Where a DevOps Services Company Fits in Operations

Author : James Smith | Published On : 10 Sep 2026

Our business improvement programme and our engineering programme ran separately for two years. Both delivered. Neither knew what the other was doing. About a third of the work overlapped.

The lesson took longer than it should have. A DevOps services company and a process partner often solve the same problem from different ends. The value sits where they meet.

We audited eleven slow processes to find out why each one was slow.

Six were procedural. Too many approvals, unclear ownership, work batched for no reason.

Five were technical. Two systems that did not talk, so somebody re-keyed data between them. A report assembled by hand each month, because no integration existed. A form that could not be submitted twice, so a correction meant a phone call.

Our business process services partner found all five and could fix none. That work needed engineering. It sat in a different budget with a different sponsor.

A process partner watches people work and sees the workaround.

An engineering team sees the system and rarely watches anybody use it.

Both views are partial. The workaround exists because of a system limitation. The limitation persists because nobody in engineering knew it cost forty minutes of manual work a day.

We now run discovery jointly. A process analyst and an engineer, in the same room, watching the same people. That fortnight produced better findings than either function managed alone in a year.

The split we settled on, written down.

Process design, standards, and the decision about what the work should be belong to the business.

Integration, system step automation, deployment, and anything needing code sits with engineering or the DevOps services company we use.

The grey area is workflow tooling. It can go either way. We decide per case and record it. Unowned workflow tooling is how you end up with three overlapping systems.

The structural problem was budgets.

Process improvements sat in one programme with one sponsor. The engineering work they depended on sat in another, requested case by case, and lost to whatever the engineering team had already committed to.

We now hold a single prioritised list covering both, with an integration allocation inside the improvement budget.

That change removed more delay than any process redesign. A DevOps services company can deliver quickly when the work is already funded and prioritised rather than negotiated each time.

Automating a bad process gives you a fast bad process.

We fix the procedural problems first, then automate what remains. The other order is the most common failure I have seen. Automation encodes whatever it was pointed at.

One example. We nearly automated an approval with no policy behind it. The process review removed the step entirely. That was cheaper and faster than automating it.

Once we looked, the pattern was clear.

Almost every improvement past the obvious ones needed two systems to exchange data reliably. That is engineering work. It takes time, and it was not in the process programme's budget.

We now fund an integration allocation inside the business improvement programme rather than requesting it separately each time. The DevOps services company delivers against it, prioritised by the process team.

That single funding change removed most of the friction between the two functions.

Proposals in this space use loose language and it causes real confusion in steering meetings.

Somebody will ask about the difference between chatbots and AI agents. A chatbot answers a person. An agent takes a goal and acts on its own, then reports afterwards.

The distinction matters commercially. A chatbot deflecting queries is a contained decision. An agent that changes records in your systems is a delegation. It needs limits enforced in code, an approval path for anything material, and a log you can reconstruct from.

We allow agents on a short list of reversible tasks. Everything else proposes and waits.

• Discovery is joint, with a process analyst and an engineer together.
• Integration work is funded inside the improvement programme.
• One prioritised list covers both procedural and technical fixes.
• Findings go to whichever function can act, rather than to whoever found them.
• Both partners attend the same monthly review.

The last one sounds trivial. Before, we relayed information between two vendors who had never met. Both worked from a partial picture we had assembled badly.

Where should process work sit?

With the business, informed by both partners. Business process services and engineering see different halves of the same problem, and neither view is complete alone.

Should one supplier do both?

Rarely. The disciplines are different. What matters is that they are in the same room and working from one prioritised list.

Who should lead?

The business, on what the process should be. A DevOps services company leading process design tends to produce technically elegant work aimed at the wrong problem.

How do we split the budget?

We do not, any more. One programme budget with an integration allocation inside it, prioritised jointly.

What is the fastest improvement available?

Usually removing a manual step between two systems. It is unglamorous, it is engineering work, and it is where most of our time savings came from.

Look at your process improvement backlog and count how many items need engineering.

In our case it was five of eleven, and every one of them was stalled because it sat in the wrong budget with the wrong sponsor.

A DevOps services company will not fix your processes. A process partner cannot fix your systems. Most of the value in either programme sits between them. Getting it out needed nothing more than one shared list and one review.