What does AI consultancy involve?
An AI consultant starts by looking at how the work runs today: which team uses which software, how many people a job passes through, and where it waits or goes wrong. AI is then fitted onto that picture. Doing it the other way round, choosing a tool first and then looking for work to give it, is where projects most often stall.
A typical consultancy engagement covers:
- Process analysis: a map of the repetitive work in sales, marketing, HR and operations.
- Prioritisation: how long each job takes, how often it repeats and what a mistake costs.
- Model and infrastructure choice: cloud models such as OpenAI, Google Gemini or Anthropic Claude, or open-source models running on your own servers?
- Data and security decisions: which data is processed where, who can see it and how long it is kept.
- A pilot: a first, measurable set-up running on a single process.
- Training and roll-out: teams adopt the new flow, and other departments follow as results come in.
Which processes are suitable for AI?
For the first job you hand to AI, look for four things. It should repeat often, be heavy on text, documents or images, have rules that can be written down, and any mistake should be caught at a human approval step.
| Process | What AI does | What stays with people |
|---|---|---|
| Preparing quotes | Reads the enquiry, pulls out products and quantities, prices them by your rules | Approving the price |
| Handling emails and forms | Classifies, summarises, sets urgency, drafts a reply | Sending the reply |
| Order and form entry | Copies details from documents into ERP fields | Checking exceptions |
| Reporting | Gathers data, drafts a summary and commentary | The decision |
| Internal questions | Searches company documents and answers with the source | Keeping documents up to date |
One-off strategic decisions, work with no data behind it and steps that cannot be undone are not good first projects. In that kind of work AI can at most prepare; the final word stays with a person.
Where should a company start with AI?
- Pick one process. Choose the job people complain about most and that repeats every day. Don’t try to change the whole company at once.
- Map how it runs today. Who does what, in which software and in what order? Note how long one job takes on average.
- Split the steps. Which steps will AI prepare and which will a person approve? Write the approval points down at the start.
- Decide on data and access. Be clear about the data AI will read, the systems it will connect to and the permissions it needs.
- Run a small pilot and measure it. Define the measures before you begin: time per job, error rate, time to reply to the customer.
- Extend as results come in. Once the first process has settled, take the same method to the next one.
How do you measure the result?
An AI project is judged by the difference it makes to the process, not by the name of the model. The most practical measures are time saved per job, the share of outputs that need correcting, and time to reply to the customer. Measuring these for a week before the pilot makes a fair comparison possible afterwards.
What should you look for in an AI consultant?
- Model independence: can they choose a model to suit the job, or are they selling one product?
- Integration skills: can they connect to your CRM, ERP and other software, or do they only write reports?
- Data security: do they set out in writing where your data will be processed?
- Going live and support: who monitors and updates the system and trains your team?
- Sector experience: do they know your sector’s data and language?
Frequently asked questions
Who is AI consultancy for?
It suits companies of any size that have daily, repetitive work built around documents and correspondence. Small businesses usually start with a single process such as quoting, order entry or customer emails.
Do we need to buy new software before an AI project?
No. The process and the data are reviewed first, then a solution that connects to your existing systems is chosen. In most cases the software you use stays in place.
Will our data be sent to AI models?
That is a decision made together at the start of the project. For sensitive data, cloud services under enterprise contracts or open-source models running on your own servers can be used.
This guide was prepared by the dipro.uk team and last updated on 16 November 2026. Questions: [email protected].
