Business productivity with AI: emails and meetings

By AiKitOctober 6, 202612 min read

Person checking their email — Cover image generated with AiKit Image. The people shown do not exist.

Contents

If you reach the end of the day without really knowing where it went, it may be worth looking at email and meetings. Messages that pile up, meetings that could have been an email and tasks left half-done: that's where a good part of business productivity goes.

It isn't always easy to bring order when that work is part of everyday life. AI productivity tools can take some of the repetitive work off your hands and give you more control over your time. Here we tell you where to start.

Why do emails and meetings hold back business productivity?

Email and meetings hold back business productivity because they take up hours that aren't always billed and leave agreements without anyone responsible for them.

It's easy to think the problem is discipline and that everyone should organise themselves better. But not everything depends on how each person gets organised. If the team receives more emails than it can handle, or there are no priorities and no one responsible, messages pile up, get answered late or simply go unanswered. Then the problem isn't that someone organised themselves badly, but that there's no clear way to share out the work.

Every client email should have someone assigned to answer it, and no agreement should leave a meeting without action items, an owner and a date.

Example

An architecture practice

The site meeting might end with a “we'll look into it this week”. As nobody would say who's looking into it, a few days later the client would ask by email and two architects would reply with different answers. The practice would end up working on the same query twice, and the client's trust would start to wear thin.

To find out whether this happens to your team, ask two colleagues what was agreed at the last client meeting and who's in charge of it. If their answers don't match, the bottleneck is in how work is shared out. That's where the rule starts, and AI comes in afterwards so that sticking to it doesn't depend on anyone's memory.

How do you use AI for emails and turn them into tasks?

AI in your emails helps you optimise time you could be billing and makes sure that no client is left waiting without someone looking after them.

Before asking it for anything, check what permission you're giving it over your email. The most prudent approach is to start with read-only access: the AI reads, searches and summarises, but doesn't send or leave drafts in your inbox. With the ERP, on the other hand, it can write, and it creates tasks with an owner and a date. That leaves the team with two decisions: who reviews and sends each reply, and which emails become tasks. If you're not sure what a connector or a language model is, we explain it in the guide to AI for SMEs.

A small law firm would see it clearly.

Sort: Ask the AI: “Read the unread client emails. Split them into two groups. At the top, the ones that can be answered in two lines. At the bottom, the ones that ask for work.” That way you see first what you can close today.
Answer the ones at the top: Ask the AI for the reply to each one: “Confirm Thursday's appointment and remind them to bring the contract.” The text appears in the conversation. Copy it into your email, check names, dates and deadlines, and send it.
Turn the ones at the bottom into tasks: Ask the AI connected to the ERP: “Create a task from this email, assign it to the employment lawyer and set it for Friday.” The task is saved in the firm's ERP, with an owner and a deadline.

This way, no client email ends the day without a reply or without a task with an owner.

The typical mistake is to turn every email into a task, because the list grows so long that the team stops looking at it. Keep tasks for what really is one.

How do you run effective meetings and summarise them with AI?

Effective meetings are the ones that end with an agreement assigned to a person and a date. That way, the agreed work reaches the client on time. For them to be effective with artificial intelligence, those agreements have to be said out loud: that way the summary picks them up and they don't get lost afterwards.

AI works in two steps. First it transcribes the recording, that is, it turns it into text and separates the voices. Then the model reads that text and extracts the agreements. That's why it only picks up what was said. If nobody said who or by when, it can't extract it, and sometimes it fills the gap with whatever is most likely.

Every meeting ends with each agreement said out loud, with a name and a date, and a person reviews its summary before sharing it.

Example

A small wedding planning business

For their Monday meeting, whoever coordinates would first ask the AI for the tasks due that week, grouped by owner. That list would serve as the agenda. At the end, each person would repeat out loud what they're responsible for and by when: “I'll confirm the menu with the caterer before Thursday”. Afterwards, the AI would pull out the agreements and flag any without an owner or a date. Before moving them to the ERP, someone would compare them with their notes.

Tips to make your recordings useful to the AI:

Put someone in charge of the transcription. The most common problem is that nobody remembers to turn it on. Ask one person to start it when the meeting begins. If your video-call tool allows it, leave it on by default.
Say the name of whoever is in charge, because the AI separates the voices but doesn't always know whose each one is. “Lucía will confirm the menu” has a name attached; “I'll handle it” doesn't.
Say the full date, since “Thursday, before twelve” can be checked and “next week” forces the model to guess.
Speak one at a time and close to the microphone, because when two people talk at once the transcription mixes up sentences and agreements get lost.
Let everyone know before recording, at every meeting. A voice that identifies a person is personal data. That's why the General Data Protection Regulation (GDPR) requires you to inform anyone being recorded. Notice given for one meeting doesn't cover the next.

If the meeting is on Teams, check which transcripts the connector can reach. It may only access those from meetings organised by whoever connected it, and only if transcription was turned on.

How do you organise the work in your company each morning?

Work organisation in a company is decided in the first hour of the day. If nobody decides it, it's set by the order in which emails arrive. When each person starts with their own priorities, those first hours go to the clients with a deadline looming.

If your day starts with firefighting, planning seems like a luxury. However, the slow part of planning is gathering the information. And AI can do that part by reading your calendar, your email and the tasks in the ERP. What's left for the person is what the AI can't do for them, which is choosing.

How do you set up the board and the roadmap in the ERP?

A board by stages helps you choose the morning's priorities knowing where each job stands. It's a view in the ERP made of columns, one per stage, and each task is a card that moves from one column to the next. You set it up once, in three steps:

Create the stages. For example: to do, in progress, waiting for the client and ready to deliver. With few stages, everyone knows where to move their card.
Fill in each card. Give it an owner, a deadline (the day the delivery is due) and the client's name, because that's what the AI reads later.
Sort by date. The roadmap is those same cards sorted by deadline up to the end of the month. A card without a date doesn't appear in it or in the list the AI prepares for you each morning.

That way, the AI connected to the ERP can read the cards by stage and date every morning. With that, it lists what's due this week and what's already overdue without having been delivered. Then it moves or assigns the cards when you ask it to.

How do you increase productivity in the company and know whether it's working?

To increase productivity in the company, change just one rule and measure before and after. If you change the tool, the rules and the schedules in the same week, you won't know what worked. In the first week you count without changing anything, and in the second you try out the rule.

Ideally you would measure how long each task takes, but we understand that in many teams that feels like surveillance. In that case, count what reaches the client, such as unanswered emails or agreements without an owner. And explain to the team beforehand what you're measuring and why.

In this table you have the rules from the guide, each with a prompt to try it out. Start with the row where your team gets most stuck and check everything before sending it.

In this table we give you tips on what to do depending on where your team gets stuck
Where the team gets stuck	Team rule	Prompt for the AI
Unread email that nobody claims	Anything that takes two lines gets answered today	“You are my admin assistant. Read today's unread client emails. Split them into two lists: the ones that can be answered in two lines and the ones that ask for work. For each one, give the client, what they're asking for and whether there's a deadline. Don't add anything that isn't in the email.”
Replies to clients	A person sends every reply	“You are my customer service assistant. Draft the reply to this email in five lines at most. Use the same tone as our last reply to this client, which I've pasted below. Confirm the appointment on Thursday morning and ask them to bring the signed contract. If any information is missing, ask me before writing.”
Emails that ask for work	Anything that asks for work is a task	“You are my admin assistant. Turn this email into a task. Short title with the client's name. In the description, what they're asking for and what documents they've attached. Owner: Ana. Deadline: Friday. If the email asks for several things, create one task for each.”
Meetings that start from memory	No meeting without a purpose	“You are my sales assistant. Before the meeting with this client, read our email thread from the last few weeks. Make two short lists: what we committed to do and what they still have to confirm. Flag whatever has gone longest without a reply.”
Agreements that get lost	No agreement without an owner and a date	“You are the person taking the minutes of this meeting. Read the transcript and pull out only the agreements, not the narrative. For each one: what's to be done, owner and date, exactly as they were said. If the owner or the date is missing, write 'unassigned' instead of guessing.”
Agreements nobody carries out	Every agreement ends up as an assigned task	“You are my coordination assistant. Using these agreements, create one task for each in the ERP, with the owner and date given. Don't create the ones marked 'unassigned': leave them in a separate list for me to decide.”
Mornings decided by the inbox	Everyone brings three priorities	“You are my organisation assistant. Look at my calendar, today's email and my tasks in the ERP. List what's due this week, sorted by date, and my free slots of more than half an hour. Don't choose the priorities: I'll decide them with this list.”

Where is it best not to use AI with email and meetings?

We don't recommend using it where what matters isn't written down, because the model only works with the text you give it.

These are the most common cases:

A complaint with a history: A long-standing client writes in angry. The AI drafts a proper apology, but what you promised them last time isn't in the email it reads.
A price or a discount: An email asking for a reduction doesn't show how much margin that client leaves you, so you decide the figure.
What was agreed verbally: If a deadline was settled with the recording off, the summary doesn't pick it up, and nobody misses it until it falls due.
An email with legal consequences: A formal claim is answered by whoever signs or by your adviser, because that reply commits your company.

It depends on the permission the connector has. If it's read-only, it reads, searches and summarises, but doesn't send or leave drafts in your inbox. The assistant prepares the reply in the conversation. Then a person checks names, dates and tone, copies it into their email and sends it. That way, sending, deciding and signing are still your team's job.

It depends on how each inbox is connected. In AiKit, each person connects their own Gmail or Outlook as read-only, and the assistant works only with that inbox. So whoever coordinates can't see other people's email from there. Company documents, on the other hand, are shared with permissions by department, so each person is connected only to the parts that concern them.

Yes, as long as someone takes notes, because the language model works on text: a recording is transcribed first, and notes are already text. What changes is what they contain. A transcript captures what was said while recording, and notes only what someone wrote down. If the notes don't give an owner or a date, the AI can't pull them out, so whoever takes notes writes who and by when for each agreement.

Business productivity with AI: emails and meetings