Seven questions about your highest-consequence ticket journey, and what we're doing with the answers.
CX automation works - that’s the first point we want to make.
Handle times are down, self-service deflects a growing share of routine contacts, and copilots are surfacing accurate answers in seconds. If you run a CX operation today, your tooling is better than it's ever been, and your metrics probably say so.
But automation removes volume, not variance.
This isn't a complaint about AI, and it’s not an argument that you bought the wrong software either; most of what the market sold you does what it said on the tin and does it better today than it did yesterday.
Each generation of tooling absorbs the simple, repeatable contacts: password resets, order status, return windows. What's left in the queue is the residue, the complex, sensitive, regulated work: a subject access request, a vulnerable-customer escalation, a claim with a reporting obligation attached.
These journeys now make up a smaller share of tickets, but a larger share of everything that can hurt you.
Because automation takes the easy cases, agents get fewer repetitions at the hard ones than their predecessors did. Lisanne Bainbridge made this point in 1983 in Ironies of Automation, a paper about industrial control rooms. The more you automate, she wrote, the more you depend on the human for exactly the cases they're least practiced at handling. She was writing about power plants; but swap "control room" for "support queue," and it describes the experience most CX leaders are feeling right now, even if they haven’t fully articulated it yet.
Rates and instances
Most CX metrics describe populations: average handle time, CSAT, automation rate, first-contact resolution. Audits ask about instances, and the arithmetic flips. A 99% automation rate is a triumph; "99% of our subject access requests followed the required steps" is a failed audit. In a regulated environment, the question is run per ticket: did every required step happen, and can you show me the proof?
The assurance ladder
For most ticket types, each is the proportionate answer. The question is where your highest-consequence journeys land.
- Level 0: Undocumented. The process is stuck in experienced heads.
- Level 1: Documented. Written down in a wiki or SOP folder, with no relationship to the ticket in front of the agent.
- Level 2: Prompted. The right information surfaces at the right time, and adherence stays voluntary. Macros, knowledge bases, and AI copilots belong here.
- Level 3: Observed. You know your adherence rate, after the fact, from a sample.
- Level 4: Enforced. The ticket can't advance past an unsatisfied checkpoint.
- Level 5: Evidenced. Enforcement leaves a version-stamped, instance-level record as a byproduct of doing the work.
The critical threshold is between levels 3 and 4. Lower levels track broad population metrics, whereas level 4 and above guarantee instance-level certainty.
Your required level depends entirely on the stakes.
If a failure on your most critical journey simply results in churn, level 2 is sufficient. But if that same failure triggers regulatory action, fines, or licensing issues, operating at level 2 means you have a dangerous compliance gap - whether anyone has admitted it yet or not.
The seven questions
Design: do we know what we do?
- Is this process mapped, or does it live in tribal knowledge and macros?
- Do CX, Compliance, Legal, and Ops agree that's what it is? Who owns the map?
Execution: do we actually do it?
- Is the designed process the executed process, and how would we know if it diverged?
- For any single ticket, can we evidence who did what, in what order, when?
Evolution: does it work, and were changes controlled?
- Does the process produce the outcome it exists to produce?
- When the process changes, how fast does the change reach every agent, identically?
- Which version of the process was in force on the date of the incident?
Before you start:
First, software won't save a misaligned team. If three stakeholders describe the same journey differently, buying a tool won't fix it. Agreement has to come before enforcement - software can make a map binding, but it can't make it true. Second, if you can already answer all seven questions, keep your money. A zero-dollar budget is a win!
Where we stand
Because we build process enforcement for Zendesk, we sit at the top of this ladder and have a commercial stake in your compliance. But you don’t need to buy our software to use this assessment.
Our hypothesis is that regulated CX teams are carrying a dangerous gap between what they are required to prove and what their systems can actually validate. We believe this problem is concentrated in complex, ownerless journeys, and we’re putting it to the test.
Over the next few weeks, we’re running these seven questions with a dozen organizations, from current clients to cold contacts; and we’ll publish the raw results, even if they completely kill our theory.
Our ask
You can use this in one of two ways.
- Do it yourself: Take these questions into your next operations meeting. Put the right people in the room, pick a journey, and keep your findings to yourself. If it works, that’s half our job done!
- Do it with us: Give us thirty minutes to map one journey. You get a concrete, one-page compliance report you can hand straight to Risk. There’s no demo and zero obligation - even if the conclusion is that you don't need our help.
If you’re ready to take that step, email us!
