A founder running a small professional services firm described her weekly one-to-ones to me recently. She had twelve people reporting in at various levels, and she was spending an hour every Sunday evening on preparation: tracking what had been agreed the week before, pulling notes from three different places, assembling talking points she knew she would partly forget by Monday. When she started using AI to draft the agenda and surface the previous session’s notes, that hour dropped to about ten minutes. The conversations became sharper. She stopped arriving half-prepared and apologising for it.
The practical upside is real. There is also a data-protection question that small firms routinely skip, and I will come to it.
What does AI actually do in a one-to-one workflow?
AI in this context means using a language model or integrated meeting tool to handle three categories of admin: drafting the agenda before the meeting, transcribing the conversation during it, and producing a summary with agreed actions after. Each task is something a capable assistant would once have done. The AI works from your existing notes and returns a first draft in seconds, which you review and edit before it goes anywhere.
The UK Government’s AI Playbook specifically names “summarising content, generating agendas and drafting documents” as appropriate uses of AI in professional settings, provided teams keep meaningful human control and follow their own data policies.
In practice this means: before the meeting, you prompt the tool with the previous session’s notes and any open actions, and it returns a structured agenda. During the meeting, a transcription tool records and labels speakers. After the meeting, the AI summarises what was said, pulls out decisions, and extracts next steps with owners and dates.
The common thread is that AI handles the repetitive recall and drafting work. The judgement about what to raise, how to handle a difficult conversation, what commitments to make: that stays with you.
Why does the prep burden hit small teams hardest?
In a business with twenty or thirty people, the owner or a senior manager typically carries one-to-ones with every direct report plus a handful of key clients. At that scale, weekly prep and follow-up can amount to three or four hours. The admin does not scale with the quality of the conversation; it scales with the number of relationships you are sustaining without a personal assistant alongside you.
Microsoft’s 2023 Copilot early-access research found that drafting and summarisation tasks, including meeting agendas and notes, could reduce time on routine writing work by up to 50%. A study published via the National Bureau of Economic Research by researchers at MIT and Stanford found a 37% average productivity uplift on writing and editing tasks when workers used AI assistance, with the biggest gains for less-experienced staff.
For a small firm those numbers land differently than they do for an enterprise. You do not have an executive assistant to prepare the brief. You carry the context for every relationship in your head, which means forgetting something in a one-to-one is a real risk, not just a mild inefficiency. AI gives you a retrieval layer that does not need a salary.
Where in the one-to-one workflow does AI actually fit?
The short answer is all three stages: before, during, and after. The risk level is not equal across them. Pre-meeting agenda drafting is the lowest-risk entry point, because the AI works from notes you already hold and produces something you review before sharing. Live transcription is more sensitive, because it captures the conversation as it happens, and informed consent becomes a clear requirement.
Before the meeting, integrated tools like Microsoft 365 Copilot can pull from previous notes, email threads, and calendar context to draft a structured agenda. Notion AI works similarly if your notes live there. The prompt matters: the more specific you are about format and focus, the more useful the output. Asking the model to use the last two sessions’ notes and highlight overdue actions will get you something worth editing; a vague request will get you something generic.
During the meeting, tools like Zoom AI Companion, Otter.ai for Business, and Plaud Note can transcribe in real time with speaker labelling. Before using any of them, you need the other person’s informed agreement. A simple script works: “I’d like to use a transcription tool so I can focus on the conversation. The notes will be stored in [system], accessible only to [who]. Are you comfortable with that?” The ICO expects transparency when you use AI to process personal data from employee or client conversations.
After the meeting, the AI generates a summary, pulls out agreed actions with owners and dates, and can draft a follow-up email. Your job is to review it before anything is shared or stored. Generative AI can produce inaccurate outputs, and the ICO notes that human review is an expected safeguard in professional contexts.
When does AI prep help, and when does it get in the way?
The strongest case for AI-assisted prep is a regular cadence of structured one-to-ones where you already keep digital notes. The weakest case is a sensitive wellbeing conversation, a performance warning, or any discussion where the presence of a recording tool might reduce what the other person is willing to say. Reading the situation accurately matters more than any general rule about when AI is technically appropriate.
A few situations where AI prep earns its keep: weekly team check-ins with a consistent structure, client relationship calls where you need to recall detail from months ago, and project review meetings where tracking actions against owners genuinely matters.
Where it gets complicated: conversations involving health, performance, or anything where the person might disclose something sensitive. The ICO’s guidance on employee monitoring makes clear that using AI tools to capture personal data requires a lawful basis under UK GDPR, and that health or other special-category data mentioned in meetings requires additional legal protections.
There is also a cautionary example worth knowing. In April 2023, Samsung engineers pasted proprietary meeting notes into ChatGPT. The company restricted employee use of generative AI across the board after the incident. The engineers used a consumer tool with no data agreement in place. That is the mistake to avoid. The fix is choosing enterprise tools where your content stays within your own systems, not free tiers that may use your data for model training.
What connects to this: tools, rules, and where to go next?
Using AI for one-to-one prep sits at the intersection of three broader practices: building a personal information layer that compounds as your notes grow richer over time, understanding your UK GDPR obligations when AI processes employee or client conversations, and choosing collaboration tools whose data-protection posture you have actually verified rather than taken on trust from a vendor pitch.
On the tool side, the options most relevant to a UK services firm in the five to fifty-person range are Microsoft 365 Copilot if you are already on that stack, Zoom AI Companion for video calls, and Otter.ai for Business for standalone transcription. The National Cyber Security Centre’s guidance on using cloud services securely recommends choosing enterprise configurations over free consumer tools, specifically because of how data is stored, accessed, and retained.
On the regulatory side, you remain the data controller; the AI tool is a processor. That means you need a written Data Processing Agreement with any tool that handles personal data from conversations, and your staff privacy notice needs to reflect the tools you use. If your firm operates in financial services, the FCA’s Consumer Duty adds a further layer. Client conversations that touch suitability or advice need to sit within your conduct framework, and that extends to how you capture and summarise them using AI.
The practical starting point is the simplest one. Pick one regular meeting, choose a tool that integrates with what you already use, and try the agenda-drafting step first. Transcription and summarisation come after you are comfortable with the data-handling side.
