IBM’s Institute for Business Value has published two reports that frame AI in service as more than cost cutting: one on customer-centric automation, personalisation and sales opportunities in customer service, the other on building a connected enterprise that runs with AI agents. Both point in a direction we recognise. Neither, on the material we hold, supplies figures we can quote, so this article reports their themes and then sets out what operating experience adds.

We run automated telephone contact for banks, insurers, healthcare providers, telecoms and utilities. Our interest in these reports is practical: what would it take for a customer-service operation to deliver the productivity and connection IBM describes?

The two reports#

The two IBM Institute for Business Value reports read here
“AI-powered productivity: Customer service”“The blueprint for agentic operations”
Date15 August 202512 May 2026
PublisherIBM Institute for Business ValueIBM Institute for Business Value
ThemeCustomer-centric automation, personalisation and finding sales opportunities in serviceBuilding a connected enterprise able to operate with AI agents
ScopeCustomer service broadly, wider than telephone contactEnterprise operations, not specific to contact centres
Figures we quoteNoneNone

Sources: IBM Institute for Business Value, 15 August 2025 and 12 May 2026. We report themes only; read the originals for the data behind them.

Two caveats apply. IBM is a technology vendor and its Institute for Business Value is its research arm, so the reports sit alongside products and services IBM sells. And both are broad: the first covers customer service across channels, the second covers enterprise operations as a whole. We draw only their central themes from them and attribute no statistics to them.

Productivity is more than handling time#

The customer-service report puts three ideas side by side: automation that starts from the customer, personalisation, and the discovery of sales opportunities inside service contact. The combination matters. It treats a service interaction not only as a cost to be reduced but as a moment that can produce value.

We agree, with a caution about how productivity is measured. The traditional measure in contact centres is average handling time (AHT). It is easy to track and easy to misread. A shorter call is not automatically a better one, and a longer one is not automatically wasteful. In banking, our cases ending with a promise to pay averaged 116.5 seconds (historical), but that number describes the conversation, not the result. AHT is not the outcome.

116.5 sAverage handling timeMean duration of completed conversations.Scope: banking collections cases ending in a promise to pay, historical. AHT is not the outcome.

That is also why new Vee agreements are no longer billed by the minute. A minute rewards long calls, while results often come from shorter ones. Productivity, for us, is measured on a ladder: reach, continuation, obtaining the key information and the real business effect, such as a payment, a contract or an attended visit.

If we are paid by the minute, we are not playing towards the same goal as our client. – Dawid Wójcicki, Founder & CEO of Vee

Personalisation and sales, in regulated contact#

Personalisation on the phone does not mean improvising. In regulated processes our replies come from a closed, versioned library, and there is no learning in production. Personalisation happens through structure agreed with the client: the conversation reflects where the customer is in the process and what has happened before, within approved content.

Sales opportunities in service are real, but they have to be handled with care. In telecoms, active outbound sales is a large cost when done by people; we take the first contact on retention, renewals and pre-sales, and pass the interested customers to the operator’s salespeople, paced to their capacity and hours. In insurance, contact during the life of a policy creates warm conversations for renewal and cross-sell, rather than waking up only at renewal. In one lead pre-selection campaign we measured 26% more interested leads and a 73% lower customer acquisition cost (historical, that campaign only).

What we would not do is turn every service call into a sales call. A customer in hardship, or one disputing a debt, is not a sales opportunity. Those situations lead to a transfer, a stop or a referral, by design.

Agentic operations need a connected enterprise#

IBM’s second report is about operations: how an enterprise becomes connected enough to run with AI agents. We read it for one idea only, because it is the one that matters most on the phone. An agent, human or automated, can act only as far as the systems around it allow.

For automated contact that means three things. Data must reach the conversation before it starts, in the right format, by REST API, SFTP flat file or HTTPS. The outcome must reach the client’s systems after the call, as a status its teams can act on. And the boundaries of what the conversation may decide must be fixed in advance and approved by the client. Allocation strategy, meaning which accounts are contacted and when, stays with the client’s analysts.

We also avoid describing this as autonomy. Our processes include agreed transfers to the client’s own staff, every call can be reconstructed from its recording, logs, statuses and script version, and every content change in a regulated set-up is versioned and approved. A connected operation, in our experience, is one where every step can be traced, not one where every step is left to an agent.

Scale is infrastructure, not headcount#

There is one productivity effect of connected, automated operations that rarely appears in reports and that we have lived through. When demand jumps, a human contact centre has to hire and train; an automated one has to add capacity. In March 2020, as the pandemic began, demand from our large clients rose fivefold, and channel capacity at one mobile operator was raised from 250 to 900. That was possible because scaling depended on infrastructure rather than on the availability of staff.

The same property works in the other direction. A whole appointment-confirmation campaign for the coming days can be completed within hours rather than spread across a week of staff time, which matters when the point of the call is to free a slot early enough for someone else to use it. Productivity, in this sense, is the ability to act at the moment the business needs it, not the number of calls per hour.

Both reports imply something they do not dwell on: productivity and personalisation depend on learning from the contact itself. That is a common gap. A 2026 study by EY-Parthenon with PSMB of around 50 organisations in Poland found that 98% record their calls but 69% analyse less than 5% of the volume. The sample is small and the answers are self-reported, but the pattern is familiar.

Our own operating data shows what is at stake. By 31 October 2026 we had accumulated 28,569,847 minutes of AI conversation (cumulative, Vee operating data; AI conversation time only, human agents not included). Contact on that scale is a record of how customers respond to timing, wording, voice and channel. Our analysts work on correlations and A/B tests across it, and our QA team listens to selected calls every day in every project, especially unusual ones.

  • Statuses on four axes. Technical, dialogue, processing and business, so a problem can be traced to its layer.
  • Comparators, not anecdotes. Results compared with a control on the same buckets and portfolios.
  • Fast change. One approved change reaches every live conversation at once.

What to ask before you invest#

  1. Which productivity measure will you use, and is it an outcome or an activity?
  2. Where in the service journey are sales opportunities appropriate, and where are they excluded?
  3. Which systems will the conversation read from and write to, and who builds the interfaces?
  4. What share of contact is actually analysed, and who acts on what is found?
  5. Can every automated decision be traced and reconstructed?

IBM’s reports point in a sensible direction: from cost to value, and from isolated tools to connected operations. The distance between that direction and a working service is covered by process design, integration and daily analysis. That is the part we run for regulated enterprises.

Follow The First Second

Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.

Follow via RSS