Digital transformation fails quietly. Not with a cancelled programme, but with a portal nobody uses, a dashboard nobody trusts, and a personalisation strategy that met the same wall every other one meets: fragmented data that nobody is willing to stand behind.
Digital operating models
Deloitte’s 2026 research found that just 1% of IT leaders report no major operating model change underway. That is close to universal, and it means the differentiator is no longer whether you are changing but whether the change is designed or merely happening.
A digital operating model is not a technology decision. It is a decision about where work happens, who decides, and what the organisation stops doing. The most common mid-market error is layering digital capability on top of an unchanged structure: new tooling, same handoffs, same approval chains, same people doing coordination work that the tooling was bought to remove. The tools get blamed. The tools were not the problem.
If a digital programme does not remove work, it has added it. There is no neutral outcome.
Process digitisation and automation
Operational speed in mid-market businesses stalls in a specific, recognisable place: staff moving data between disconnected spreadsheets, PDF invoices and email threads. It is invisible in most reporting because it is nobody’s job title, and it is expensive because it consumes exactly the people you can least afford to have doing it.
The work is unglamorous and the returns are reliable. Intelligent document processing where the input is genuinely unstructured. Automated pipelines where the handoff is currently a person forwarding an attachment. Straight-through processing for the transactions that do not need a human at all, so the ones that do get proper attention.
A caution we apply consistently: automating a broken process makes it faster, not better. Where the underlying process is wrong, the automation should wait until the redesign is done — and saying so usually costs us scope in the short term.
Data architecture and platforms
Organisations lose an average of 5.87% of annual revenue to poor data quality — incomplete records, inconsistent formats, duplicates, silos. For a mid-market business that is a material number before AI enters the picture, and a compounding one afterwards, because a model trained on unreliable data produces unreliable output at scale and at speed.
The customer data platform market has matured past its marketing-technology origins into genuine data infrastructure, which is useful — but only if the foundations underneath it are sound. Our work here is establishing a single version of truth for the datasets that actually drive decisions, building lineage so you can trace and defend a number rather than merely produce it, and being honest about which parts of the estate are not yet fit to feed anything.
The data questions worth answering before you buy a platform
- Which datasets genuinely drive revenue or margin decisions, as opposed to being reported on
- Where does the same entity exist under different identifiers, and which one wins
- Can you trace a number end-to-end to its source, or only to the report that produced it
- Which systems are feeding decisions despite nobody owning their data quality
- What breaks if a regulator asks you to explain a specific automated decision
Customer and colleague digital experience
Experience work usually gets scoped for customers and stops there, which is an odd decision given that the colleague experience determines how much of the customer experience is deliverable. A support team fighting four systems to answer one question is a customer experience problem with an internal cause.
We treat both sides as the same problem: reduce the number of places someone has to go to complete a task, remove the steps that exist only because of a system boundary, and make the digital route genuinely faster than the workaround. That last test is the one that decides adoption, and it is the one most often skipped.
Personalisation sits here too, with a prerequisite attached. Ambitious personalisation on untrusted, fragmented data does not underdeliver — it actively erodes trust, because getting it visibly wrong is worse than not attempting it.
Signals this is the conversation you need
- Significant staff time goes to moving data between systems that should be connected
- Two reports answer the same question differently and both are defended
- A digital tool was deployed, adoption is low, and the workaround is still faster
- Personalisation or analytics ambitions keep meeting the same data quality wall
- Customers experience your internal system boundaries as delays
The fastest diagnostic here is usually the least technical one: follow a single transaction end to end and count the handoffs.