When everyone owns the AI, nobody owns the phone call

NoJitter ran a good piece last week on who actually owns AI in the enterprise. The short version: AI isn't concentrated in IT any more. Marketing spins things up, ops tweaks them, finance buys a tool, and everyone ships. The question stopped being who owns AI and became how do you govern when everyone does.

The stat that jumps out, from IDC:

68% of agentic AI investment decisions are driven by business leaders, while only 4% are owned by AI centres of excellence.

Right. So ownership is already spread across the building whether you planned it or not. The article's answer is sensible: reposition your governance people from gatekeepers to enablers, get cost visibility, keep everyone aligned. All true.

But it's a conversation about the inputs. Budgets, approvals, who signs what. And it quietly skips the one thing your customer never sees a budget line for: the call. 📞

TL;DR

• Fragmented ownership is a governance problem for your inputs and an accountability problem for your outputs. The article covers the first and skips the second.
• The output that matters most is the one with a human on the end of it. On the phone there's no "reload the page" - just a caller stuck in a loop at 2am and nobody whose job it was to catch it.
• When five teams touch the voice agent, "does it actually work for the caller" belongs to all of them, which means it belongs to none of them.
• You don't have to settle who owns the AI to make it accountable. You need one thing every stakeholder trusts: the agent tested from the outside, from the customer's side, on a schedule.

Governance without a source of truth is just a meeting

The three things the article recommends - pre-approved use cases, cost visibility, workforce alignment - are all real. But notice they're all about deciding to build the thing and paying for the thing. None of them tell you whether the thing you shipped is any good once a customer is talking to it.

That's the gap. And it's widest exactly where ownership is most fragmented.

Picture a customer-facing voice agent. Marketing owns the greeting and the brand voice. Ops owns the call flow. A vendor owns the model. Someone in product tuned an intent on Tuesday. Compliance signed off on a version that no longer exists. Every one of those people is doing their job, and not one of them is on the hook for the specific moment where an 82-year-old caller says "I just want to speak to someone" and the bot cheerfully starts reading them the data protection notice for the third time.

That failure isn't in anyone's swim lane. It's between them. And the between is where voice goes wrong.

"Aligned on priorities" doesn't survive contact with a live call

The article's honest caveat is that fragmented ownership is fine as long as stakeholders stay aligned, and dangerous the moment alignment slips. I'd go further: alignment on a slide is cheap, and it drifts the second the thing goes live.

Everyone agrees the agent should be helpful, compliant and on-brand. Lovely. Then it ships, someone swaps a voice, someone adds an intent, the model gets an update nobody flagged, and the version taking calls this afternoon is not the version everyone aligned on last month. A voice channel is a living thing. Alignment is a snapshot. The two come apart continuously, and no amount of governance-as-enabler catches it, because governance is looking at the roadmap, not listening to the calls.

You can't align your way to quality on a moving target. You have to measure it.

The bit that makes ownership stop mattering

Here's the useful reframe. The whole debate assumes you need to decide who owns the AI before you can hold it to account. You don't. You need something outside the org chart that everyone treats as the truth.

The only way to know whether your voice agent works for a caller is to have something that isn't your voice agent phone it up and check. From the outside. Over a real line. Against what a real customer would need.

That's what we built. You point it at your actual number and a synthetic caller runs the journeys that matter: does the hand-off land, does it disclose it's an AI, does it slow down for a confused caller, does it take "uh-huh" as consent, does it survive someone going off-script. Every result quotes the transcript, so whoever owns that step - whichever swim lane it turns out to live in - can see exactly what broke and fix their bit.

Three things fall out of that, and they're the three the governance debate can't hand you:

• It's the shared truth nobody can argue with. Marketing, ops and the vendor don't have to agree on who owns the failure. The transcript owns it. The finger-pointing is over before it starts.
• It runs on a schedule, not once at sign-off. Your agent drifts every release. The check drifts with it, so "does it still work" is a test suite, not a memory of a meeting in April.
• It's evidence, not opinion. A dated record that the journey did the right thing on the day it mattered. The thing you actually want when a regulator, or a furious head of CX, asks what happened on that call.

This is the same argument I keep making about where testing belongs in the lifecycle: the expensive failures live at the boundary between the tool and your specific use of it, and no org-chart decision reaches that boundary. Only a real call does.

So, who owns it?

By all means have the orchestration conversation. Pre-approve your use cases, get your cost visibility, make your governance team enablers instead of gatekeepers. That's all good, and the article's right to push it. 👍

Just don't mistake governing the inputs for knowing the output is good. You can run the tidiest AI operating model in your sector and still ship a voice agent that dead-ends a vulnerable caller, because that failure was never on anyone's budget line to prevent.

You don't need to win the ownership argument. You need one honest, outside-in answer to does the thing our customer just called actually work - and you need it to keep answering, every release, forever. Settle that, and it stops mattering quite so much who owns the rest.

Frequently asked questions

Isn't this a governance problem we can solve with clearer ownership? Clearer ownership helps with the inputs - who builds, who buys, who approves. It doesn't tell you whether the shipped agent works for a caller. That's an output, and outputs have to be measured, not assigned.

We already do QA on the bot before release. Point-in-time QA is a claim about a version that has since changed. A voice channel drifts every release. The question a customer hits is about the version answering now, which is why the check has to run continuously, not once at sign-off.

Who should own the testing, then? Ideally nobody's private property. The value of an outside-in check is that it's neutral ground every stakeholder trusts, so it settles disputes instead of adding another thing to fight over.

Does it matter what platform the agent runs on? No. We test over real PSTN, so anything that terminates a phone number is testable, whichever vendor or stack sits behind it.

---

If your AI is owned by everyone and your phone line by no one, that second bit is the job we built TotalPath for.

Talk to us

Or register for a free account and run one against your own number this afternoon.