Phase 1: your first AI assistants
Phase 1 puts an AI assistant in every team, connected to the tools that team already uses, and switches on the first background automations. It is quick and cheap to prove, and a person checks every output.
Your first AI assistants should sit inside the tools each team already uses and take on its most repetitive writing, sorting and summarising. Add background automations for fixed, rule-based tasks. Start with one system, one workflow and one measured win, keep a person checking every output and expand only once the pilot proves itself.
What should our first AI assistants and automations be?
Your first AI assistants should sit inside the tools each team already uses and take on that team's most repetitive writing, sorting and summarising. Your first automations should handle fixed, rule-based tasks in the background, such as filing, reminders and logging.
This is Phase 1 of the AI-Native Roadmap, and it comes after Preparation. It is where most organisations see their first measurable return, and where staff build the habits every later stage depends on.
Adoption is rising fast, and the gains are not automatic. The government's AI adoption survey 2026 (opens in a new tab) found that one-third of UK businesses were using at least one AI technology, 32% compared with 16% in 2025. Half of adopters (50%) reported an overall increase in productivity, gains the survey describes as generally modest.
Phase 1 moves every team one step towards the goal the roadmap names: An AI-native business is one where AI assistants, department brains and agents do the repetitive work in every team, with people keeping the judgement.
What does an AI assistant do for each team?
An AI assistant drafts, summarises, sorts and answers for one person at a time, using the documents and systems it is connected to. The person asks, the assistant produces a first version, and the person checks and finishes it.
| Team | What the AI assistant does | What the person still does |
|---|---|---|
| Sales | Drafts follow-up emails and proposals from call notes | Decides the offer and the price |
| Customer support | Suggests replies to common questions from the help articles | Checks the reply and handles complaints |
| Marketing | Drafts posts, newsletters and first versions of copy | Sets the message and approves what goes out |
| HR | Drafts job adverts, policies and letters from templates | Makes every decision about a person |
| Operations | Summarises meetings and turns notes into action lists | Agrees priorities and owners |
| Fundraising (charities) | Drafts thank-you letters and first versions of grant applications | Builds the relationship with each funder |
| Programmes (charities) | Summarises case notes and drafts reports for trustees | Works with beneficiaries and judges outcomes |
The evidence for assistants is strong. A study of 5,179 customer support agents published by the National Bureau of Economic Research (opens in a new tab) found that those given an AI assistant were 14% more productive on average. In the UK government's Microsoft 365 Copilot trial (opens in a new tab), civil servants saved an average of 26 minutes a day, and the trial found training was essential to getting those benefits.
A secondary school shows what this looks like day to day. The office drafts parent letters in minutes, the head of year summarises behaviour notes before a meeting and a teacher turns a scheme of work into a parents' guide. Every output is read by the person who sends it.
What are the first background automations?
The first automations are rule-based tasks that run on their own: when this happens, do that. They do not think or make choices, which is exactly why they are safe to start with.
- Save every invoice that arrives in the accounts mailbox to the right folder, with a consistent name.
- Send a renewal reminder 30 days before a membership or contract ends.
- Log every website enquiry in the customer system and tell the right person.
- Turn a recorded meeting into a summary and file it with the project.
The difference between an automation and an agent matters. An automation follows fixed steps; an agent decides which steps to take within limits a person sets. Phase 1 uses automations. Agents that reason arrive in Phase 2, once the department has data and habits they can rely on.
A sports club might start with a single automation: when a membership payment comes in, update the register and send the welcome pack. Nobody misses it until it stops. The workflow automation page goes further into how these are scoped and built.
Why start with the tools we already have?
We always start with existing tools, because buying a proven platform is faster and lower-risk than building custom automation. Most office suites and business systems now include an AI assistant, or can add one cheaply.
Starting with what you already pay for has three advantages. Staff already know the tool. The data is already in it, with permissions already set. And if a pilot fails, you have lost a few weeks, not a development budget.
The choice of assistant matters less than the habits around it. A team that knows how to ask well, checks every output and shares what works gets more from any tool. Hartz AI's ChatGPT training and the ChatGPT for Teams workshop build those habits in a single session.
Bespoke build has its place, but not here. It enters the roadmap in Phase 2, at the point where a platform cannot do the job.
How do we pick the first workflow?
Pick one system, one workflow and one measured win. The best first workflow happens every week, takes real time, uses a single system and has a result anyone can check.
Measure it before you start. Time the task for a fortnight, note how often it goes wrong and agree what success would look like. Then pilot the assistant or automation with the people who do the work, and measure the same things again.
Keep the scope small with the 30% rule, as Hartz AI uses it. If a proposed system has to automate far more than a third of a process a person still signs off, the scope is too wide and the pilot should be split.
The getting started with AI guide has more examples of good first workflows for small teams.
What does ready for Phase 2 look like?
A team is ready for Phase 2 when most of its people use their AI assistant every week, the first automations run without anyone watching and the questions people ask need information the assistant cannot see.
That last sign is the important one. When a fundraiser asks "what did we send this trust last year?", or a support agent needs the full history of a customer, a general assistant runs out of road. Answering those questions across a department's own data is the job of a department brain.
Phase 1 never ends for good. New staff need training, new tools arrive and some teams move on while others stay. Phase 1 succeeds when the assistant is part of the job rather than an experiment.
Common questions
Frequently Asked Questions
Find your first AI workflow
A discovery call picks one system, one workflow and one measure, so your first pilot proves something.