A practical way to implement AI in a business is a 90-day plan: two weeks to audit workflows and choose one, four weeks to pilot it on real data, four weeks to roll it out and three weeks to start a second. With Praxen, the first 30-minute audit call is free, and a pilot takes 2–4 weeks and costs £2,900 (£2,030 with the launch offer), excluding VAT.
Which AI use cases come first
The AI use cases in business that pay back first tend to share three features: they happen every day or week, the input arrives in writing or by phone, and the rules fit on a page. Answering enquiries, entering invoices and orders, updating the CRM, drafting routine replies and screening applications all qualify. The plan below takes one of them from audit to a measured result, then starts the next.
A 90-day plan forces one decision at a time. Each stage has an owner, an output and a date, and the second workflow starts only after the first has numbers behind it.
The 90-day plan
Thirteen weeks, split into four stages. The dates assume your systems can be reached through an API or an export; if access depends on a software vendor or an outside IT provider, ask for it in week 1.
- Weeks 1–2: auditShort interviews with the people who do the work. Workflows are scored by hours, volume and risk, one is chosen, and its metric and current figure are recorded. The usage policy is drafted alongside.
- Weeks 3–6: pilotThe chosen workflow is built on your real data and connected to your tools, then tested on last month’s cases. After a week in shadow mode, it goes live with a person approving every output.
- Weeks 7–10: rolloutThe rest of the team starts using it, with a short guide and training. Exceptions are reviewed weekly, and the first full month is compared with the baseline.
- Weeks 11–13: second workflowThe next workflow from the audit list is built on the same connections, knowledge base and access rules. The sponsor reviews the first one’s results and decides what follows.
Who does what
- Sponsor: a director who picks the first workflow, signs off the budget and clears blockers. About an hour a week.
- Workflow owner: the person who knows the process best. They explain the rules, check test results and approve go-live, which takes a few hours a week during the pilot.
- Approvers: the staff who check AI drafts in daily work. They need training and a simple way to report errors.
- IT and data contact: grants access to systems, checks security settings and keeps the list of approved tools.
- Implementation partner: builds, tests, documents and hands over. At Praxen, one named lead runs the project from audit to launch.
Agree a usage policy in the first fortnight
Staff are often using AI tools before the company has decided anything about them. One or two pages, agreed during the audit, set the rules for the pilot and for everyday use. They should cover five points:
- Approved tools: which ones, on which business accounts and with which settings switched on.
- Off-limits data: personal details of customers and staff, health records, passwords and figures not yet published.
- Review: who reads AI output before it reaches a customer, a supplier or a regulator.
- Errors: how people report a wrong answer and who corrects it.
- Ownership: the person who keeps the policy current and signs off any new tool.
How to measure the result
- Record a baseline in week 2: the current figure for your metric, taken from an ordinary week.
- Pick one main metric, such as minutes per invoice, time to first reply or calls answered out of hours.
- Add one quality measure, usually the share of AI drafts approved without changes.
- Measure again at the end of the pilot and after the first month of rollout, the same way each time.
- Turn the time saved into money using the real hourly cost of the people involved, then set it against the build and monthly costs.
What the 90 days cost
Most of the plan runs on fixed prices. Before the launch discount, training is £850, a pilot £2,900, a custom solution from £6,500 and care £350 a month; the cards show launch offer prices, all excluding VAT. AI usage goes on your own provider accounts at cost, and we estimate it before the build starts.
Common mistakes
- Buying the tool first: licences bought before a workflow is chosen leave people with an empty chat box and no reason to open it.
- Three pilots at once: results blur together and nobody owns any of them.
- No baseline: without last month’s figure, nobody can say whether the pilot worked.
- Invented test cases: last month’s real cases include the awkward ones that break a polished demo.
- Removing approvals too early: keep a person on every output until a month of data shows where it is safe to stop.
- No owner after launch: when a price list or policy changes and nobody updates the rules, the answers go out of date.