AI strategy that starts with your business, not the model.
Most businesses do not lack AI ideas. They lack a way to tell the valuable ones from the expensive distractions. We help you pick use cases, test readiness and plan pilots, advised by engineers who build with AI every working day.
The problem is rarely a shortage of ideas.
By the time a company asks for AI consulting, it usually has a spreadsheet of forty ideas, three vendor demos in the inbox and staff already pasting customer emails into public chatbots. What it does not have is a shared view of which ideas pay back, which ones the data can support, and which ones carry risk nobody has priced.
Good AI strategy is mostly subtraction. We look at where time and money actually go in your business, where errors are costly, and where a probabilistic system is acceptable. A use case that drafts a reply for a human to check is a very different investment from one that approves refunds on its own, even if the demos look similar.
Our advice comes from practice, not slideware. FusionOne engineers build software AI-first, running coding agents in parallel, generating tests and reviewing AI-written code daily. That gives us a grounded sense of what current models do reliably, where they still break, and what it really costs to take a pilot into production.
Not all AI use cases carry the same risk
We sort candidate use cases by how much they are trusted to do. The further right a use case sits, the more data, testing and governance it needs before it earns its keep.
| Assist a person | Decide within rules | Act on its own | |
|---|---|---|---|
| Typical examples | Draft emails, summarise calls | Classify tickets, flag anomalies | Issue refunds, update records |
| Cost of a wrong answer | Low, a human edits it | Moderate, caught downstream | High, reaches customers or ledgers |
| Data needed to start | Little beyond good prompts | Labelled historical examples | Clean data plus system access |
| Time to first value | Days to weeks | Weeks | Months, with staged rollout |
| Where to begin | Usually the right first step | Second wave, once measured | Only with proven guardrails |
Many successful AI programmes start in the left column, gather evidence and move right one step at a time.
An honest AI readiness assessment asks these questions
Readiness is less about having a data lake and more about whether a specific use case has what it needs. We score each shortlisted idea against the points below.
Is there a measurable baseline today?
If nobody knows how long invoices take to process now, nobody will be able to prove the AI version is better.
Does the data exist, and can you legally use it?
Check where the records live, how messy they are, and whether customer contracts or privacy law restrict sending them to a model provider.
Is there a named business owner?
Pilots owned only by IT tend to stall. Someone accountable for the process must want the change and define "good enough".
What error rate can the process tolerate?
Agree this before building. It decides whether you need human review, a narrower scope, or a different approach entirely.
Would a simple rule or an off-the-shelf tool do?
A surprising share of "AI" requests are solved by a report, a form or a feature already in software you pay for.
Who handles it when the model changes?
Providers update and retire models regularly. Someone has to re-test prompts and outputs when that happens.
A typical AI advisory sprint
Most AI consulting engagements run as a two to four week Advisory Sprint. Larger organisations sometimes follow it with an Ongoing Partnership to oversee delivery.
Map the work
Week 1Interviews with the people doing the work, a look at systems and data, and a long list of candidate use cases gathered from every department.
Score and shortlist
Week 2Each idea is rated on value, feasibility, data readiness and risk. We quickly prototype the top few with real examples to see how current models cope.
Design the pilots
Weeks 2 to 3For the chosen one or two use cases: success metrics, a test set of real cases, build or buy recommendation, rough running costs and a kill criterion.
Roadmap and guardrails
Weeks 3 to 4A sequenced AI roadmap, a short acceptable-use policy for staff, and a clear view of who owns what once the sprint ends.
Decisions you can act on, not a trend report
A ranked use-case portfolio
Every idea scored on the same criteria, with the reasoning written down so your team can challenge it and revisit it later.
Build, buy or wait, per use case
Many needs are met by features in tools you already license. We say so, and reserve custom work for where it creates an advantage.
A practical AI usage policy
Plain rules on which tools staff may use, what data must never be pasted into them, and how to report problems.
AI-native delivery advice
Guidance on bringing AI coding agents and AI-assisted review into your own development team, including the security and IP questions to settle first.
Questions we often hear
What does an AI consultant actually do?
An AI consultant helps you decide where AI will create real value, whether your data and processes are ready, and how to test ideas without wasting months. A good one also tells you when a problem does not need AI at all. At FusionOne the same people can then help build or oversee the chosen solution.
How much does AI consulting cost?
It depends on the size of the organisation and how many departments are in scope. Our AI consulting usually runs as a fixed two to four week Advisory Sprint, and after a free 30-minute discovery call we send a clear proposal within 48 hours so you know the cost before committing.
How do I choose the first AI use case for my business?
Pick something frequent, measurable and low-risk, where a human still checks the output: drafting, summarising or classifying are common starting points. Avoid starting with anything that acts on customers or money without review. Our AI use case finder can suggest ranked options by industry and department.
Do we need our own data scientists before starting with AI?
Usually not. Most business AI today is built on existing models accessed through APIs, which calls for solid software engineering and good evaluation rather than a research team. You do need someone internal who owns the business process and can judge whether the output is good enough.
How does your own use of AI affect your consulting?
Our engineers work AI-first every day, with multiple coding agents running in parallel, AI-assisted code review and generated tests. Our open-source CLI, perchd, grew directly out of running agents side by side. That hands-on use keeps our advice grounded in what current models reliably do, rather than what vendor demos suggest.
Related reading and tools
Tell us where AI keeps coming up.
Share the ideas on the table, the vendors pitching you and what worries you about getting it wrong. Within 24 hours we will reply with an honest view on where to start, and where not to.
Working with companies globally · Response within 24 hours