The Three-Minute Gap: Auditing Where Language Access Slows UK Service Channels

Most UK organisations can tell you what they spend on interpreting. Very few can tell you what language access costs them. The difference is not a matter of accounting pedantry; it is the difference between paying an invoice and understanding a channel.

A Welsh-language caller in Cardiff, a British Sign Language user on a video relay call in Leeds, a Polish-speaking customer in a webchat queue at eleven at night: each of these is not an edge case bolted onto a standard service, but a contact type with its own latency, its own failure modes and its own economics. When that contact type has no owner, no measurement and no design, the cost does not disappear. It relocates — into queue time, into repeat contacts, into informal bilingual cover and into translation work that never appears on any ledger.

Language is a channel property, not a supplier contract

Every service channel has three defining characteristics: how long it takes, who is on the other end, and what record survives the interaction. An interpreted contact preserves all three and adds a fourth party — the interpreter — whose availability, booking rules and connection method sit between the customer and the resolution.

That is a design problem, not a procurement problem. Yet in most operating models the interpreter is bought as a framework service, with per-minute rates, minimum call durations and out-of-hours premiums, while the channel itself is designed as though everyone speaks English at conversational speed. The consequence is a service that performs acceptably in aggregate and poorly in precisely the segments where the statutory and reputational risk is highest.

Three clocks, not one

A standard voice contact has a single clock: queue to answer. An interpreted contact has at least three, and organisations rarely measure more than the first.

• Connection time. The interval between the agent recognising a language need and an interpreter joining. On a booked appointment this is close to zero. On an ad hoc call it can run to many minutes, and in less common languages considerably longer.
• Comprehension time. The contact itself runs at roughly double the length of a monolingual equivalent, before account is taken of the explanation the interpreter needs and the summary the agent must give.
• Reconstruction time. The point at which the case is handed to a back office that cannot read the original exchange. Notes are written in English, often abbreviated, and the reasoning that produced them is lost.

Add these together and the true handling time of an interpreted contact is not a marginal uplift on the English baseline. It is a different service, with a different cost profile, and it deserves to be modelled as such.

The statutory floor has been rising

The Welsh Language Standards, made under the Welsh Language (Wales) Measure 2011, impose enforceable duties on specified bodies covering correspondence, telephone calls and published material. The British Sign Language Act 2022 signalled a wider expectation of BSL provision across public services, and the Accessible Information Standard has, since 2016, required health and social care providers in England to identify and meet communication needs. The Equality Act 2010 duty to make reasonable adjustments sits underneath all of it.

These are floors, not targets. Compliance is generally demonstrated through documents and process statements; it is rarely demonstrated through measured customer outcomes. An organisation can be fully compliant on paper and still leave a BSL user waiting eleven days for an appointment slot that a hearing customer would have had in ninety seconds.

Where the duplication hides

Three patterns recur in almost every audit of multilingual service operations, and none of them appear in the interpreting invoice.

Unlogged machine translation. Agents paste customer messages into free translation tools and paste the output back. The customer receives an answer; the organisation receives no record of which language was used, what was translated, or whether the rendering was accurate. In regulated sectors this is the most common unaudited processing activity in the estate.

Informal bilingual cover. A colleague who speaks Portuguese is asked to step in. It is quick, it is unpaid, and it is invisible — until that colleague leaves, or until the same request arrives during their annual leave.

Translation that is done twice. Customer-facing forms, letters and help articles are translated by one team for the website, again by another for contact centre scripts, and a third time by a regional office for local distribution. Nobody compares versions, and consistency drifts.

Five registers worth building

An audit does not require a new platform. It requires discipline in five places.

1. Contact register. Where and how a language preference is captured — at the point of contact, in the CRM, or not at all. If it is not captured, it cannot be routed.
2. Time register. Timestamps for queue, interpreter connection, interpretation minutes and resolution. Even a fortnight's sample will reveal the shape of the problem.
3. Handoff register. How many parties touch a contact before resolution, and how much of the original exchange survives each transition.
4. Content register. Which pages, forms and letters exist in which languages, at which version, and who signs them off.
5. Cost register. Per-minute rates are the least interesting number. Minimum durations, no-show charges, out-of-hours premiums and rework are where the money actually goes.

Read together, these registers convert a diffuse complaint about complexity into a set of addressable design choices.

Designing the interpreted channel deliberately

Where the volume justifies it, the sensible pattern is to split the estate by intent rather than by language. High-volume transactional intents — balance enquiries, appointment confirmation, status checks — should be served through verified written channels that are properly translated once and maintained thereafter. Complex, judgement-heavy contacts should be booked, with the interpreter scheduled alongside the specialist rather than dialled in on demand. Video should be the default for BSL, not an escalation.

None of this is novel. It is simply what happens when language access is treated as part of the channel map rather than as a supplier relationship maintained at arm's length by a procurement team.

The organisations that get this right tend to measure four things: first-contact resolution by language group, interpretation minutes per resolved contact, the proportion of interpreted contacts that could have been self-served, and the number of handoffs a contact endures before it closes. The first tells you whether the service is fair. The second tells you whether it is efficient. The third tells you whether the channel estate is correctly drawn. The fourth tells you whether the resolution will survive contact with the back office.

Language access is not a tax on good service. It is evidence of where the channel design was never finished.