Data Analytics Consulting and Data Strategy

Data Analytics Consulting: From Numbers People Trust to Decisions That Hold

Data analytics consulting is the work of bringing operational data together, naming it and preparing it so that a figure in a report means the same thing it means in the department it came from. The subject rarely turns urgent because a new tool appeared. It turns urgent the moment two units walk into the same meeting with different numbers and the discussion circles the origin of the values instead of the decision that was on the agenda. What that calls for is less a set of fresh reports than a workable data strategy, a platform that joins the sources reliably, binding definitions with named ownership behind them — and people who keep both alive in day-to-day operations. That is the working ground of business intelligence consulting, and it is where a data consulting engagement starts.

Sort Out Your Data Picture See What the Work Covers

Two specialists working through a metric system on a laptop

Leading companies trust our network

The Metric Chain

What Data Analytics Consulting Delivers — and What It Does Not

Consultant tracing the path of a single metric across several source systems

Data analytics consulting — also called business intelligence consulting, BI consulting or data strategy consulting, depending on who is asking — works on a single question: how scattered operational data becomes a basis a whole company can agree on. Its subject is the distance a number travels before anyone reads it, not the choice of a reporting tool.

Every metric passes four stations. It is captured when someone posts a document, a machine reports a value or a form is submitted. It is joined when extracts from several systems meet. It is defined when it is settled what exactly counts as revenue, as an order or as an active customer. And it is used when someone decides something on the strength of it. The report is the last of those four stations — which makes it the place where an error becomes visible, and almost never the place where it was made.

That is why the work starts at the weakest station, not at the report. Contradictory numbers usually have one of three causes: a shared definition is missing, the join drops or duplicates records, or whatever happens to fit the mandatory field gets typed in at capture. All three produce the same symptom — and two of them sit outside any reporting tool.

And the work only ends where someone takes over the upkeep. A metric without named business ownership decays: systems change, fields take on new meanings, exceptions get built in. A reporting system is therefore not a result but a condition that has to be held.

What it does not do. It does not procure data that was never collected. It does not replace a business decision either: whether an order counts at the purchase order or at delivery is a question about the business, not about the data model. And an analytics consulting engagement is not an artificial intelligence engagement — where machine methods, generative applications or their oversight are the subject, AI consulting is the right place, although it assumes exactly the state of data that gets built here.

Decision Points

When Outside Help With Data and Reporting Is Worth the Effort

Not every data question needs help from outside. Where definitions are agreed, a platform is in place and a reporting system has already been built once in-house, doing it yourself is the quicker path. The six situations below tip that calculation the other way. What is missing in them is never knowledge of your own business; it is routine in untangling a data landscape that grew over a decade. A company rebuilds its reporting every few years. Our network does it several times a year.

1. Two Units Present Different Numbers

The same metric, two values — and the meeting circles the origin instead of the decision.
Almost never a tool problem: a binding definition is missing, and so is someone who owns it.

2. Reports Are Produced by Hand

Extracts get copied together into spreadsheets every month, and one or two people know how it works.
The effort is the smaller problem — the bigger one is that nobody can check the result.

3. A Reporting Tool Is in Place but Goes Unused

Licences are running, dashboards exist, and people still calculate in their own spreadsheet.
Usually what is missing is trust in the number or a link to the reader’s own job — neither can be bought with more licences.

4. The System Landscape Is Changing

A change of core system, a move off the data centre or a consolidation after an acquisition.
Anyone who looks at the data path only after the switch builds the reporting twice.

5. A Method Fails on the State of the Data

A forecast, an automation or a model cannot be calculated dependably.
The cause almost always sits in history, in completeness or in shifted meanings of single fields.

6. The Knowledge Hangs on Individuals

Whoever built the queries is retiring, moving on or tied up in day-to-day work.
Documentation, ownership and upkeep routes get settled before the knowledge leaves the building.

Do you recognise your own situation in one of these points? Twenty minutes are enough for a first sort: at which station of your metric chain the losses arise, what has to be settled first — and whether you actually need someone from outside for this step.

Talk Through Your Numbers

Areas of Work

From Data Strategy Consulting to Governance: The Areas of Work

The areas below get staffed singly or in combination, depending on which station of the metric chain is the sticking point. A data strategy consulting engagement usually opens with the first of them and pulls the others in as the picture clears. What follows describes what is asked for most often; narrower cuts are possible.

Data Strategy and Target Picture

Which decisions should rest on data in future, and what state of data does that presuppose? A data strategy answers this as a ranking: which question is worth answering first, which source has to be opened up for it, what gets left. The result is an order of work that still holds when the budget is cut.

Data Platform, Data Warehouse and Data Engineering

The substructure that brings the sources together: connecting the upstream systems, modelling, historisation, loading routes and the monitoring of those routes. Part of it is the sober question of whether a classic data warehouse, a data platform or a lean intermediate layer fits — and when what already exists is better extended than replaced. Vendor-neutral, with no licence to sell.

Business Intelligence, Reporting and Metric Systems

From the metric tree to the report someone actually opens: choosing the leading figures, deriving them and cutting them to their readers — a management board needs a different level of aggregation than a planning desk. The measure is usage; a report without readers is a cost item.

Data Governance, Data Quality and Master Data

The unspectacular part that decides whether a number still holds in two years: binding definitions, named business ownership per metric, rules for master data, measurable quality criteria and a route for reporting deviations. Data protection requirements and retention periods are placed here, not settled here — the binding assessment stays with your legal advisers.

Analytics Use Cases and Decision Support

Where an analysis goes beyond the rear-view mirror: sales and demand forecasts, anomaly detection, segmentations, simulations. What matters is not the method but the follow-up question — which decision would come out differently on a different result. Leave it open and you have produced a number, not a benefit.

Data Literacy and Operating Model

Who is allowed to analyse, who signs off, where does self-service end? An operating model settles responsibilities between the business units and a central data function, fixes which analyses are maintained, and puts the units in a position to answer their own questions without raising a ticket. Without this part every platform stays a bottleneck with a queue in front of it.

Which of these areas has to take hold first in your organisation is something twenty minutes can settle. Describe the state of your data; what comes back is a judgement, not a presentation.

Sort Out Your Data Picture

Engagement Formats

Four Formats in Which a Data Project Gets Supported

What decides the outcome is often not depth of expertise but form: how much capacity comes from outside, with what mandate and over what stretch of time. Our network covers four formats, and switching from one into the next happens more often than not. All four rest on the same frame: a named internal owner for the definitions, goals written down before anyone starts, and a date on which the work passes back.

Settle the Format

A Short Look From Outside

Second Opinion on the State of the Data

One experienced person examines a bounded question: does the planned architecture hold? Is the history sufficient for the forecast you want? Why do two reports contradict each other? Typically a few weeks, often part-time — sensible where the work is being done in-house and only a gap in experience has to be closed.

Building With Handover

Mixed Build Team

Two to five external specialists work alongside your own people on modelling, loading and reporting, with the professional lead staying internal. This is the format most often chosen while a platform is going up. What it buys you is the handover: because the work happens in your rooms, the knowledge is already where it has to stay.

Leadership With a Mandate

Ownership for a Period

One external person takes the lead of a data function or of a programme and holds the authority to decide: to bridge a vacancy, to steer a special situation, or wherever an arbiter is wanted who stands outside the running interests of the units. Always with a handover date agreed up front.

Holding Rather Than Building

Support in Regular Operations

Recurring support after the build: keeping the definitions current, developing the models further, standing by when upstream systems change. The format that is planned least often and needed most often — because a data landscape only stays true for as long as someone follows it.

Industry Practice

Data Work by Sector: Where the Chain Breaks First

The four stations of a metric are the same everywhere; the weak one is not. Three things decide which it is: where the data comes from at all — from machines, from documents or from customer contact —, how many upstream systems have to be joined, and how strictly a number has to be provable before anyone acts on it. Where values arise automatically, the problem usually sits in the join; where they are entered by hand, in the definition. That is why we staff along sector experience. Whoever knows the upstream systems, the customary metrics and the evidence duties of a sector spots earlier which station is losing the value — and which analysis will fail against a requirement that appears in no specification. The six sectors below are the ones our network covers most densely; further ones are reachable through the category pages.

Project Types

Data Projects That Are Commissioned Again and Again — and How They Are Judged

What gets commissioned in data projects falls for the most part into four project types. Each has a typical starting situation, a sequence that has proved itself and a figure that lets you agree in advance how the result will be judged.

A Metric System Everyone Signs Up To

Starting situation: several reports, several truths, a separate derivation per unit. The sequence that works: first name the twelve to twenty figures that are actually steered by, then write a definition per figure with named business ownership, and only after that build anything. The measure is how many figures can be taken from one source without a follow-up question.

Building or Replacing a Data Warehouse

Starting situation: a reporting layer that grew until nobody can see across it — or none at all. The sequence that works: settle sources and keys first, then a model for the leading business objects, then loading in waves rather than in one cut, with parallel running. The measure is loading time, the error rate of the routes and the time until a new source is usable.

Setting Up Data Governance and Master Data

Starting situation: duplicates, inconsistent spellings, fields whose meaning has shifted. The sequence that works: begin with the few master data objects that everyone needs, name a business owner per object, make quality criteria measurable and only then clean up — a clean-up without a rule for upkeep is back within two years. The measure is the duplicate rate and the completeness of mandatory fields.

Bringing a Forecast Into Regular Operations

Starting situation: a sales, demand or liquidity plan produced by hand whose accuracy nobody tracks. The sequence that works: measure the existing plan first so there is a benchmark, then check the history, then run a simple method against it. The measure is the deviation between forecast and actual — and whether the planning desk decides differently as a result.

From the First Metric to a Reporting System That Is Maintained: The Stages

How long a stage takes is decided by the number of source systems. Their order, by contrast, is fixed: first settle what the company is steered by, then check what the data will actually support, then write the definitions down, then build, then pass it back. None of the six gets left out, and shortening one is only defensible where dependable groundwork is already on the table.

Stage 1: steering questions and existing reports being recorded

1. Take Up the Steering Questions

What gets recorded is which decisions are taken regularly and which figures are actually used for them — not which reports exist.

Conversations with the business units, controlling and the people who build the analyses today make the workarounds visible.

The result is a list of the leading figures together with the definition questions still open.

Stage 2: source systems, history and data quality being checked

2. Check the State of the Data

For each figure it is checked which system it comes from, how far the history reaches and whether the meaning of the field stayed stable.

Completeness, duplicates and broken keys between the systems are measured rather than estimated.

The result is a sober statement of which analyses would fail on this basis today.

Stage 3: definitions and ownership being agreed and written down

3. Settle the Definitions

For each figure it is put in writing what it contains, what it excludes and from which event it counts.

Every definition carries a named business owner; without one the agreement is open again after the next system change.

The result is an agreed metric catalogue as the basis for the build.

Stage 4: model, loading routes and reporting layer being built

4. Build the Model and the Loading

Modelling of the leading business objects, connection of the sources, loading in waves.

Old and new run in parallel for a while; deviations are explained before anyone switches over.

The result is a reporting layer whose values can be traced back to the previous source.

Stage 5: business units being enabled and operations being settled

5. Enable People and Settle Operations

It is settled who may analyse for themselves, which reports are maintained centrally and where self-service ends.

Training runs on your own metrics, not on sample data; monitoring and a route for reporting deviations are set up.

The result is an operating model with named roles on both sides.

Stage 6: usage being measured and the work being handed over

6. Measure Usage and Hand Over

What gets measured is which reports are opened and how long a new question takes to reach an answer.

Analyses that go unused are switched off rather than maintained — they cost trust and maintenance time.

At the handover point the upkeep sits in the building.

Budget Frame

What Business Intelligence and a Data Warehouse Cost: Daily Rates and Budget Tiers

Outside support in a data analytics consulting engagement is invoiced through our network by the day worked. There is no lump sum for the project and no fee component that depends on the result. Six things move the number up or down.

Seniority. Whoever carries definitions through against several units, or owns an architecture, sits above a role that supplies.
State of the data. A maintained source with documented fields costs less onboarding than a landscape that grew over years without a description.
Tool stack. Widely used reporting and platform tools are staffed more broadly than niche environments or older in-house builds.
Industry. Sectors under evidence duties ask for experience with documentation and auditability.
Share of remote work. Data work can largely be done from a desk; a high share of on-site presence raises the rate.
Length of engagement. Longer mandates price lower than short assignments with a high share of onboarding.

The bands below come out of our own role pages, where each of them is published in the open. Analytical profiles sit at roughly €650 to €1,050 per day — Freelance Data Analyst and Freelance Forecasting Analyst. Build-side profiles for platform and loading move at €800 to €1,250 — Freelance Data Engineer or Freelance Platform Engineer. Statistical specialisation as well as definition and steering work sit at €750 to €1,300 — Freelance Anomaly Detection Specialist, Freelance Time-Series Forecasting Specialist, Freelance Knowledge Graph Engineer, Freelance Data Scientist, Freelance Business Controller, Freelance FP&A Consultant and Freelance Group Controller. At the upper end sits the connection of the core system at €950 to €1,500 — Freelance SAP BTP / Integration Consultant and Freelance SAP FI/CO Consultant. These are bands, not fixed prices; where a particular placement falls inside them is settled by the actual brief.

Budget in four tiers. The larger budget figure is rarely the daily rate but the number of days — and that can only be cut if the project has decision points. Four tiers have proved themselves, each ending with a sign-off of its own. The data assessment settles sources, history and gaps in the definitions and ends with a statement of which analyses would hold today — the cheapest tier with the highest leverage. The prototype takes a few metrics through to a usable analysis; its budget follows the number of source systems, not the number of wishes. The platform carries that across to the remaining units. Operations is the only tier that produces running cost, and the one most often forgotten. Sign all four off as one amount and the moment at which calling a halt would still be inexpensive disappears with it.

How this differs from a software vendor and from a large consultancy. A software vendor supplies the tool and its rollout — sensible once the product decision has been taken; which metric the company is steered by is not part of that brief. A large consultancy brings method for programmes spanning several legal entities; below that size a considerable share of the fee goes into managing the engagement itself. Specialists for a period are engaged by effort: on a fixed-scope contract the experience walks out with the supplier once the work is signed off, whereas a placement leaves it sitting where the metric is maintained. Licence fees, compute and cloud charges from third parties sit outside the daily rate and are not priced on this page.

Which profiles a project needs follows from the station that is the sticking point: settling definitions asks for closeness to the business, a loading route for depth of craft. The individual role profiles with their price frames sit under Data Engineering & Data Science. Steering and reporting profiles are found under Finance & Controlling, statistical specialisations under AI & Machine Learning, the connection of the core system under SAP & Enterprise Systems and the running of the platform under Cloud, Infrastructure & DevOps. Where the bigger picture beyond the data is at stake, digital transformation consulting is the right entry point; where models are to run on top of this foundation, AI consulting is.

12 September 2025

is when the EU Data Act applies. It governs access to data from connected products — and turns the question of where which data sits and who has to hand it over into a documented task. How it applies in a given case belongs in a legal review.

European Commission, Data Act

Article 10

of the AI Act requires documented data governance for high-risk applications: the origin, the suitability and the quality of the data used. Anyone planning to run machine methods needs that foundation beforehand — the classification in a given case remains a legal question.

European Commission, AI Act

Four Stations

are passed by every metric before it appears in a report. Only the last one is visible — that is where the error shows up, while it was usually made one or two stations earlier.

Business intelligence is the systematic preparation of operational data into metrics and reports a company is steered by. It answers mainly retrospective and status questions: how high revenue was, how the order book is developing, where a unit deviates from plan. Data analytics reaches beyond that and asks about relationships and predictions — why a development is occurring and what is likely to follow. In practice both rest on the same foundation: shared definitions, a dependable join of the sources and a maintained history. Anyone who wants to answer the second question without having ordered the first is calculating on soft ground.

A data strategy settles which decisions are to rest on data in future, what state of data that presupposes and in which order the work is done. It does not start from the data that happens to exist but from the decisions: the steering questions determine which figures are needed, which sources would have to supply them and where a gap sits. Out of that comes a ranking by contribution and effort. What matters is the reasoned omission: a strategy that takes in everything is a wish list.

A data warehouse is a reporting layer in which data from several upstream systems is brought together, unified and historised — structured, following an agreed model and built for recurring analyses. A data lake, by contrast, takes data in largely unchanged, unstructured data included, and defers the ordering to the moment of use. The difference is less technical than organisational: the warehouse demands the definition work up front, the lake allows it to be postponed — with the risk that it never happens at all.

Data governance is the whole of the rules, responsibilities and routines with which a company husbands its data: binding definitions per metric, named business ownership, quality criteria, access rules and a route for reporting deviations. It is needed because a data landscape decays without upkeep. Data protection requirements and retention periods can be mapped inside this frame; their binding assessment is a case-by-case legal question and belongs with your legal advisers, not in a data project.

Three groups move alongside one another in the market for business intelligence consulting: software vendors and system integrators, who introduce tools; large consultancies, who steer programmes across several legal entities; and specialists who work inside existing structures. Which group fits hangs on the cut of the work. Where the tool is already settled, the vendor is the shortest route. Where definitions and responsibilities are the subject, what BI consulting has to supply are people who work inside your own units and leave their knowledge there.

As a rule a data consulting engagement is carried by a small pairing rather than a single role. The strategic side — which decisions should be taken on the basis of data — is carried by profiles close to the business, for example out of controlling. The architectural side — model, source connection, operations — is carried by data engineers and platform specialists. Both need a named internal owner who decides between the units. The task profiles together with their daily rate bands are held under Data Engineering & Data Science and Finance & Controlling.

Work through our network is invoiced per day. Analytical profiles sit at roughly €650 to €1,050 per day, build-side profiles for platform and loading at €800 to €1,250, steering and definition work at €750 to €1,300, and the connection of the core system at €950 to €1,500. The total, though, is rarely decided by the rate but by the number of days — and that hangs on the number of source systems and on the state of the definitions. All figures are bands, not fixed prices.

A serious figure can only be named after the data assessment, because three things determine the effort: the number of upstream systems to be connected, the condition of their keys and histories, and how many definitions are still open. Two maintained sources with agreed metrics are a different project from eight systems that grew over years without a description. What makes it plannable is the tiering: data assessment, prototype, platform, operations — each tier with a sign-off of its own. The daily rate bands named above apply. Licence fees, compute and cloud charges from third parties come on top of them and are deliberately left unquantified here.

You can rely on us

Excellent. We are not the only ones who think so.

consultingheads has received several awards from leading trade magazines and independent third parties.

consultingheads-award-top-company-2025-kununu

consultingheads-kununu-top-company-2023 (1)

Next Step

Let’s sort out the state of your data.

Twenty minutes to sort out your metric chain

One named station at which the work begins

Free of charge, with no tool recommendation and no sales pressure

Twenty minutes in which we sort out the state of your data and name the station where the losses arise — and whether our network is the right place to look for the people.