Bain argues that a large share of the work in customer support does not happen inside any one system but between them, and estimates that roughly 40–60% of workflow tasks in customer support are automatable. We think the framing is right, and the phone is where it is easiest to see. A conversation is worth little until what was said reaches the systems that act on it.
Our reading is deliberately narrow. We report only what Bain’s article supports and then widen the article with what we see running automated telephone contact for regulated enterprises.
What Bain published, and what we can confirm#
The analysis is titled “The $100-Billion SaaS Opportunity Hiding in Cross-System Labor”. Bain published it on 7 May 2026 as part of its Technology Report 2026. Its estimate for our field is that customer support, together with R&D and engineering, sits at the high end, with roughly 40–60% of workflow tasks automatable; the cross-system work it describes spans ERP, CRM, billing and support ticketing.
| Item | Detail |
|---|---|
| Publisher | Bain & Company |
| Date | Announced 7 May 2026 |
| Type | Analysis and estimate, not a survey |
| Key estimate | Roughly 40–60% of customer-support workflow tasks automatable (R&D and engineering similar) |
| Access | Part of Bain’s Technology Report 2026; the market size in the title refers to the US only |
Source: Bain & Company, May 2026. We report only the headline estimate; method and sample are not covered here.
Three caveats matter. First, this is an estimate of potential, not a measurement of automation achieved. “Automatable” describes what could in principle be done, not what organisations have done or will do. Second, the range is wide, and the lower and upper ends imply very different business cases. Third, Bain itself advises applying its automatability factors to sub-processes rather than whole functions, so a function-level range is a starting point, not an answer for any one process. We do not repeat the market-size number in the title as a finding of our own.
Why the framing matters more than the number#
The useful idea in Bain’s title is “cross-system labour”. Much of the effort in customer support is not talking to customers. It is looking something up in one system, copying it into another, checking a third, and making sure the record says the same thing everywhere. Agents spend their day as the glue between applications that were never designed to talk to each other.
That is why automating the conversation alone often disappoints. A bot that can talk but cannot read the account, write back the outcome or trigger the next step simply moves the cross-system labour to someone else, usually a back-office team working through a queue the next morning.
The work between systems, on a phone call#
Take a typical early-collections call for a bank. Before the call, the right accounts must be selected by the bank’s analysts and passed to us with the data the conversation needs: delinquency stage, amount, previous promises. During the call, identity must be verified before any debt detail is mentioned, and the conversation must reflect the customer’s history. After the call, the outcome must go back to the bank as a status it can act on: a promise to pay with a date, a dispute, a request for a person, a hardship flag.
Each of those steps crosses a system boundary. Most of them are invisible to the customer, and all of them decide whether the call was worth making.
- Input. Case data arrives by REST API, SFTP flat file or HTTPS, in the format agreed in the integration specification.
- Verification. Identity is checked before any detail is disclosed; this is a platform rule, not a line in a script.
- Conversation. The dialogue reflects the case data and the customer’s history.
- Exceptions. A request for a person, failed verification, withdrawn recording consent, a legal dispute or financial hardship leads to a transfer, a stop or a referral.
- Output. A daily case-level status in the client’s own terms, which its teams can act on.
- Follow-up. Retry rules, telephony and SMS, tuned for each process.
In healthcare the shape is similar. A confirmation call is internally a cancellation call: its purpose is to learn early that a visit will not happen so the slot can be offered to someone else. The value appears only if the cancellation reaches the calendar in time. A perfect conversation that leaves the slot blocked has achieved nothing.
We call it appointment confirmation. Internally we call it cancellation, because that is the business goal: learning early that a visit will not happen. – Dawid Wójcicki, Founder & CEO of Vee
When the other system is another organisation#
Sometimes the system on the far side of the boundary belongs to someone else entirely. In medical-benefit booking for insured persons, we take the request from the insured customer and then call the receptions or IVRs of partner clinics until we find a slot that matches the customer’s preferences. The cross-system labour here is literal: a sequence of phone calls between organisations whose calendars are not connected, usually done by a person today.
In energy and utilities the boundary runs between the customer, the call and the field schedule. Confirming meter access, a reading or a replacement before the technician sets out only saves a wasted visit if the confirmation reaches the schedule. The phone call is the visible part; the value sits in the update that follows it.
Why we specify integration before dialogue goes live#
Our specification for every new process has three parts: the dialogue, the integration and the telecoms. They are written together because they fail together. A beautifully designed conversation with a broken status feed is a failed process, and so is a working integration behind a number that customers do not pick up.
Units matter here too. We count accounts, cases and attempts separately. An account is the customer relationship, such as a loan contract; a case is each assignment of that account for handling, and one account can produce several cases a month; an attempt is a single dial. Statuses run on four axes: technical, dialogue, processing and business, the last being the client’s version of the truth, in its own language. Agreeing those definitions before launch removes a large share of the cross-system arguments that otherwise surface later.
Where we would be cautious with Bain’s estimate#
In regulated sectors, not every task that can be automated should be. Allocation strategy, meaning which accounts are contacted and when, stays with the client’s analysts in our model, because it is credit policy, not contact execution. Decisions about hardship, disputes and complaints often belong with the client’s own staff. A high automatable share is an upper bound on what is technically possible, not a target for what is appropriate. Bain’s range is also an estimate across a whole function; in a single regulated process the automatable share may sit well below or above it.
We would also note that cross-system work brings cross-system risk. Every interface is a place where data can be wrong, late or exposed. That is why our data exchange is specified, tested atomically and end to end, and covered by the client’s data-processing agreement and, where needed, a data-protection impact assessment. Our operating company, Vee Center Sp. z o.o., holds ISO/IEC 27001:2022 certification. Clients under DORA and the AI Act audit us as a sub-processor, and we help them assess their own configuration.
What to ask a supplier#
- Which of our systems will the process read from and write to, and through which interfaces?
- What exactly comes back after each call, in which format, and how quickly?
- Who agrees the definitions of account, case, attempt and status before launch?
- Which decisions stay with our staff and analysts, and how is that enforced?
- How is each interface tested before real customers are called?
Bain’s estimate is a reminder that the real opportunity in support automation lies between systems, not inside the chat window. On the phone that is doubly true. The conversation is the execution layer; the result depends on what happens on either side of it. That is the part of the work we run end to end.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS