How to adopt AI at work: a practical 90-day plan for growing businesses
Econy Editorial Team · · 7 min read

Key takeaways
- Start with two or three repetitive, text-heavy workflows, not with a tool.
- Write a one-page AI policy before the pilot: what data can be used, who reviews output, what is off-limits.
- Keep a person responsible for every AI output that leaves the team.
- Measure time saved and rework, not 'number of prompts'.
- Scale only what worked, and redesign the workflow around it.
Almost every company now has someone experimenting with AI. The harder question is how to turn scattered experiments into everyday work that is faster, safer and better. McKinsey's State of AI survey found that nearly nine in ten respondents report regular use of AI in at least one business function — yet far fewer see a clear impact on the bottom line. The difference is rarely the model. It is whether the company changed how the work is done.
This guide gives founders, operations leaders and team leads a practical 90-day plan to adopt AI at work. It assumes you have no AI team, limited time and real customers to look after.
Why most AI pilots stall
Pilots usually stall for predictable reasons:
- They start with a tool. Someone buys licences and asks people to "find uses". Without a target workflow, usage spikes and then fades.
- Nobody owns the output. AI drafts a reply, a summary or a plan, but it is unclear who checks it. Mistakes slip through and trust drops.
- Data rules are unclear. People either paste sensitive information into public tools or avoid AI entirely because they are not sure what is allowed.
- Success is not measured. Without a baseline, it is impossible to say whether the pilot saved time or simply added a new step.
The plan below addresses each of these in order.
Days 1–15: choose the work, not the tool
Begin with a short list of workflows where AI is genuinely good today: turning messy input into a first draft, summarising long threads, classifying incoming requests and extracting information from documents.
Good first candidates in most growing companies:
- Drafting — job descriptions, policy drafts, client updates, follow-up emails.
- Summarising — long support tickets, meeting notes, project status.
- Triage — suggesting a category and priority for incoming requests.
- Screening — comparing résumés against the requirements you wrote.
For each candidate, write down three things: how often it happens each week, how long it takes today, and what "good" looks like. Pick two or three workflows that are frequent, text-heavy and low-risk if a draft is wrong — because a person will review it anyway.
Days 1–15: write a one-page AI policy
Before anyone starts, agree on a short policy. It does not need to be a legal document. It needs to answer:
- What data may be used? For example: internal project information yes; customer personal data only in approved tools; passwords and payment details never.
- Who reviews output? The person who would normally do the task stays responsible for the result.
- What is off-limits? Decisions about people (hiring, performance, pay) are never made by AI alone.
- Where are approved tools? List them, so people stop guessing.
The US National Institute of Standards and Technology publishes a free, voluntary AI Risk Management Framework organised around four functions — govern, map, measure and manage. You do not need to implement it in full, but its structure is a helpful checklist for what your policy should cover as you grow.
If you employ people in India, remember that the Digital Personal Data Protection Act, 2023 applies to the personal data you process; the DPDP Rules were notified on 14 November 2025 with an eighteen-month period for phased compliance. Take advice on your specific obligations.
Days 16–45: run a focused pilot
Run the pilot with a small group — ideally the people who do the work every day. Give them:
- A clear brief for each workflow ("Draft the first version of the job post from the hiring manager's notes").
- A place to record when the AI output was used as-is, edited, or discarded.
- A weekly 20-minute check-in to share what worked.
Keep the pilot inside the tools where the work already happens. Copying text between a chat window and your project tool adds friction and loses context. AI that works next to the record — the ticket, the candidate, the project — has the information it needs and leaves a trail.
What to measure
Measure outcomes the business cares about:
- Time to first draft for each workflow, compared with your baseline.
- Edit rate — how much of the AI draft survived review.
- Rework — mistakes found later that the reviewer missed.
- Adoption — how many eligible tasks actually used AI.
Avoid vanity metrics such as the number of prompts. They say nothing about whether work improved.
Days 46–75: redesign the workflow around what worked
This is where most of the value appears. In McKinsey's survey, nearly three-quarters of the highest-performing organisations report fundamentally redesigning workflows because of their AI use. In practice, redesign usually means small, concrete changes:
- The hiring manager's intake form now captures exactly what the AI needs to draft a good job post.
- Incoming support requests are triaged with an AI suggestion, but a person confirms before routing.
- Project leads get an AI summary of the week's activity every Friday and edit it into the client update.
Write the new workflow down in one page per process. Who triggers it, where the AI step happens, who reviews and where the result is stored.
Days 76–90: scale carefully
Scale only the workflows where the numbers improved and reviewers trust the output. For each one:
- Train the wider team with real examples from the pilot.
- Add the workflow to your onboarding material.
- Keep the review step — especially for anything that reaches customers or affects people decisions.
- Review the policy every quarter as tools and regulations change.
Then pick the next two or three workflows and repeat.
Common mistakes to avoid
- Automating judgement. AI can draft a performance summary; a manager must own the assessment.
- Hiding AI use. Tell customers and employees where AI assists. Transparency builds trust.
- Skipping the baseline. If you cannot say how long a task took before, you cannot prove the improvement.
- Spreading too thin. Ten half-adopted use cases deliver less than three that are embedded in daily work.
A worked example: a 40-person agency
Consider a 40-person digital agency with three teams: delivery, sales and a small operations group that also handles HR. Here is how the 90-day plan might play out.
Days 1–15. The operations lead lists candidate workflows and times them for a week. Two stand out: weekly client updates (project leads spend about an hour each per client every Friday) and first-draft job posts (the hiring manager and HR go back and forth for days). Both are text-heavy and already reviewed by a person. The team writes a one-page policy: client data may be used only in the approved workspace, every client update is reviewed by the project lead, and no AI is used to rank candidates.
Days 16–45. Five project leads pilot AI-drafted client updates built from the week's project activity. They record whether each draft was used as-is, lightly edited or rewritten. HR pilots AI-drafted job posts from the hiring manager's notes.
Days 46–75. The numbers show client updates taking a fraction of the time, with most drafts needing light edits. The team redesigns the workflow: project leads now log decisions and milestones as they happen, because better input makes better drafts. Job posts improve once the intake form asks for must-have and nice-to-have skills separately.
Days 76–90. Both workflows roll out to everyone. The next candidates — ticket summaries for the support rota and policy drafts for HR — enter the pilot.
The lesson is not that AI wrote the updates. It is that the team changed how they capture information so AI could help reliably.
A checklist you can copy
- ☐ Three candidate workflows listed, with weekly volume and current time per task
- ☐ One-page AI policy agreed and shared
- ☐ Approved tools listed; everything else off-limits for company data
- ☐ Pilot group chosen from people who do the work today
- ☐ Baseline measured before the pilot starts
- ☐ Weekly 20-minute check-in booked for the pilot period
- ☐ Review step defined for every AI output that leaves the team
- ☐ Decision at day 45: scale, change or stop each workflow
Where Econy fits
Econy is built around the principle that AI assists and a person decides. AI features sit next to the work they help with: drafting job posts and offer letters in hiring, suggesting replies and summarising tickets in the service desk, breaking a scope into a first project plan and drafting HR policies. Every suggestion is reviewed by a person before anything is saved or sent, and AI only works with information the person using it is allowed to see. You can read more about how AI works in Econy.
Conclusion
Adopting AI at work is less about choosing the perfect model and more about choosing the right work, setting clear rules and measuring honestly. Start small, keep people in charge of the output, and redesign the workflow around what proves useful. In 90 days you can move from scattered experiments to a few dependable, measurable improvements — and a repeatable way to find the next ones.
Frequently asked questions
What is the best first AI use case for a small business?
Repetitive, text-heavy work that a person already reviews — such as drafting job posts, summarising long support tickets or suggesting a category for incoming requests. Mistakes are caught in review and the time saving is easy to measure.
Do we need an AI policy before we start?
Yes, but it can be one page: what data may be used, which tools are approved, who reviews output and which decisions AI must never make alone.
How do we measure whether AI is helping?
Compare time to first draft, edit rate and rework against a baseline taken before the pilot. Count how many eligible tasks actually used AI.
Should AI make decisions about employees?
No. Use AI to draft or summarise, and keep hiring, performance and pay decisions with accountable people.