What an AI operations firm does, and why companies are starting to need one
Most companies have bought AI tools. Far fewer have AI doing real work. Here is the gap between access and use, what an AI operations firm does about it, and what to look for in one.
Over the last two years, most companies went through the same cycle with AI. The tools arrived. Leadership heard what they could do. Licences were bought, a few training sessions were run, and someone was asked to look for use cases. Then the year moved on.
Today, in many of those companies, the licences are still being paid for. A handful of people use them to tidy up emails. Nobody is quite sure what data is safe to paste in. And when finance asks what the company is getting for the money, nobody can give a clear answer.
This is not a story about bad tools. The tools are good. It is a story about the gap between having AI and having AI do work.
Access is not use
Buying access to an AI tool is a purchasing decision. Getting value from it is an operations problem, and in most companies those two things are handled by different people.
The numbers on this are fairly consistent. Weekly active use of paid Microsoft 365 Copilot seats is commonly reported at around 20 to 30 percent of licences. A Recon Analytics survey of more than 150,000 US enterprise users found that about 35.8 percent of employees with Copilot access actively use it. In a typical company, most of the seats are paid for and sitting idle.
Closer to home, KPMG’s Global Tech Report 2026 found that only 24 percent of organisations in Kenya achieve a return on investment across their AI use cases, and 53 percent lack the expertise they need for their digital goals.
The reasons are usually ordinary. A general chat tool does not know how your approvals work or where your customer records live. People are not sure what they are allowed to do with it. Their managers have not changed what they expect from them. And nobody owns the job of connecting the tool to the actual work.
Why forward deployed engineers became a job title
Software companies ran into this problem first. They sold capable products, then watched customers struggle to put them to use. The answer some of them settled on was the forward deployed engineer: a technical person who embeds inside the customer’s business, learns how the work really moves, and builds the product into real workflows on site, instead of handing over a login and a manual.
As AI tools spread, demand for this kind of person has grown fast. Job postings for forward deployed engineers rose by more than 1,000 percent year on year through early 2026. That says something about where the bottleneck is. It is no longer the model. It is the people who can sit with a team, understand the process, and make the technology fit.
Most companies cannot hire that person. They are scarce, expensive and usually employed by the AI vendors themselves. And one engineer on their own does not fix the people side of the change.
What an AI operations firm is
An AI operations firm does that forward deployed work for companies that do not have it in house. It does not sell a particular tool. It comes into the business, finds where AI fits in the way work actually gets done, builds it, and gets the people using it. Then it stays to keep it working.
The difference from a typical IT vendor or software agency is where the work starts and where it ends. It starts with the operation, not the technology. And it does not end at launch. It ends when the people doing the work are using what was built, and leadership can see what it is doing.
How a typical engagement runs
Operations review. A short conversation about one process that takes too long or goes wrong too often. The aim is to understand the problem well enough to say whether AI fits there at all, and to write that down.
Assessment. Most of the thinking happens here, in four parts.
- Listen: talk to the people who do the work, not only the people who manage it.
- Map: follow how the work actually moves, step by step, including the workarounds nobody wrote down.
- Measure: put rough numbers on where time and money go, such as waiting, rework and missed follow ups.
- Rank: list the gaps by what they cost and how hard they are to fix, and agree which come first. Some fixes will not involve AI at all.
Build.
- Design the fix with the people who will use it.
- Build it inside the systems the company already has: the CRM, email, documents and accounts.
- Test it on real work, with real cases, before anyone depends on it.
- Launch it in stages, one team at a time.
Adoption.
- Map who will use what, and for which tasks.
- Set clear rules: what data can go where, what needs a human check, and who is accountable.
- Train by role, so a loans officer learns on loan files and a customer service agent learns on real enquiries.
- Measure use and results every month and report them to leadership.
Ongoing support. Watch performance, fix what breaks, add new uses as people find them, and keep the reporting going.
An example: loan applications at a mid sized SACCO
Take a SACCO with a busy loans desk.
Before: applications arrive by walk in, email and WhatsApp. An officer types the details into the core system by hand, then chases missing documents by phone. Files sit in a queue until someone has time to check them against the lending policy. Members call to ask where their application is, and each call pulls an officer away from the work. Management sees the backlog at the end of the month.
After: applications from every channel land in one place. AI reads the documents that come in, pulls out the details and flags what is missing, and the member gets a message asking for it straight away. Each complete file is checked against the policy and reaches the officer with a short summary and any issues marked. The officer still makes the decision. Members can check their status without calling. Management sees the queue every day.
None of this replaces the loans officer. It removes the typing, chasing and searching that kept the officer from the part of the job that needs judgement. And it only works because the process was mapped and cleaned up first. Automating a messy process only gives you a faster mess.
What a decision maker should look for
If you are considering an AI operations firm, a few questions separate the useful ones from the rest.
Do they start with your operation or with a product? A firm that arrives with a tool already chosen will find a way to use it.
Will they talk to the people who do the work? A plan built only from conversations with managers will miss how things really happen.
Do they build inside your existing systems? AI that lives in a separate tab does not get used.
What is their plan for adoption? Ask how training will work, who sets the rules and how use will be measured. If the answer is vague, the project will probably end at launch.
Will they tell you where AI does not fit? Some problems are better solved with a clearer process or a simple form. A firm that says so is one you can trust on the rest.
What happens after launch? Someone needs to watch, fix and improve what was built.
Where Eflapy Digital comes in
This is the work Eflapy Digital does. We are an AI operations firm based in Nairobi, working with organisations of 30 to 5,000 people. We find where AI fits, build it, and get people using it. We now offer this to companies in Kenya on site, and to companies abroad remotely.
If there is one process in your business that takes too long or goes wrong too often, book a free operations review. Bring that one process. In 30 minutes we will show you where AI fits, and you leave with a written summary whether you hire us or not.