Ragioneer Decision Intelligence, built on your Data Intelligence foundation.

Spot where results drift. Test the fix first.

Decision Intelligence watches your results, finds where they move away from the plan and why, and lets you try a change before any rule is rewritten.

Explore a pattern in your business See it in your industry
Week 1 · Plan set

The group allows 21 days to re-let an empty unit, from keys returned to the new tenant moving in.

Example decision for an illustrative property group with 14 buildings.

Learning from patterns is older than AI.

In 1924 Walter Shewhart sketched the first control chart at Bell Labs. The idea holds: learn what normal looks like, and look into only the changes that matter.

A business now has far more to watch than a production line. Transactions sit alongside project records, customer conversations, market data and years of operating history. Decision Intelligence brings statistics, machine learning and language models to all of it, with one purpose: to learn which of your assumptions still describe the business, where circumstances have changed, and what deserves a closer look next.

Every operating rule carries an assumption about the world. A contractor allows a standard period for closing out a project. A retailer applies one shipping threshold in every region. A property manager allows a fixed number of days to re-let a unit. Each rule was reasonable when it was written. Its continued use deserves evidence.

Consider one rule

21 days
to re-let an empty unit
  1. Written2023. The contractor could start work the day the keys came back.
  2. ChangedThis spring. A new contractor, who starts once the unit is formally handed over.
  3. Today34 days on average in the two buildings that use the new contractor. 21 everywhere else.
Right when it was written. Worth checking now.

From history to a change you can test.

Analysis runs on a schedule, as results arrive or when a team asks. Every step keeps its evidence, so each finding traces back to its records.

  1. 01Record what was expected.

    The forecast, estimate or rule that applied at the time, kept apart from anything changed later. You can always see what the business believed, and when.

  2. 02Compare like with like.

    Statistical models, machine learning and time-series analysis look for trends, similarities and unusual outcomes, allowing for season, location, project type and market conditions.

  3. 03Say what the evidence supports.

    Language models read the related documents and put the finding in plain words. Every claim points back to records. A possible cause is labelled as something to test, not as a conclusion.

  4. 04Test the change.

    Forecasts are checked on periods they have never seen. Where practical, a change is tried on a small scale and compared with the current approach before anyone commits to it.

  5. 05Return what you learned.

    After review, the estimate, forecast or rule is updated, with its owner, its evidence and the date it took effect. The next results show whether it helped.

A dashboard shows the number. This shows what to do next.

The same re-let figure, as a dashboard reports it and as Decision Intelligence reports it.

What the dashboard shows

Average time to re-let34 daysAcross the portfolio, this month
Units empty today9
Rent collected97%

The number is right. It is also the whole report. Someone now has to work out whether 34 is a problem, where it is coming from and what, if anything, to do about it.

What Decision Intelligence reports

The average rose from 21 to 34 days in week 6, and only in two buildings. The other twelve are still re-letting within plan.

Both buildings changed refurbishment contractor that week. The new contractor starts after the keys come back, not at notice stage. That is the most likely reason, and it is marked as something to test rather than a conclusion.

The suggested test: book the contractor at notice stage in those two buildings for four weeks, and compare the result with the other twelve.

Four weeks later, both buildings are back to 22 days. The property head approves the change for every building, with the evidence attached.

A dashboard tells you that something moved. Decision Intelligence tells you where, since when, the most likely reason and what is worth trying, then measures whether the change worked.

Good analysis also tells you what is missing.

22%

of UK construction firms never measure productivity, the highest share of any region surveyed. Use of industry benchmarks ranged from 5% to 16% across regions.

RICS Construction Productivity Report 2026, drawing on nearly 3,000 professionals. A survey of the industry, not a Ragioneer deployment.

Decision Intelligence names the gaps and the evidence needed to close them. An analysis is allowed to conclude that the data does not support a reliable finding or forecast. That answer is worth more than a confident guess, and it is often the most useful first result.

  • If lost tenders have no recorded reason, bid analysis has very little to learn from.
  • If project records stop at handover, the cost of closing the account never enters the next estimate.
  • If a whole season is missing from the history, a demand pattern is not ready to support a forecast.

Every finding comes with its evidence.

A finding is a proposal, not an instruction. Nothing in the business changes until the person who owns the rule has reviewed it.

Each finding shows the period and the population it looked at, the source records, how strong the evidence is and where it is thin. What was observed is kept separate from what might explain it.

Reliability is measured for the job in hand: forecast error, missed events, false alerts and performance on data the model has never seen. It is measured again when operating conditions change.

Nothing changes on its own. A named person approves every update to a rule, and the record shows what was proposed, what was decided and what happened next.

The finding, as the property head received it. An example.

Re-let time rose from 21 to 34 days in two buildings.

What it looked at. Weeks 1 to 7, and the 41 units re-let across all 14 buildings in that time. The records come from the lease system, maintenance tickets and contractor invoices.

How strong it is. Strong. The rise shows in all eleven re-lets in those two buildings and in none of the other thirty.

Where it is thin. There is only one contractor change to compare against, and winter demand is not yet in the data.

Still to test. That booking the contractor at notice stage brings the time back down.

Property head, week 7: test approved in two buildings for four weeks. The proposal, the decision, the result and the plan change that followed are all on record.

The same method, in your industry.

Three places we apply it. The records change from one industry to the next. The discipline doesn’t.

Construction and infrastructure

Win work with a clearer view of how it will finish.

Two projects start with the same estimated margin. One reaches handover with a short defects list. The other carries unresolved variations and a final account that takes months to settle. The next estimate needs to learn from both.

We connect bid assumptions, win and loss outcomes, contract terms, programme changes, supplier performance, costs, defects, variations and settlement dates, as they were known at the time.

So you can ask: Which opportunities resemble work we delivered well? Where do our estimates repeatedly miss?

Measured by estimate accuracy, margin at final settlement and the time from handover to closeout.

Winning and delivering are judged separately. Delivery records only cover the work you won.

Discuss construction analytics

Commerce and customer growth

Learn which growth is worth repeating.

An offer lifts orders, so the business expands it. Weeks later, returns, shipping costs and service contacts tell a more complicated story. The offer works for one group of customers and erodes margin in another.

We connect campaign and referral sources with product mix, discounts, fulfilment costs, returns, support contacts and repeat purchases.

So you can ask: Which customers stay profitable after the cost of serving them? Which offer should we repeat, and for whom?

Measured by incremental contribution, repeat purchasing and customer outcomes after the change.

A pattern in the history is a reason to design a test. The test shows whether the change helps.

Discuss commerce analytics

Healthcare and research operations

Plan capacity around the work a protocol requires.

A programme plans capacity from the number of available appointments. Some visits take a short assessment. Others need equipment, several staff and a lab result. Then the protocol is amended, and the scheduling assumptions stay the same.

We connect scheduling, staffing, visit durations, equipment, lab turnaround and completion records, with the protocol version in force at the time.

So you can ask: Where do delays concentrate, once visit type and site are allowed for? Do our capacity estimates still describe this work?

Measured by scheduling accuracy, completed visits, waiting time, staff overtime and resource use.

Gaps in site coverage and changes in the population stay visible when a finding is read.

Discuss healthcare analytics

Each approved finding strengthens the foundation.

Data Intelligence keeps the connected record of your business. Decision Intelligence learns from it and returns what it finds.

The foundation

Your records, with their meaning, rules and permissions, maintained by Data Intelligence.

Analysis

Runs in the background, at the pace of the work: daily, monthly, or when a project completes.

A finding, reviewed

Put in plain words with its evidence, and taken to the person who owns the rule.

An updated rule, dated

The definition, forecast input or rule is changed, with the old version kept alongside.

People and AI applications work from the improved rule. The next results show whether it helped.

Common questions.

What teams usually ask before they bring us their first assumption.

Still have a question?Ask us about your systems, your data and where to start.Talk to us
Does this need to run in real time?

No. Some questions benefit from frequent monitoring. Others need a month, a season or a completed project before there is anything to learn. The cadence follows the pace of the work and the amount of evidence needed.

How much history do we need?

There is no row count that makes a prediction reliable. What matters is relevant examples, comparable outcomes, consistent definitions and coverage of the conditions you want to predict. We assess the evidence for each use case before we set expectations.

Does it change our rules automatically?

No. A finding is a proposal. The person who owns the rule reviews the evidence, decides whether to test it and approves any change. Every step is recorded, with the old version of the rule kept alongside the new one.

Do we need Data Intelligence first?

Yes. Decision Intelligence works on the connected, defined and checked record that Data Intelligence maintains. Without it, the analysis would be comparing figures that don’t mean the same thing.

What if our data isn’t good enough?

The analysis says so, names the gap and says what would close it. That is a legitimate result, and often the most useful first finding.

Bring one assumption you’d like to check.

We’ll look at what your records can already say about it, what is missing, and what a first test would look like. It starts from your Data Intelligence foundation.

Explore a pattern in your business Explore Data Intelligence

The property group, its buildings, figures and people on this page are illustrative, not a client.