Data analytics consultancy or data science: which do you need?

You have data in several systems, reports that disagree, and a board asking what AI could do for the business. When you look for outside help, you find two kinds of firm: data analytics consultancies and data science consultancies. The names sound close, and their websites often use the same words. The work they do, the people they send and the results you can expect are different.

A data analytics consultancy helps you understand what has happened and why, using your existing data to build trusted reporting, dashboards and analysis. A data science consultancy builds models that predict, classify or optimise what happens next. Most organisations need analytics first, because a model is only as reliable as the numbers beneath it.

This guide is for the person who has to make the choice: a head of data, a finance director, or a CEO in a smaller organisation. It sets out what each kind of consultancy solves, what it delivers, what drives its cost and when you need both. It also explains how to brief either one, and how AI changes the decision. Shipshape Data does both kinds of work. We have tried to describe each fairly, including the cases where a smaller engagement is the right answer.

The difference in one paragraph

Data analytics looks at the past and the present. It answers questions such as what sold last quarter, which region is behind target and why margin fell in March. Data science looks forward. It builds statistical and machine learning models that estimate what will happen, sort records into categories or recommend an action. An analytics consultancy gives you numbers people agree on and the reporting to see them. A data science consultancy gives you a model that turns those numbers into a prediction or a decision. The two share tools and often share people, but they start from different questions.

The roles behind them follow the same split. Our guide to data engineering vs data science covers the people who build pipelines and the people who build models. For the difference between reporting and analysis itself, see business intelligence vs data analytics.

Analytics tells you what happened. Data science estimates what happens next. Both depend on the same data being right.

What each one solves

Problems analytics consultancies solve

• Reports that disagree. Finance and sales count revenue differently, and meetings start with an argument about whose number is right.
• Manual reporting. Analysts spend days each month exporting spreadsheets and rebuilding the same pack.
• Slow answers. A director asks why churn rose, and the answer takes two weeks.
• Dashboards nobody uses. A reporting tool was bought, but the reports do not match how the business runs.
• No agreed definitions. Terms such as active customer, gross margin or on-time delivery mean different things in different teams.

These are problems of trust, speed and consistency. The fix is usually a cleaner data model, agreed metric definitions and reporting built around real decisions. Poor source data sits behind most of them, and our guide to data quality explains how to measure it.

Problems data science consultancies solve

• Forecasting. Demand, cash flow, staffing or stock levels need an estimate with a confidence range, not a guess.
• Classification. Tickets, documents, transactions or leads need sorting into categories at a volume people cannot handle.
• Risk and anomaly detection. Fraud, equipment faults or unusual orders need flagging before they cost money.
• Optimisation. Prices, routes, schedules or marketing spend need setting against many constraints at once.
• Testing. You need to know whether a change caused an improvement or simply happened alongside it.

These are problems of prediction and decision. They need historical data with enough depth and quality to learn from, and a business process ready to act on the output. Our guide to predictive analytics explains how forecasting models use the history you already hold.

What each one delivers

Analytics deliverables

• A data model or semantic layer that holds one agreed definition of each business measure.
• Written metric definitions that finance, sales and operations have signed off.
• Dashboards and reports built around the decisions people make each week.
• Analysis of specific questions, with written findings and the method behind them.
• Training and documentation so your team can answer new questions without the consultancy.

If your reporting runs on Power BI, our guide to what a good Power BI consultancy engagement looks like sets out these deliverables in more detail.

Data science deliverables

• A framed problem with an agreed business measure of success.
• Prepared training datasets, with the features the model learns from documented.
• A trained model with test results on data it has not seen before.
• Deployment into a system or workflow where the output is used.
• Monitoring for drift, and a plan for when and how the model is retrained.
• Documentation that explains what the model does, what it was trained on and where it falls short.

A model that sits in a notebook is not a deliverable. Ask how the model will reach production and who will maintain it once it is there. Our guide to machine learning consulting covers the stages from problem framing to a model in production.

Team shape

An analytics engagement usually brings analytics engineers, BI developers and analysts who understand how the business measures itself. A data science engagement usually brings data scientists and machine learning engineers. Both need a data engineer, because both depend on pipelines that deliver clean data on time. Both also need someone who understands your sector well enough to tell a real pattern from an artefact of how the data was recorded.

What drives the cost of each

Neither type is cheaper by default. The cost depends on the state of your data and what you ask for. An analytics engagement on clean, well-organised data can be short. A data science engagement on the same data can also be modest if the question is narrow. The same work on scattered, inconsistent data costs more in both cases, because most of the effort goes into making the data usable.

• Data readiness. How many source systems there are, how clean they are, and whether a warehouse already exists.
• Definition work. How much agreement is needed on metrics before anything can be built.
• Scope. The number of reports, models or business areas covered.
• Production requirements. A model that runs live inside a workflow needs engineering, testing and monitoring that a one-off analysis does not.
• Ongoing support. Whether the consultancy stays to maintain, retrain and extend the work after go-live.
• Knowledge transfer. How much your own team needs to learn to run the work afterwards.

Data science work carries more uncertainty at the start. You do not know how well a model will perform until it has been tested on your data. A good consultancy scopes a first phase to answer that question before you commit to a full build. That first phase should produce something useful on its own, so the money is not wasted if the answer is to stop.

When you need both

Most organisations that need data science also need analytics, and usually first. A forecast is only as good as the history it learns from. If your sales figures disagree between systems, a demand model will learn the disagreement.

• You want forecasts, but your current reports are not trusted.
• You have a working model, but the business cannot see its results alongside the measures it already tracks.
• Your analysts answer the same predictive question by hand each month.
• Leadership has asked for AI, and nobody has checked whether the data can support it.

In these cases the sensible order is to fix the data foundation, agree definitions, build reporting people trust, and then add models on top. Data engineering sits under both. Our guide to data engineering consulting services explains what that layer includes.

A model trained on numbers nobody trusts gives you a confident answer nobody should trust.

Using one consultancy for both can reduce handovers, because the people who define the metrics are the people who build on them. This works when the consultancy has real depth in each discipline, rather than one strong practice and a thin second one. Ask to see examples of both kinds of work, and ask who will do each part.

How to brief either

A clear brief saves time and money. It also tells you quickly whether a consultancy understands your problem. Write it around the decision you want to improve, not the tool you want to buy.

• The decision. Which decision this work will change, and who makes it.
• The current state. Which systems hold the data, how it reaches reports today, and where it breaks.
• The measure of success. How you will know it worked, expressed as a business number.
• The users. Who will read the reports or act on the model output, and how often.
• The constraints. Budget, timeline, platforms already chosen, and regulatory limits such as UK GDPR.
• The ownership plan. Who will run and maintain the work after the engagement ends.

Questions to ask in reply

• What will you need to see in our data before you can scope this?
• What would make you advise against this work?
• How will the output reach production, and who maintains it?
• Who will do the work, and will they stay after go-live?

A consultancy that answers the second question honestly is more useful than one that agrees to everything. For a wider checklist, see our guide on how to shortlist a UK data consultancy.

Where AI changes the answer

Generative AI has blurred the line between the two kinds of consultancy. Many requests that once went to a data science team, such as classifying documents or summarising support tickets, can now use a pre-trained language model instead of a model trained from scratch. That lowers the modelling effort. It does not lower the data effort.

Language models used on company data need the same things analytics needs: agreed definitions, clean sources and clear rules about who can see what. A chat assistant that answers questions about revenue gives wrong answers if revenue has three definitions. In our experience, AI work turns into data work quickly. That is why we build the data foundation first.

What this means for your choice

• If you want AI to answer questions about your business data, start with analytics foundations: a governed data model and one definition per metric.
• If you want AI to read documents or route work, you may need less classic data science and more engineering, evaluation and governance.
• If you want bespoke predictive models on structured data, data science still applies, and so does the foundation under it.

Our guide on why AI projects fail covers the checks that predict trouble early. If you are unsure where you stand, the free AI readiness assessment scores strategy, people, data and change on the spot, with no email gate.

Where this leaves you

Choose a data analytics consultancy when the problem is trust, speed or consistency in the numbers you already report. Choose a data science consultancy when those numbers are trusted and you need to predict, classify or optimise. Choose one that can do both when you need both, and expect it to fix the foundation before it builds models.

Whichever you choose, judge the consultancy on how it scopes. The right partner names what the result depends on, starts with a first phase that pays for the second, and tells you when the work is not worth doing yet.

Our data consultancy service covers data strategy, engineering, analytics, data science and governance on one governed foundation, so your teams work from one version of the truth. We scope first, build on governed data and stick around past go-live. If you are deciding between analytics and data science, we will tell you which you need first, including when the answer is a smaller piece of work or not yet. Talk to us and you will get a personal reply, usually within one working day.

Frequently asked questions