The pilot was going well. For three weeks, every call the team took for an energy client had been scored automatically, and the operations director finally had what a QA team of four never could give her: a view of every conversation, not a sample of them. She put the numbers in the monthly report.
Two days later the client’s data protection officer wrote back. Not about the numbers. About the supplier: who is this company processing our customers’ calls, where do they process them, and when did we approve that?
The pilot was paused for six weeks.
When an outsourced contact centre adds AI to a client’s calls, the calls are the client’s data, the quality standard should be the client’s, and the client has to approve the new supplier before the first call is processed. Settle four things up front: permission, scorecard, visibility and calibration. Do that, and the DPO approves the supplier before the pilot starts, not six weeks into it.
The calls are your client’s, so the yes is theirs
Under the GDPR, your client is usually the controller of the customer data in those calls, and you are the processor. An AI scoring service you bring in becomes a sub-processor: someone processing the client’s data on your behalf.
Article 28(2) of the GDPR is plain about what that takes. A processor may not engage another processor without the controller’s prior written authorisation, either specific (for this supplier) or general (for suppliers in general, with the right to object to each change). Which of the two you have is written in your data processing agreement with the client.
- Specific authorisation: you need the client’s written yes for this supplier before any call is processed.
- General authorisation: you still have to tell the client about the new supplier in advance, and give them time to object.
Either way, the client hears about it from you, before it happens. Send who the supplier is, what it does with the calls, which data it sees, where it processes it and which further suppliers it uses. We publish our full sub-processor list for exactly this reason: your client’s DPO can check it without waiting for anyone.
The client may also need a data protection impact assessment before it can say yes. Send what they need for it with the notice, so the assessment doesn’t wait on you.
Score on the client’s scorecard, not yours
Most outsourced contact centres have their own quality form, and it is tempting to put every client through it. Resist that. Your client already has a definition of a good call, often in a QA form their own team uses or in the targets written into your contract. If the AI scores against your form instead, every report starts with a translation exercise, and the client will trust it less.
Start with their criteria, and use the rollout to sharpen them. Criteria like “showed empathy” or “was professional” mean different things to different listeners, people and AI alike. Rewriting them into things a reviewer can actually hear, such as “acknowledged the problem before offering a fix”, is work you do together with the client. We’ve written about how to word criteria so they can be scored consistently.
If you serve several clients, keep their scorecards apart. A bank’s checks and a retailer’s checks do not belong in the same average.
Decide what the client sees before they ask
Once every call is scored, the client will want to see the results. That is the point. But a client with access to everything will soon be managing your agents, and that is not a role either of you wants them to have.
Draw the line in advance:
- What the client sees: scores on their criteria, trends, why their customers call, and the calls themselves with the evidence behind each score.
- What stays with you: individual agent rankings, disputes and reviews, and the coaching you give. That is how you run your operation, and agents who feel watched by the client stop being coachable.
In Future Ready this is a separate client role. Clients see the dashboard, calls and the reports you share with them, but not agent lists, agent pages, disputes or training. Whatever tool you use, agree on the line in writing, so it doesn’t move every time a number dips.
Calibrate with the client’s QA team, not only your own
A client loses faith in AI scoring the first time they listen to a call and disagree with its score. Get ahead of that: show them how often their own reviewers agree with each other, and how often the AI agrees with them.
Ask the client’s QA team to score a sample of calls blind, alongside your reviewers and the AI. Compare criterion by criterion. Where everyone agrees, the AI can carry the work on every call. Where the client’s reviewers and yours disagree, you have found a difference in how the contract is read, which is worth knowing whether or not you automate anything. We explain how to set this up in QA calibration for AI call scoring.
Bring it to the renewal
Most QA teams hear between 1 and 5 percent of calls, by the estimates of ISG and Gartner. Your client’s team probably does too. A view of every call, on their own criteria, is something they rarely have.
So bring three things to every account review: the share of their calls you scored, the criterion that improved most since go-live, and their top contact reason with what you changed because of it. By renewal, the conversation has a year of evidence behind it, not just an hourly rate. We’ve written more about becoming the partner a client can’t afford to replace.
A checklist for the first client
- Read your data processing agreement with the client. Does it give specific or general authorisation for new sub-processors?
- Send the supplier notice before any call is processed: who, what, which data, where, and which suppliers they use.
- Get the client’s scorecard, and rewrite the vague criteria together with them.
- Agree what the client sees, and what stays inside your operation.
- Run a blind calibration with the client’s QA team on 50 to 100 calls.
- Agree on the monthly report: coverage, trends on their criteria, top contact reasons and what you did about them.
This adds a week or two at the start. Skipping it cost this pilot six.
Running several clients on one floor? We’ll score a sample of one client’s calls on that client’s scorecard, so you can show them the result before you ask for anything else.
