Advice is the hardest thing in the world to sell twice.
Most products get made once and shipped a thousand times. Advice needs a person in a room giving their full attention to one other person for an hour, and then it needs them to do it again tomorrow. That was the business MidShift was in when the work started, and it’s the same shape as a lot of small and medium businesses: the thing customers pay for only exists while a member of staff is doing it.
They went from an idea to a live platform. Over 20,000 professionals have now been through it. This is what got built, what stayed human, and why the second part matters more than the first.
If you want the wider picture before the specifics, start with AI for small business: the complete guide. This post is one worked example of what it describes.
The arithmetic of one person at a time
Careers guidance in the UK has a supply problem, not a demand problem.
Research from Phoenix Insights and Standard Life puts around 4.5 million UK adults in the position of wanting to change jobs, being at risk in their current role, or being very unhappy at work. Against that, only 15% of 45 to 54 year olds have had careers advice in the last 3 years, and 51% of that age group have never heard of any careers advice service at all.
So the people who need it mostly don’t get it. And when they do go looking, the market price of an hour of someone’s attention is what you’d expect. UK career coaching runs between £75 and £150 an hour, averaging about £100, from £63 a session in the North East up to £125 an hour in London.
That’s not a criticism of the coaches. An hour of skilled attention is worth what it costs. It’s just that the unit of delivery is an hour of a human being, and there is a hard limit on how many of those exist.
You’ve probably got a version of this. The surveyor who has to write the report. The accountant doing the year-end review. The recruiter reading the CVs. The consultant preparing for the meeting. The bit the customer pays for sits at the end of a lot of preparation, and the preparation is where the hours go.
What we built
The brief was to get from a concept to a working platform quickly, and to make the guidance itself do the work. Not a library of articles with a search box on it.
The engine takes what somebody already has (their experience, their current role, where they want to get to) and produces a personalised career plan for that specific person. Not a template with their name at the top. It works out where the gaps are between where they stand and where they’re aiming, what they need to learn to close each one, and what order to do it in. Skill gap analysis, then a plan with concrete steps against it.
On top of that sits 1:1 mentorship. A real person, having a real conversation.
That combination is the whole design. The software does the part that used to eat the mentor’s preparation time, and the mentor walks in already knowing this person’s situation, gaps, and options. The conversation starts where it used to finish.
The line we drew
This is the part worth copying, whatever business you’re in.
The AI does the preparation. It does not make the decision.
Nobody’s career gets decided by the engine. It reads a situation, sorts it, and lays out the options with the reasoning attached. The person, with their mentor, decides what to actually do. That’s deliberate, and it’s the same line we draw on every build: automate the half that repeats identically, route the half that needs judgement to a human being.
The test we use is simple. Can somebody check the output at a glance and know whether it’s right? A skill gap between “operations manager” and “head of operations” is checkable in seconds by anyone who knows the field. “Should this person leave their job” is not. One of those is safe to generate at volume. The other is a conversation.
We’ve written the general version of this test in what AI can actually do for a small business (and what it can’t). MidShift is what it looks like when you apply it honestly and accept that half the work stays with people.
The numbers
Over 20,000 professionals served. More than 1,000 AI-generated career plans. And 92% faster progression through the process it replaced.
That last figure is MidShift’s own measure against their previous way of working, not a controlled study, and it’s worth saying so plainly. The first two are counts, which is why they carry more weight.
The 20,000 is the number that actually tells the story. Work out what it would have taken to deliver that by hand. At an hour of guidance each, that’s 20,000 hours. A full-time UK working year is somewhere near 1,750 hours once you take out holiday, so you’re looking at roughly 11 people doing nothing else for a year. At the £100 market average, it’s about £2m of chargeable time.
That business was never going to exist. MidShift would have been glad to hire 11 careers advisers. The trouble is that a service carrying 11 salaries stops being something an ordinary person can afford, so the people in that 4.5 million never get near it.
The platform didn’t make the old service faster. It made a different service possible.
Three things to take from this
1. The expensive half was the repeatable half.
The mentors weren’t the bottleneck. The preparation in front of every mentoring hour was. Same in your business: the costly part is rarely the skilled bit at the end, it’s the identical groundwork that happens before it, every single time, and that nobody has ever put a number on.
Put a number on yours. We’ve laid out the method in the cost of manual work, and the manual work cost calculator does the arithmetic if you’d rather not.
2. It shipped because it changed the shape of the work.
MidShift didn’t bolt AI onto an existing process and hope. The process was redesigned around what the software could do reliably, with the mentor moved to the point where a person adds most.
That distinction is most of the difference between projects that work and projects that quietly stop. McKinsey’s State of AI research found only about 6% of organisations qualify as AI high performers (attributing 5% or more of EBIT to AI use), and what separates them is that they rebuild workflows around AI rather than layering it on top. Everyone else adds a tool and keeps the old process running underneath. We’ve gone through why that fails in why most small business AI pilots never ship.
3. It was new capacity, not fewer people.
Nobody lost a job. The mentors are still mentoring, and there are more people to mentor. The work that went away was the work nobody wanted: assembling the same background pack over and over so the useful conversation could start.
That’s the honest version of what this does in a small business. You don’t cut the team. You stop paying skilled people to do unskilled preparation.
Where the pattern fits, and where it doesn’t
It fits when the process runs often, the shape is the same each time, and somebody can eyeball the output and tell if it’s wrong. MidShift ticked all 3. So did Founderise, where a 4-stage coaching framework became a self-serve platform in 9 weeks and took 12 hours a week off the founder.
It doesn’t fit when the work changes every month, when every job is genuinely bespoke, or when the task is the last human contact a customer has with you. Build on ground that keeps moving and you’ll pay to build it twice.
Size is the main thing standing between most UK businesses and this. ONS figures put 29% of UK businesses using at least one AI technology in June 2026, rising to 49% among those with 250 or more staff. The bigger you are, the more likely you’ve started. Which is backwards, because the smaller you are, the more one person’s week is worth to you.
Finding your version of this
You’re looking for the preparation that happens before the valuable thing, not the valuable thing itself.
Work backwards from whatever your customers pay for. What has to happen before someone can do it? How often does that run, how long does it take, and how much of it is identical every time? The candidate you want is boring, high volume, and currently invisible because it’s buried inside jobs people are already being paid for.
Rank what you find before you build anything. Which processes to automate first sets out how, and the systemisation scorecard will sort a list for you in a few minutes. There’s more on how we work on the AI consultancy page.
MidShift didn’t begin with an AI strategy. It began with one process that repeated too many times to keep doing by hand.
Related reads
- Shadow AI: what your team is already doing with company data
- AI agents for business: what they actually do
- AI for small business: the complete guide
- What AI can actually do for a small business (and what it can’t)
- The cost of manual work: what repetitive admin really costs
- How much does it cost to add AI to a small business?
- Which processes to automate first (and how to rank them)
- Why most small business AI pilots never ship
- Case study: how Founderise automated delivery
- How to choose an AI consultant (and the red flags)
Got a service that only works while somebody is doing it?
That’s the shape the AI audit is built for. Two weeks, fixed price agreed in writing before anything starts, every repeatable process in your business mapped and costed in hours and pounds, ranked by what to do first, plus one working proof of concept running on your own data. If you go ahead with a build within 90 days, the audit fee comes off it in full.
Book a free call and we’ll find the process in your business that’s doing what MidShift’s was.