Department publishes record outlining how technology has reduced a process that took 30 minutes to one lasting just 60 seconds, with plans afoot to use tool for 80% of correspondence
The Department for Education has revealed that it has begun using a tool based on ChatGPT to draft email responses to communication from citizens.
In a newly published transparency record, the department has released details of the Correspondence Drafter tool, which “supports internal teams by generating first-draft responses to external queries”.
The technology uses the GPT-4o system developed by OpenAI – the second-most recent version of the artificial intelligence firm’s flagship large language model.
The DfE platform, which is hosted in an Azure environment, enables departmental officials to “input [an] external query into the tool”, which then “finds the most relevant information to draft a response”. It is able to create such a response within one minute – compared with the half an hour it currently takes a human to do so, according to the operational record.
The department currently receives about 1,000 pieces of external correspondence each month and intends that, once it is running at full capacity, the AI system will be used to help address about 800 of these.
In its current pilot phase, the Correspondence Drafter calls on four data sets, each of which is a document of outlining the DfE “drafting team’s briefing packs/standard lines to take” in response to differing kinds of queries. These documents range from 20 to 200 pages in length and “the complete scaled-up version [of] the tool will be trained on circa 400 documents”.
Related content
- EXCL: DWP reverses policy prohibiting staff use of ChatGPT and other LLMs – except DeepSeek
- Government signs collaboration agreement with ChatGPT-maker OpenAI
- Government LLM pilots to assess potential ‘employee satisfaction’ uplift
“The user pastes the body of the email they have received into the text box on the tool… [and] the LLM pulls relevant information from the vector database,” the record says. “The returned vectors pass through the LLM and returns text in human language. The text is presented to the user in the front end of the tool, in the format of an email response. The tool enables features to modify output for tone and intended audience. If needed, the user can adapt the output of the generated response before using it in their response email.”
Once a response has been sent, “the individual who has been communicated with can reply back to the relevant correspondence address to ask further questions”.
“A human is in the loop and reviews every correspondence draft recommendation prior to being sent to the public,” the record adds. “The policy teams have a reporting mechanisms if they find any issues with the tool.”
All uses of the technology will be provided with “operational documentation [and] training”, which will encompass “guidance [on the] purpose of the tool, [and an] example end-to-end process, outlining what the tool can and cannot do”.
In development of the tool, the department indicates that it did explore the potential use of other LLMs, but that “Azure’s Open AI’s models provide the extra security the team needed and had received pre-approved by the department’s chief technology officer, [while] open source LLMs have not yet been approved by the DfE”.
Once the tool goes into full operational mode, it will be managed on behalf of the department by cloud provider ServiceNow.
“DfE will undertake reviews and plan in changes to development and test environments before pushing to live,” the transparency record says.
