By Shirish Kulkarni, Thought Leader in Data & Artificial Intelligence, 02 November 2026
Data is valuable only once it has been cleaned, connected and understood, just like gold ore. Most organizations hold far more data than they use, so the advantage goes to the teams that refine it into decisions. Analytics and data visualization are that refining stage.
Key takeaways
- Volume is not value. Most enterprise data go unused, so collecting more is no longer an advantage. Refining what you already hold is.
- Trust comes before insight. Two systems can both be correct and still disagree. The remedy is an agreed definition, not another report.
- An insight is finished only when a decision changes. Visualization shows, analytics explains, and someone then must change a parameter.
Raw data, like raw ore, isn’t wealth yet
Gold in the ground is not value until it is found, extracted, purified and shaped.
Business data follows the same path. ERP records, customer interactions, spreadsheets and sensor feeds pile up every day, but piling up is not using.
The gap is large. A 2020 IDC survey of 1,500 enterprise leaders, commissioned by Seagate, found that about 68% of the data available to enterprises went unused. Collecting more is no longer the edge. Refining what you already have is.
Bill and Alex, who appear throughout this article, are a composite scenario for illustration, not a client case. All figures in the scenario are illustrative.
On the plant floor: It is Monday, and Bill, a plant manager, has a purchase order due by noon. His ERP says 12,000 units are in stock. The warehouse system says 10,800. The plant already holds data from the ERP, the warehouse system, supplier portals, and a handful of spreadsheets. Data is not what Bill lacks. What he lacks is one number he can trust.
Unrefined data can be fool’s gold
Data nobody trusts looks like an asset and behaves like a liability. A peer-reviewed study of nearly 370,000 inventory records from one retailer’s 37 stores found about 65% were inaccurate.
In Gartner’s 2023 supply chain survey, one in three leaders named lack of trust in data as a top hurdle to adopting analytics.
AI raises the stakes. In an IDC study reported with Kinaxis, 62% of supply chain leaders said better data quality and integration is the factor most likely to speed up AI investment. Before data can be gold, someone must refine it.
One-point leaders often miss: when two trusted systems disagree, the cause is rarely “bad data”. It is usually a difference in definition or timing, such as whether quality-hold or in-transit stock counts as available. That makes it a governance and consulting question before it is a technology question.
On the plant floor: Alex, an operations data consultant, does not pick a number. He asks what “available” means to each system. In this scenario, the 1,200-unit gap breaks down into about 500 units on quality hold (counted by the ERP, excluded by the warehouse), 450 in transit (shipped, not yet received) and 250 reserved for open orders. Each figure was correct by its own definition. Once the teams agree on one definition, Bill places the order on a single number.
How data is refined: five stages
Refining is a sequence, and each stage removes a different kind of doubt. The stages below follow the plant’s story and apply to any function that depends on data.
Stage 1: Assay.
Start from the decisions, not the datasets
Miners assay ore before investing in extraction. The data equivalent is a short list of the decisions that matter most, such as what to order, where to hold stock or which customers to prioritize, with the metric each one depends on and a named owner.
Common mistake: Beginning with an inventory of every dataset, which produces a large platform and few answers.
On the plant floor: Bill’s list starts with one decision, how much to reorder each week, and one metric, available inventory.
Stage 2: Extract and connect.
Put the sources on a common footing
Data engineering brings the ERP, warehouse system, supplier portals, and spreadsheets into one governed environment. The detail that matters is alignment: one code per item, consistent units of measure, and comparisons taken at the same moment, so two systems are never compared at different times.
Common mistake: Copying data without aligning keys and timing, which only moves the disagreement to a new location.
On the plant floor: Alex maps both systems to one item code and compares them as of 06:00Pm each day. What remains of the 1,200-unit gap is genuine differences in definition.
Stage 3: Purify.
Govern the definition and test the quality
Each important metric gets one written definition: Its name, formula, what it includes and excludes, owner, refresh time and source of truth. The definition is then built once into a shared metric layer, so every report uses the same logic. Automated checks reconcile systems, monitor freshness, completeness and duplicates, and keep lineage so any number can be traced to its source.
Common mistake: Treating governance as a document that nobody enforces.
On the plant floor: Quality-hold, in-transit and reserved stock become named categories, and a daily check flags any variance above tolerance before a purchase order is raised.
Stage 4: Refine.
Move from “what happened” to “why”
Analytics separates the drivers behind a movement. A useful method is to compare actual against expected and split the difference into demand effects, supply effects such as lead time and reliability, and policy effects such as reorder parameters. Segmenting supplier, site and product shows where the cause sits, and agreeing what normal variation looks like keeps teams reacting to signals rather than noise.
Common mistake: Moving to prediction or AI before the diagnostic layer is trusted. Generative BI and AI assistants inherit whatever definitions lie beneath them.
On the plant floor: Inventory is up 18% across three warehouses while demand is flat. The decomposition shows that supplier lead times improved while reorder parameters stayed the same. Because required stock depends on demand during the lead time, a shorter lead time means less stock is needed. Good news produced excess inventory.
Stage 5: Shape.
Present for decisions, then close the loop
A good view leads with the exception, compares against expectation, shows the metric definition and last refresh, and answers one decision question. Refresh frequency should match the decision cycle: a weekly reorder does not need the same cadence as a daily allocation. Most importantly, every insight needs an owner and a trigger, and the change should be made in the system where it takes effect.
Common mistake: Celebrating the dashboard launch instead of tracking whether decisions changed.
On the plant floor: Bill and the planning team update the lead-time assumptions and safety stock for the affected products and set a trigger so that shifts in supplier lead times prompt a review. The excess stock stops building. Looking back, the plant moved through the stages in order: a decision to serve, connected sources, a governed definition, an explained cause, and a changed parameter. Skipping at an early stage makes every later one less reliable.
Three questions to ask about your data
- Can you trust it? Is there one agreed definition and a traceable source?
- Can people see it? Do leaders see the exception, not a table of thousands of rows?
- Does it change the decision? Who acts on insight, and how fast?
If any answer is no, there is a refining left to do.
How VCreaTek helps:
VCreaTek starts from where it should i.e. Data.
The end objective could be to deliver Insights through Analytics to help our stakeholders make decisions which create value. But we always insist on straightening out the inputs going into the decision engine. Once that is taken care of, Analytics & Data Visualization service focuses on the later stages of this journey, turning governed data into insight that people can see, trust and act on. It works alongside our data engineering team at the earlier stages.
The strategy works across multiple industries such as CPG, manufacturing and MedTech and are available in ours case studies.
Want to run an assessment of whether your data is working for you?
Please reach out to our analytics and data visualization team here: Analytics & Data Visualization services · Contact us
Sources and References
All figures are paraphrased and linked for reference. No source text is reproduced. Please review each source’s terms before reusing its material.
- Seagate and IDC, Rethink Data (2020). View source i A vendor-commissioned survey of 1,500 enterprise leaders. It is older but remains widely cited.
- Management Science, Inventory Record Inaccuracy: An Empirical Analysis. View research i A peer-reviewed study based on inventory records from one retailer’s 37 stores.
- Gartner, 7 Data Quality Focus Areas to Ensure Effective Analytics and AI (March 2024). View summary i Gartner guidance covering data-quality priorities for reliable analytics and artificial intelligence.
- IDC and Kinaxis, Making Supply Chain AI Accountable, reported by Just Style. View article i Findings from an IDC study commissioned by Kinaxis and reported by Just Style.
