order-takers. Conversely, the CSM teams admit where their software might not be the one to go for. It’s a two-way street. The technology is just the engine; the vendor relationship is the steering wheel. If you buy a Ferrari but refuse to talk to the mechanic, you are going to crash.

Reality Check: Three Imperatives for Enterprise CX Buyers

If you are a CIO, CXO, or IT leader currently evaluating AI for your service operations, you are swimming in an ocean of marketing fluff. Vendors are promising to cut your headcount in half while doubling your customer satisfaction. To avoid making a catastrophic and expensive mistake, here are three critical learnings you must apply to your buying cycle today.

Stop Buying “Deflection” and Start Buying “Resolution”

Deflection is a vanity metric. Often, a high deflection rate simply means you have made your IVR or chatbot so incredibly frustrating to use that the customer gave up and went to a competitor. Do not reward vendors for preventing customers from reaching you.

Instead, force vendors to prove their “fully resolved” rate. Demand to see how their AI handles a complex workflow end-to-end without a human ever touching it. If the AI can only handle password resets and order status checks, it is just a basic automation script wearing a tux. You are paying an “ego tax” for a label. Demand actual resolution.

Clean Your Pipes Before You Buy the Pump

Do not spend a single dollar on an AI agent if you have not invested time and resources to clean your data. The most sophisticated LLM in the world cannot resolve a billing dispute if your billing data is housed in an on-premise server from 2012 that only updates in batches every 24 hours and doesn’t even offer an API.

You need to establish a rigorous knowledge management practice. You need clean, structured data and clear workflow documentation. If you skip this step, your AI implementation will fail, your ROI will evaporate, and you will spend the next two years blaming the software for a problem that you created on your own.

Calculate the True “Human-in-the-Loop” Maintenance Cost

Vendors love to show ROI models based on how many tier 1 support agents you can eliminate. What they conveniently leave out of the spreadsheet is the cost of the highly skilled engineers, data scientists, and workflow managers you will have to hire to babysit the AI.

An AI system requires constant tuning, monitoring for hallucinations, management of drift, and edge-case management. You might eliminate fifty low-cost roles, but you will need to hire five very expensive experts to maintain the system. The need for humans in the loop further reduces the systems autonomy and hence “efficiency”. If your ROI calculation does not account for this shift in personnel costs, you are presenting a work of fiction to your board. Focus on the payback period, don’t forget about ongoing maintenance costs, and never buy a black box that your own team cannot audit.