Chapter 12 — Strategic Vendor Selection and Implementation

Chapter 12 walkthrough from The AI Contact Center Handbook by Sho Shimoda. Available on Amazon.

The fourteenth floor, three demos, one blank room

The conference room is on the fourteenth floor of a downtown hotel in Chicago, and it has been laid out in the way these rooms always are: a horseshoe of tables covered in white cloths, a projector at the head, water pitchers every six seats, and the specific brand of hotel coffee that never tastes like coffee but is always the correct temperature. Seventeen people are in the room. Six are the customer — the CIO, the head of CX, the head of contact center operations, the head of procurement, the head of information security, and a project manager assigned to run the RFP end-to-end. The other eleven are vendors, arriving in staggered ninety-minute slots.

Vendor number one is a pure-play CCaaS company. Their solutions engineer is very polished. He shows a demo of an AI agent handling a mock customer conversation about a wireless bill dispute. The demo goes well.

Vendor number two, ninety minutes later, is a hyperscaler. Their solutions engineer is also very polished. She shows a demo of an AI agent handling a mock customer conversation about a wireless bill dispute. The demo goes well.

Vendor number three, ninety minutes after that, is a CRM incumbent extending into CCaaS. Their solutions engineer is likewise polished. He shows a demo of an AI agent handling a mock customer conversation about a wireless bill dispute. The demo goes well.

At 4:45 in the afternoon, after all three vendors have packed up and left, the six customer people are still in the room. They are quiet. The head of CX finally says what everyone is thinking: "The three demos were almost identical. We just spent a day on a bake-off that told us nothing."

The CIO, who has been through this before, says: "That's because the demos are the wrong thing to evaluate. Every one of them can do the demo. What we actually need to know is what happens at month twelve, when we're two-thirds of the way through the migration and half the plan has gone sideways. And nobody demos that."

The vendor selection decision is not really a technology decision. It is a decision about which company you trust to be on the other end of the phone at 2 AM on a Sunday in year three of a five-year contract, and no PowerPoint deck answers that question.

12.1 The threat map: four camps competing for every deal

Before you can pick a vendor, you have to know who is in the market. The 2026 landscape has four distinct camps, and they are competing with each other in ways that were not true even three years ago.

The hyperscalers. Amazon Connect powers major deployments at Capital One, John Hancock, and (in 2025) the customer service operation of GE Appliances. Its pay-per-use pricing is genuinely different from the incumbents, and for spiky or seasonal operations it can be dramatically cheaper. It is less feature-complete than Genesys or NICE on workforce management, though the gap is narrowing. Microsoft's play is subtler — Dynamics 365 Customer Service plus Copilot for Service is pitched to any enterprise already running Microsoft 365 as the office suite, Teams as the collaboration platform, and Dynamics as the CRM. The integration story is compelling if you are all-in on Microsoft. Less so if you are not.

The CPaaS players. Twilio Flex is a viable option for enterprises with strong engineering teams and specific requirements that no off-the-shelf platform satisfies. It is a bad option for enterprises whose IT function is small, offshore, or already stretched.

The CRM heavyweights. Salesforce is aggressive here. The 2024 launch of Agentforce, followed by rapid customer acquisition through 2025, put Service Cloud in nearly every large CCaaS evaluation. The integration between CRM data and AI agent is genuinely tight. The counter-argument is cost — Salesforce's per-user CX pricing is high, and TCO on large deployments can exceed the CCaaS incumbents by 40 percent or more. Zendesk targets 50–500 seat operations with a pricing model designed for companies that would find Salesforce or Genesys too expensive.

The pure-play CCaaS incumbents. Genesys Cloud is the market leader by revenue and deployed seats. Five9 is number two in North America. NICE CXone has arguably the strongest quality management and workforce engagement suite. Talkdesk is the younger cloud-native alternative growing in mid-market. Content Guru has quietly won large European government and utility deployments. 8x8 combines UCaaS and CCaaS into one platform.

Each camp is now competing directly for every deal above a certain size. The threat map is not just which vendor do we pick — it is which category of vendor do we pick, and what are we implicitly buying into by picking that category.

12.2 The phased rollout playbook

Then comes the implementation. This is where the money goes, the reputations get made and broken, and the difference between the successful projects and the disasters is largely visible in the first quarter.

The single most important principle: do not do a Big Bang cutover. Never. Under any circumstances. Every horror story in the industry — and there are many — starts with a leadership team that decided, against the advice of nearly everyone with operational experience, that the migration would happen on a single weekend, in a single move, with the old system turned off on Friday and the new one turned on Monday.

In plain English: A cutover is the specific moment when live customer traffic stops flowing through the old system and starts flowing through the new one. A Big Bang cutover attempts that switch for the entire operation at once, on one weekend; a phased cutover does it for one queue, one region, or one customer segment at a time, over months, so that when something breaks it breaks in a small enough slice of the business to fix.

Two composite patterns from the industry are worth remembering. In the first, one large multinational's CX migration attempted a global cutover in a single weekend, ran into data-quality and integration problems the RFP had not surfaced, and ended up rolling back three months later — on top of an eighteen-month remediation project. In the second, another major retailer's system-wide modernization went through three postponed launches before the executive team switched to a much more gradual, store-by-store rollout after the initial pilots exposed data-migration issues at scale. Both organizations eventually got to modern platforms. The path was longer, more expensive, and considerably more painful than a phased rollout would have been from the start.

Phase 1 — Foundation (months 1–3). Stand up the platform in a non-production environment. Wire it to identity, warehouse, and one or two representative back-end systems (not all). Migrate a small, non-critical queue as pilot. Nobody outside the project team should notice.

Phase 2 — Parallel operation (months 3–6). Add the second queue. Run it in parallel with legacy for a defined period. Compare metrics. Fix surprises. Success criteria: new platform handling real customers, metrics at parity, team comfortable enough to advocate for expansion.

Phase 3 — Departmental cutover (months 6–9). Migrate the largest queue or largest customer segment. This is where the political conversation gets real, because you are now taking scale-critical operations off the legacy system.

Phase 4 — Full migration (months 9–12). The remaining queues, locations, segments. By this point the project team has been through three cycles of surprise, then fix, and the fourth should be smaller than the first three.

Phase 5 — Optimization (months 12–18). Migration done; optimization begins. AI features disabled during migration get turned on. WFM retunes. QA models retrain on new data. This is where the real business benefits show up — because the migration itself typically returns the operation to parity, and the AI-enabled optimization is what produces the compounding gains.

12.3 Contract terms that actually matter

The contract negotiation phase is where most enterprises give away leverage they could have kept. The reason is that negotiation is usually run by procurement and legal, working from a template, while the operations team — who understands what actually matters — is not in the room.

A 3 percent escalator sounds reasonable. Over a five-year contract, it compounds to about 16 percent total, which changes the economics of the deal materially. The AI-specific provisions did not exist five years ago and matter now: who owns the model outputs from AI agents deployed in your instance, who is liable if the AI makes a decision that harms a customer, is your customer data used to train the vendor's base models (yes is a red flag), does the vendor have the right to change the underlying model version without notice, and if the new version regresses on your specific use case, who bears the remediation cost.

Change of control matters more than it looks. The CCaaS market has consolidated repeatedly — Nice acquired inContact, Genesys acquired Interactive Intelligence, Cisco acquired BroadSoft — and there is no reason to think the next round of consolidation is done. Your contract should give you the right to terminate without penalty if the vendor undergoes a change of control that materially affects the product or the support commitments.

12.4 The shift from per-seat to outcome-based pricing

The pricing model itself is changing, and 2026 is roughly the inflection point. The traditional model is per-seat, per-month — somewhere between $75 and $200 per seat depending on vendor, bundle, and negotiation. If seat count doubles, bill doubles. If seat count stays flat but AI handles half the volume, bill stays flat. You can already see the problem.

The negotiation question is not which pricing model is best? It is which pricing model best matches how we expect to use the platform over the next three years? A company expecting significant seat reduction and heavy AI usage should push hard for usage-based or hybrid. A company expecting stable seat counts should hold onto per-seat while negotiating aggressive volume discounts.

12.5 The 12-month implementation calendar

Here is what the calendar looks like at the level of specific things that go right and specific things that go wrong. Every month has both.

That last one is the most consistent pattern in the industry. Companies budget for the migration and do not budget for the optimization work that follows it. The result is a modern platform running at legacy-level performance because nobody is tuning it. Do not make that mistake. The optimization phase is where the business case actually pays out.

The one-line takeaway: The calendar is not a schedule — it is a commitment device that forces the executive team to keep making real decisions on a monthly cadence for a full year, when every instinct will be to declare victory at month six and move on.

What to take with you from Chapter 12 — and from the whole book

The vendor selection decision is a decision about which camp of the market you are joining as much as it is a decision about which specific company you are picking. Hyperscalers, CPaaS players, CRM heavyweights, and pure-play CCaaS incumbents all now compete for the same deals, but they bring very different assumptions about how the contact center should work and where the product roadmap goes next.

The implementation, once you have picked, is a phased exercise. Big Bang cutovers fail, publicly, every time. The playbook that works is a staged migration over twelve months, followed by twelve to eighteen months of optimization, with clear go/no-go criteria at each phase. The contract is where you either preserve or surrender leverage for the next five years. SLA teeth, exit clauses, data portability, indexation, AI-specific provisions, and change-of-control protections all matter more than the ones procurement usually spends time on. And the pricing model itself is shifting — from pure per-seat toward usage-based, outcome-based, and hybrid models that better reflect where the value is created in an AI-enabled operation.

This is the last chapter. Twelve chapters ago, the book opened on the legacy trap — the ten- and fifteen-year-old on-premise systems that turn every improvement into a change-order. From there it moved through the front line, the back office, the compliance and security layer, and finally, in this part, the execution — the trust architecture, the workforce transition, and the vendor decision that together determine what the operation looks like five years from now. If the argument holds, it is this: the legacy trap is real, the AI transition is real, the workforce transition is real, and the leaders who move now — with clear eyes about what will go right and what will go wrong — are the ones whose organizations will look, five years from now, like the future of the industry rather than its past.

The book does not end here. Behind Chapter 12 sit six appendices — the practical tools designed to survive the read. A full glossary of the terms this industry now uses. A working RFP template you can adapt for your own vendor process. A QA rubric for AI-generated interactions. A governance checklist for the AI Trust Council. A total-cost-of-ownership model for AI-enabled contact center operations. And an escalation playbook for when something goes wrong at 2 AM on a Sunday. Twelve chapters put the shape of the argument in front of you. The appendices are what you keep on the desk when the meeting is actually happening.

Thank you for reading.