Back to All Posts

Assurance Baselines: Most Support "Solutions" Are All Tool and No Content

Most tools leave agents to judge what proof is enough. We wrote Assurance Baselines so compliance teams set those rules before the ticket comes in.

Look at almost any Zendesk app directory and you'll find a long list of features: dynamic forms, automated triggers, checklist formats, and custom reports. It's a well-built toolkit.

But those tools won't tell your agent how much proof they need to close a deceased customer's account. They won't set the policy on how many times a locked-out user can retry a failed age check, or who can overturn a rejection. Templates don't go much further; they ship generic tokens and leave the definitions blank. The customer still has to write those definitions, and the easy option is to keep them light and leave the call to the agent, who then decides case by case what counts as "good enough."

Teams increasingly praise human agents as the people who bring judgment to a case. On a bad day, a human agent can hallucinate as readily as any LLM.

The gap shows up most in high-stakes, high-friction requests: a grieving relative trying to close an account, a user demanding a full export of their personal data, a panicked caller who lost their phone and needs back into their account right now.

Once a customer falls off the automated happy path, a human agent has to make the identity decision. In most organizations, nobody has written down the standard operating procedure. No playbook tells the agent what to check, in what order, or what audit trail to leave behind.

Capability lists and templates leave the hard part with the agent

When a team buys a capability, it still has to decide how to use it. Someone has to decide which document to request, when to stop asking, who approves an exception, and how long to keep the passport image. Compliance teams write policy at one level of detail and agents work at another; the procedure between the two rarely gets written at all.

So agents improvise. Some ask for too little and let the wrong person in. Others ask for too much; regulators in Sweden and Spain have ruled against companies for exactly that. Either way, agents leave the documents they collect in tickets and inboxes that nobody reviews. When an auditor asks why an agent accepted a particular piece of evidence, nobody can answer.

What we found when we tried to map the frameworks

We set out to map each of our verification workflows to the governance and compliance frameworks our customers already use. We expected to find a control for each moment and build to it.

We found far fewer than we expected. Few framework authors address these service-desk moments at all, and the ones who do write their controls loosely enough that an organization can pass them without its agents ever following a consistent procedure. As far as we could find, nobody had written down the moment-level controls that sit between a policy and an agent's decision.

So we wrote them ourselves, in the structured, standards-based format NIST uses for its controls. We call each one an Assurance Baseline.

What an Assurance Baseline sets out

An Assurance Baseline is a written position for one high-risk service moment. It states, in terms a compliance lead can sign and an agent can follow:

  • which moments it covers, and which fall outside it
  • what evidence counts as enough, and what the agent must never ask for
  • the checks the agent runs, in order, and where the case waits for confirmation or a second approver
  • how long a challenge stays open, and how many retries a customer gets
  • how long the organization keeps verification records, and where it keeps them
  • who may overturn a refusal, and what the organization tells someone it refused
  • what the record on each ticket must show

Most organizations we speak to can't answer all of those questions for their own desk today. With a Baseline, the team agrees the answers before the ticket arrives; the Baseline is that agreed policy.

We write one common Baseline for each moment and map it to the standards and GRC frameworks that apply across the major jurisdictions. The rules still change from one country to the next: age-assurance appeals under the UK's Online Safety Act carry different obligations from a subject access request handled under Swedish or Spanish data protection law. So each customer adds extensions for its own organization and markets on top of the common Baseline. The desk then runs that Baseline as its working policy and control, and it becomes the defensible record of why the team handled each case the way it did.

A position you adopt, a workflow you run, a record you keep

Each Baseline comes with two other things. The first is the workflow that runs it inside Zendesk. The Baseline acts as the specification for that workflow: it guides and controls each action that carries out the policy, so agents follow its steps on the ticket and don't have to remember them.

The second is the witness record. On every case, anyone who opens the ticket can see which checks the agent ran, who approved what, and which Baseline version applied. Agents create that record as a by-product of doing the work, so nobody has to do extra work at the time, or later, to prove the team followed its policy.

When a customer updates a Baseline, we keep a register of which version each ticket was assessed against, so the compliance team can answer for an old case under the rules that applied at the time.

We describe a Baseline as an asserted position, because that is what it is. A regulator doesn't certify it; we don't claim it makes anyone compliant. In each Baseline we state what it covers and what falls outside it; a customer can read it, challenge it, change it, and adopt it as their own. With a Baseline, a compliance team gets what a capability list or template can't: a precise statement of what their desk does in a given moment, and evidence that agents did it on each ticket.

The Cloudset Recommendation

If your service desk handles identity-sensitive requests, start here:

  1. List your moments. Write down the requests that take a customer off the automated path and leave an agent to decide whether they are who they say they are, or have the right to act for someone who is.
  2. Pick the one with a deadline. Subject access and erasure requests, closures after death, and age-assurance appeals each have a deadline or a reporting duty, and compliance teams already know the risk.
  3. Answer the hard questions for that moment. Write down what counts as enough evidence, how many retries a customer gets, how long you keep verification records, who can overturn a refusal, and what you tell someone you refused. If you can't answer one, you've found the gap.
  4. Extend it for one market, not every market at once.
  5. Check one closed ticket against it. Ask whether the ticket shows what the agent checked, who approved it, and why. If it doesn't, you can't show that your control operated on that case.
  6. Adopt the position, then run it. Put the procedure into the workflow your agents use, and make the record part of the ticket.
Read More
/ Cloudset

Calibrated Navigation

Let guidance guarantee compliance with Cloudset
Topographic contour map pattern