Daniel Spence started out automating the newsletter for his own online business at university. He now runs Cybix, an AI consultancy for firms of 10 to 500 staff. He answered our seven questions on why “we already have Copilot” is not a plan, the dull work to automate first, which owner worry is the real risk, and the AI backup that quietly ran his customer emails for two months without anyone knowing.

Who is Cybix?

A UK AI consultancy for firms of 10 to 500 staff. It pairs a business consultant with an AI specialist on each client, works on site to find where AI can take work off people’s desks, and then builds the systems. It is testing Cybix X-ray, a desktop tool for the smallest firms.

Company
Cybix
Website
cybix.ai
Sector
AI consultancy

Daniel Spence answered our questions in writing, and the answers below are his own words.

From one newsletter to a consultancy

Cybix is a UK AI consultancy. We help firms of 10 to 500 staff find the work AI should take off their desks, then we build it. Our clients range from manufacturers and estate agents to names like ITV and AO. Every client gets a pair: a business consultant of Big Four calibre and an AI specialist who builds the systems. We spend time on site and talk to everyone, from the MD to the shop floor.

I started out running my own online business while at university. I automated my own newsletter, then ChatGPT arrived and I began building for other online shops. That taught me every business has different jobs eating its time. A product fixes one job for everyone. We wanted to find the right jobs for each business first.

“We already have Copilot”

We asked what state firms are in when they first call, and what they most often get wrong.

Busy, keen and unsure where to start. Most want to grow, but their team has no spare time. They see AI everywhere and know they should act, but there is no plan. Often one person inside, sometimes a younger manager, has been pushing for it.

The most common mistake is thinking AI means a chat tool. “We already have Copilot” comes up on a lot of first calls. A chat window helps, but the big wins come when AI does whole jobs inside the systems you already use.

The second mistake is waiting. Many owners think they need to replace their software first. Most don’t. AI can work with what you already have, even older systems. Waiting for the perfect moment just means another year of the same manual work.

Start with the copy-and-paste work

For a firm of 10 to 50 people with no technical staff, we asked where AI genuinely pays back first.

Start with the copy-and-paste work: quotes, chasing invoices, typing the same details into two systems, month-end packs, answering the same customer questions. It is dull, it is in every business, and AI now does it well.

The other early win is a safe company AI workspace. Give everyone a business AI account, loaded with your own documents and rules, so staff stop pasting company data into free personal tools.

Where not to bother yet: anything that talks to your customers with no person checking it first, and big custom systems before you have proved one small win. Start with one job, measure the time it saves, then grow from there.

If it will not pay for itself, don’t do it.

One newsletter, seven languages

An online retailer we work with had seven people writing its customer newsletters. They sell into several countries, but the newsletters only went out in one language.

We built a system where AI writes each newsletter in seven languages, personalised to each reader, with the right links for each country’s website. What used to take a team now takes seconds.

The biggest change was the people. Six of those seven staff moved onto growth work: YouTube, social media and other projects that grow the business. So in numbers: one language became seven, and six people went from writing the same email to winning new customers.

The backup nobody heard about

We asked about something that went wrong. His answer is the most useful thing in the interview for any business running AI.

One of our own businesses uses AI to help handle customer emails. We built in a backup: if the main AI model failed, a second one would step in. Then the main provider quietly switched off the version we were using. The backup took over, everything kept running, and the logs said “success”. It ran like that for about two months. We only found out when the provider emailed to ask why we had stopped using them.

Customers never noticed, because the backup did its job. But we had been paying a different supplier for two months without knowing.

The lesson: a backup nobody hears about is a hidden fault. Now every backup we build has an alarm on it, so a person knows the moment it kicks in.

It is a failure with no visible damage, which is exactly what makes it dangerous. Anything that switches itself over silently, from an AI model to a backup broadband line, should tell a person when it does.

Data is the real risk, staff is mostly noise

Data is the real risk, but not in the way most owners think. The danger is usually not the AI itself. It is staff already pasting customer details and company files into free personal AI accounts, because nobody has given them a safe option. The fix is simple: business accounts, clear rules on what can go in, and personal accounts switched off.

Accuracy is real but manageable. AI does make mistakes, so you plan for it. AI drafts, and a person checks anything that matters before it goes out.

Staff is mostly noise. In our experience the team does not shrink. People stop doing the dull work and move onto the work they were hired for. The firms that do best bring their staff in early, because those people know exactly where the time goes.

Five questions to ask an AI consultant

Five questions. Will you spend time with my people before you recommend anything? Do I own what you build, or am I renting it? Do you take commission on the tools you recommend? What happens when something breaks, and who finds out? And can you show me the sums, so I know it will pay for itself? If they can’t answer all five clearly, keep looking. For what it’s worth, we take no commission on any tool, and our clients own what we build.

Next for Cybix: bringing what we do to the very smallest firms. Days on site don’t make sense for a five-person business, so we have built Cybix X-ray, a desktop tool that maps how someone works over a week and shows where AI could save them time. We are testing it now.

The five questions work just as well put to any consultant, Cybix included. The one about commission is the one owners least often think to ask.

SME Leaders is Smart SME’s series with the people who own and run UK small businesses. Answers are published in the subject’s own words, with cuts only for length. If you run a business and would like to take part, write to press@smartsme.co.uk.