12For your size
AI implementation for a mid-sized company that already has a stack
A mid-sized company rarely has an AI problem. It has a pilot that worked, a stack that does not talk to itself, and no clear owner for the gap between those two things. That is the problem this page is about.
01
AI that started in one team and never spread
If a tool is already running inside one team, and a year later it is still running inside that one team, this page is written for you. Adoption climbs with company size. The clearest published split is between the largest firms and the smallest ones, and a mid-sized company sits between the two, so the figure beside this paragraph describes the ends of the range rather than your exact position on it. The useful part is the direction, not the number. By the time a company reaches your size, somebody inside it has bought something.
37%
of firms with 250 or more employees report using AI
US Census Bureau, Large Firms With at Least 20 Employees Biggest AI Users (Business Trends and Outlook Survey), by Adam Grundy, Cory Breaux and Dhanapati Khatiwoda, published . Read back .
Adoption among the smallest firms, 4 employees or fewer, is far lower. This measures adoption only, not cost or productivity, and it says nothing about pilots reaching production or about elapsed time.
02
So the decision in front of you is not whether to start. You started. The decision is what to do about the fact that it stopped at the edge of one department.
There are widely repeated figures for how often a pilot reaches production and how long that takes. We went looking for the studies behind them and could not find them, so they are not on this page and we are not going to paraphrase them into something vaguer. Every number we do use is listed with its publisher on where every number comes from.
What we will say without a number, because it is structural rather than measured, is that this size has a real advantage over a larger company. There are fewer people who can say no. There are fewer systems that have to be consulted. There is a shorter distance between the person with the problem and the person who can change how the work is done. A decision that would go through 4 review boards elsewhere goes through 2 conversations here.
That advantage has a shelf life. It is at its largest before you accumulate the governance layers that arrive with the next 1,000 people.
So the work is not to try AI. The work is taking the thing that proved itself in one team and making it the way the company runs, which is a different job with a different failure mode.
03
Joining up the systems you already own
A company your size does not need to be sold software. You have an ERP or a finance system, a CRM, a help desk, a payroll platform, a reporting tool, and a set of spreadsheets quietly holding the joins together. Somebody has already been through a selection process for most of them and does not want to do it again. So we are not here to tell you which tool to buy. If that is genuinely the question, that is different work and it is advice on which AI tools to buy. The value at this size sits in the joins. The order exists in one system, the customer exists in another, the invoice exists in a third, and a person moves information between them by hand because that join was never built. That person is the integration. They are usually excellent, usually senior enough to know better, and usually the only reason the chain works at all. 2 failure modes follow from getting this wrong.
- 01
Buying one more tool. New software that does not connect to what you own adds another place for information to sit and another manual join for a person to carry. That is a cost dressed as a purchase.
- 02
Replacing a core system to enable automation. A migration turns a short project into a long one, and it moves the ground under the measurement while it happens. We would rather work around your existing systems, and we will say so even when the existing system is not the one we would have chosen.
04
Where the buying decision sits
This is the part that most changes how the engagement actually runs. At a small business the person with the problem, the budget and the authority is one person. At your size those are 3 people, and the useful one is in the middle. The Head of Customer Operations knows exactly what is broken and has a budget line. The VP of Finance can see the number. The Chief Executive has heard the word AI too many times this year and is waiting for somebody to bring a result rather than a proposal.
- 01
We start in one function, not across the company. A single function head can sponsor, decide, and give us access to a real process without convening anybody. A company-wide programme needs a sponsor who does not exist yet.
- 02
The first result has to be something the sponsor can show their peers. Not a slide. A number their own colleagues already track and already believe. That is what turns one function's win into the second function's invitation, and it is how this spreads inside a business your size.
- 03
Sequence follows the org chart before it follows the process map. The most valuable chain to automate is often not the first one to attempt, because it crosses a boundary nobody has agreed to cross yet. We would rather finish something inside one department and earn the right to cross the boundary than start at the boundary and stall there.
None of this needs a steering committee, and we do not run one. It needs one function head who wants the problem gone.
05
Why the baseline is provable at this size
This is why the way Praxis is paid works better at your size than anywhere else, and it is not really an argument about data.
Praxis is paid a share of the measured gain, which means someone has to decide what the starting number was and whether it moved. At a small business that person is the owner, who is also the person receiving the invoice and the person who described the problem in the first place. It can be done, but every measurement runs through one interested party.
Your company has a controller, an FP&A analyst, or a finance director whose actual job is to own that number and defend it. That person did not ask us to be here and has no stake in the answer being flattering. They already produce a monthly reporting pack on a fixed cycle, from records that existed before we arrived and are signed off by people who do not work for Praxis.
So the measurement rides on a process that was running anyway. The starting point is not a figure we propose and you accept. It is a line your own finance function already publishes, chosen by them, fixed in writing before anything is built, and read the same way afterwards. If it does not move, there is nothing to share.
That independence is the thing that makes this arrangement honest rather than arguable, and it is the reason we will usually ask for the finance function to be in the room early even when the problem being solved is nowhere near finance. The mechanics are set out on how the profit share is measured, and what makes finance itself worth automating is on automate finance and accounting.
06
Work that crosses departments
A single workflow inside one team is worth something. A chain that runs from the first enquiry to the money landing is worth a great deal more, and at your size it is also the harder thing, because it crosses a line between two people who each own half of it.
Take a lead to cash chain. Marketing owns the enquiry. Sales owns the qualification. Operations owns delivery. Finance owns the invoice. Every one of those handoffs is a place where information gets retyped, delayed, or lost, and every one of them sits between two managers with different targets and different weeks.
At a small business that whole chain crosses one person's desk, so the handoff is not a negotiation. At a large company it crosses a formal process that already exists on paper. Your size is the one where the handoff is real, undocumented, and owned by nobody in particular.
Which means the technical part of joining two systems is usually the easy half. The other half is getting two function heads to agree what "qualified" means, or which system is the record when they disagree. We would rather surface that in week two than build around it and hand you an automation that quietly encodes an argument.
The chains worth looking at first in a company your size are usually the ones with the most handoffs and the least documentation: enquiry to invoice, order to delivery, hire to first day, and issue to resolution. Where each of those touches a specific function is covered on automate sales and your CRM and the other function pages, and the order we work in is on how it works.
07
What stays out of scope at this size
Short list, and it is about shape rather than principle.
We do not run 5 functions at once. We do not replace a core system to make automation possible. We do not build anything your own people cannot run after we leave, because a company your size should never be one supplier away from a broken process. And we do not take on work where the finance function will not agree a baseline, because without that the arrangement has no honest basis and both sides would end up arguing about a number instead of reading one.
Questions
What should you know before the first call?
- Why does an AI pilot stall inside one team instead of spreading?
- Usually because the pilot was built inside one team's tools and one team's process, so the handoffs on either side of it never changed. The work still arrives by hand and still leaves by hand, which means the pilot improved a step rather than a chain. Spreading it is an integration problem and an organisational one, not a question of whether the tool worked.
- Does a company that already has a CRM, an ERP and a help desk have to replace them?
- No, and replacing them is usually the wrong move. At this size the value sits in the joins between systems you already own rather than in new software, and a core system migration lengthens the work while blurring the measurement it depends on. Automation is built around the existing systems, and new tools are added only where something genuinely does not exist.
- Who in a mid-sized company should sponsor this work?
- A function head with a budget line and a broken process, rather than the chief executive. One function head can sponsor, decide and give access to a real process without convening anyone, which is why the work starts inside a single function and spreads once that function has a result its peers already believe.
- How is the gain measured in a company that has a finance team?
- It is measured from a line the finance function already publishes, chosen by them, agreed in writing before any building starts, and read the same way afterwards. Because the controller or finance director owns that number independently of the project, the before and after do not rely on anyone with a stake in the answer.
- Is it better to automate one workflow or a whole cross-function chain?
- Chains are worth far more, because most of the loss sits in the handoffs between departments rather than inside them. The harder part of a chain at this size is organisational: two function heads have to agree what a handoff means and which system holds the record. That disagreement is better surfaced early than encoded into an automation.
- What happens to the automation when the engagement ends?
- The company owns and runs it. It is built on systems already owned, handed over with written instructions, and the people using it are trained before handover. A company this size should never depend on one outside person to keep a process working.
Where this leads
This page is one part of the whole offer: the whole offer.
Start with a free consultation
Tell us which pilot worked, where it stopped, and which function head owns the problem now. We will tell you what can be joined up, what should not be replaced, and what we would build first. The call costs nothing and ends with a plain list either way.
No obligation · a scoping conversation first