Contents
The Voice of Volge is Volge’s thought-leadership series, unsigned and institutional, every figure sourced, quantified and year-stamped. Volge is a Nintex delivery partner; interest disclosed.
volge.com/voice
About Nintex
Nintex is a process-and-automation software company. What began as SharePoint-era workflow is now a single platform that maps, mines, automates and, since 2025, orchestrates AI agents against work, built out over two decades and a handful of pointed acquisitions.
The recent chapters matter most to a buyer today: Agentic Business Orchestration (September 2025), a rebuilt global partner programme, locally-hosted AI in Automation K2 5.9.1 (April 2026), and Agent Designer + Nintex Orchestration entering beta.
This report covers Process Intelligence, the Understand layer, where Nintex maps and mines how work actually runs before anything is automated.
| Understand Focus of this report | Process Manager · Process Intelligence |
| Build & run | Automation Cloud · Automation K2 (on-prem) · RPA · DocGen · Forms · eSign |
| Orchestrate | Agentic Business Orchestration · Agent Designer (beta) · Nintex Orchestration (beta) |
Source: Nintex customer stories and G2/TrustRadius, 2022–26. Customer outcomes as reported by Nintex.
This is not a young product. It is a two-decade platform that has just added governed AI and a locally-hosted engine, mature where it must be, current where it counts, and now built to orchestrate agents at scale.
You cannot automate what you cannot see.
Before the platform, before the agent, before the budget: an honest account of how the work actually runs. Seeing it is the first deliverable, and everything else depends on it.
Every process exists twice.
The process in the manual and the process that actually happens are rarely the same document. The distance between them is the subject of this book.
Every large organisation runs two versions of every process, and only one of them is written down. There is the process as designed, the flowchart in the quality manual, the swimlane in the onboarding deck, the story the head of operations tells the board. And there is the process as performed, the reworked approvals, the shadow spreadsheet, the exception that quietly became the rule. Automation and AI both act on the first. Value hides in the second.
This matters now because the tools have outrun the understanding. An organisation can buy an AI agent, a robotic process automation licence, and a workflow engine in an afternoon. What it cannot buy in an afternoon is an honest account of how its own work actually moves, which step waits three days for a signature, which invoice route the finance team abandoned in 2021, which "standard" procedure has eleven live variants across four regions.
Point automation at the designed map and you automate a fiction. The bot handles the happy path and falls over on the exceptions that make up most of the real volume. Point an AI agent at the same fiction and the failure is quieter and more expensive: the agent acts confidently on a process that no longer exists, and no one notices until the numbers drift.
The intelligent-process-automation market is priced as if this problem were already solved. Market.us estimates the category at roughly 16.81 billion US dollars in 2024, rising to 61.23 billion by 2034 at a compound annual growth rate of 13.8 per cent (Market.us estimate, 2024). Firms are spending. The question this book asks is narrower and more useful: are they spending against reality, or against the map?
Nintex answers that question with two capabilities that fit together. Nintex Process Manager holds the process as designed, the map. Nintex Process Intelligence reconstructs the process as performed, from the event logs your systems already keep, the mine. The gap between the two is not an embarrassment to be hidden. It is a worklist. Every deviation, every undocumented variant, every bottleneck is a costed opportunity to automate, to simplify, or to leave well alone.
Volge licenses, configures, and operates both on our clients' behalf. We wrote this book to make the case for buying that capability, and we have marked our commercial interest wherever it appears. Three things follow, and we will not pretend otherwise: you need the map, you need the mine, and you need the discipline to work the difference between them.
Read on for the problem, then the two instruments, then the bridge Nintex shipped in January 2026 that turns the mine back into a map a business person can read.
The cost of the gap
Market context
2 · IPA market to 2034
3 · Where automation breaks
Automation applied to a process nobody understands scales the error, not the work
Automation is a multiplier, and a multiplier is indifferent to the sign of the number it multiplies. Applied to a clean, documented, stable process, it removes cost and delay. Applied to one that is none of those things, it industrialises the confusion. The technology is neutral; the understanding is not.
Consider the ordinary shape of a failed automation programme. A team picks a process, say, supplier onboarding, because it is painful and visible. They document it by asking the people who run it what they do, and those people describe the map: the official steps, in the official order. The team automates the map. Then live volume arrives, carrying the forty per cent of cases the map never mentioned, and the automation either breaks or, worse, processes them wrongly and silently.
The pattern repeats because the diagnosis is skipped. Interviews capture what people believe they do, which is a poor guide to what they actually do. Memory smooths over the reruns, the workarounds, and the informal escalations that keep the business moving. The designed process is a sincere account. It is also, routinely, wrong in the specific places that cost the most money.
AI raises the stakes. A rule-based bot fails loudly, it stops, throws an error, waits for a human. An AI agent is built to keep going. Give it an incomplete picture of a process and it will improvise a plausible completion, acting on steps that were retired, approvers who have left, and thresholds that changed two budgets ago. The failure mode is not a crash. It is confident, well-formatted, and wrong.
This is why process understanding has become the entry point for AI rather than an afterthought to it. Niranjan Vijayaragavan, Nintex's chief product and technology officer, put it plainly in the Forbes Technology Council on 3 February 2026: process intelligence is "the AI entry point no one is talking about." An agent is only as good as its model of the work. Give it the map and it inherits the map's blind spots; give it the mine and it inherits reality.
None of this argues against automation. It argues against automating in the dark. The sequence that works is dull and reliable: see the process as designed, measure it as performed, resolve the difference, and only then decide what to automate, what to simplify, and what to leave alone.
| Dimension | Process as designed: the map | Process as performed: the mine |
|---|---|---|
| Source | Interviews, manuals, workshops, policy | Event logs from ERP, CRM, and line systems |
| Shows | How work is meant to run | How work actually ran, case by case |
| Bias | Optimistic, tidy, singular | Complete, messy, many-variant |
| Nintex instrument | Process Manager | Process Intelligence |
| Fails when | You automate it as if it were true | You never look at it at all |
The money is already committed; the understanding is the part that lags
Spending on intelligent process automation is set to nearly quadruple in a decade, which means the demand for tools has already been decided and the demand for process truth is the constraint that remains. Market.us estimates the category at 16.81 billion US dollars in 2024, growing to 61.23 billion by 2034: a compound annual growth rate of 13.8 per cent (Market.us estimate, 2024). Treat the exact figure as a directional estimate. Treat the direction as settled.
Interpolated years assume the stated 13.8% CAGR and are illustrative only.
Rising spend on automation does not, by itself, buy results. It buys capacity. Whether that capacity lands on real bottlenecks or imaginary ones depends entirely on the quality of the process picture feeding it. A firm can be a heavy, sophisticated buyer of automation and still be automating the wrong forty per cent, because the picture it is working from is the tidy map rather than the messy mine.
The corrective is cheap relative to the spend it governs. Process mining and process mapping are small line items next to an enterprise automation programme, and they determine whether that programme's budget is aimed at the steps that actually cost time and money. Skipping the diagnosis to save on the diagnosis is the expensive economy.
There is a second reason the understanding lags. Mapping felt, for years, like an administrative chore: a compliance artefact produced once and left to rot. That framing is now obsolete. A living process picture is the substrate every automation and every agent reads from. It has moved from filing cabinet to operating dependency.
Nintex sits on the understanding side of this market rather than only the execution side, which is the argument of the next three parts. First the map. Then the mine. Then the bridge between them.
Three places automation breaks, all of them invisible on the map
Automation programmes fail in a small number of predictable ways, and each is invisible to anyone working from the designed process alone. Name the three and you have a checklist for why an initiative underdelivered; see them in your own data and you have a worklist.
A process described as one path is usually many. Regions adapt, teams improvise, and a single "standard" purchase-to-pay quietly grows a dozen live variants no manual records. Automate the documented path and every undocumented variant either breaks the bot or bypasses it: which is how a shiny programme ends up covering a minority of real volume.
Delay concentrates. In most processes a few steps hold most of the waiting, an approval that queues for days, a handoff between two systems no one owns, a rework loop triggered by bad upstream data. These do not show up on a swimlane diagram, which has no axis for time. They show up in event logs.
The costly deviations are the ones nobody flags. A control skipped "just this once" that became routine. An approval threshold everyone treats as advisory. A step performed out of order because the official order was inconvenient. Each is defensible alone and dangerous in aggregate, and each is invisible until you compare the mine against the map.
| Failure mode | What the board sees | What actually happened | Surfaced by |
|---|---|---|---|
| Undocumented variants | Automation covers less than promised | Real work runs on paths the map omits | Process Intelligence |
| Hidden bottlenecks | Cycle time refuses to fall | A few steps hold most of the wait | Process Intelligence |
| Silent deviations | An audit finding, eventually | Controls quietly skipped at scale | Map vs mine comparison |
If the documented path carries 60 per cent of volume and your automation covers only the documented path, your ceiling is 60 per cent before you have written a line of configuration. The other 40 per cent is not an edge case. It is the case you did not measure.
AI agents make the invisible-process problem worse before they make anything better
An AI agent inherits every blind spot in the process picture you hand it, and then acts on those blind spots with confidence a rule-based system never had. This is the central shift of 2026. The organisations rushing agents into operations are, in many cases, pointing them at the designed map, the one document guaranteed to be incomplete, and calling the result transformation.
The distinction that matters is between failing loudly and failing quietly. Traditional automation is brittle: it stops when reality departs from its script, and a human notices. That brittleness is annoying and also protective. It is a smoke alarm. An agent, by design, does not stop. It reasons its way past the gap and produces an answer, which may be right, may be wrong, and will in either case look equally polished.
So the value of a complete process picture rises exactly as agents arrive. The agent needs to know the variants, the exceptions, the controls, and the true sequence of events. That knowledge does not live in the manual. It lives in the event logs, waiting to be mined, which is the argument for Process Intelligence as the thing you buy before, not after, the agent.
This reframes the mining category. Finding savings in existing processes is part of it, but the larger role is as a readiness layer for anything you plan to hand to a machine: the difference between deploying an agent against your operations as they are and deploying it against a flattering description of how they used to be.
Nintex's own framing places process intelligence at the front of the AI question rather than the back. We agree, with one caveat we will keep repeating: mining tells you what happened, not why, and the "why" still needs people. The instrument sharpens the judgement. It does not replace it.
Process mining reconstructs the sequence and timing of work from data. It does not explain intent. A step that looks like a deviation may be a sensible local fix; a bottleneck may be a deliberate quality gate. The instrument tells you where to look. Deciding what it means is still a human job, and Volge treats it as one.
Diagnosis first, then automation: the order that saves the budget
The sequence below is the one Volge runs on every engagement, and it is deliberately unglamorous. Its whole value is that it refuses to automate before it understands. Each step narrows the field of what is worth automating, so that the expensive execution budget lands on the few steps that actually move cycle time and cost.
Reverse any two steps and the economics collapse. Automate before you mine and you scale the current mess. Mine before you map and you have reality with nothing to compare it against. Act before you compare and you are back to guessing, which is where most programmes started. The order is the product.
Automation multiplies whatever it is pointed at. Most firms point it at the map. The map is incomplete in the exact places that cost money, variants, bottlenecks, deviations, and agents make that worse. The fix is to see the process as performed before you change it. Part II supplies the map. Part III supplies the mine.
The AI features
Governance & ownership
5 · The three verbs
6 · AI feature timeline
7 · Before / after
Nintex Process Manager turns the designed process into a document the business actually maintains
Nintex Process Manager exists to make the process as designed a living asset rather than a dead diagram, written in plain business language, owned by the people who run the work, and kept current because keeping it current is easy. Formerly Promapp, rebranded to Process Manager in 2022, it carries a deliberately modest tagline: identify, map, and model your processes. The modesty is the point.
Traditional process mapping fails for a human reason, not a technical one. The maps are drawn by specialists in a notation only specialists read, stored somewhere the doers never look, and updated on a cycle measured in years. So they drift from reality the moment they are signed off, and everyone reverts to tribal knowledge. The diagram becomes a compliance artefact, produced, filed, ignored.
Process Manager attacks that failure by moving ownership to the front line. Maps are written in readable steps, not arcane symbols, so the person who runs the process can read it, challenge it, and own it. Feedback and change requests sit inside the process itself. Ownership is assigned, so every process has a named human accountable for keeping it honest. The result is a map that stays close to the truth because the people closest to the truth maintain it.
That readability has a second payoff. A map anyone can read is a map anyone can improve, which turns process documentation from a project into a habit. New joiners learn the real procedure instead of shadowing a colleague, and auditors read the same document the operators use. Change becomes visible and versioned rather than whispered.
"Model", the third verb in the tagline, matters as much as the first two. A map records the steps; a model captures the logic, the decisions, the branches, the roles, the handoffs, in a form precise enough to reason about and to hand downstream. That precision is what makes the later moves possible: comparing the model against mined reality, and generating the artefacts that drive automation.
| Attribute | Detail |
|---|---|
| What it is | Business process mapping and modelling: the process as designed |
| Former name | Promapp (rebranded to Nintex Process Manager, 2022) |
| Tagline | Identify, map, and model your processes |
| Owned by | Front-line process owners, in business language |
| Core value | A living, current, readable map rather than a filed diagram |
Three verbs, three jobs, one living document
The tagline hides three distinct jobs, and it is worth separating them because organisations tend to be good at one and poor at the others. Identifying, mapping, and modelling are not the same task, and a process programme that skips any of them leaves a predictable gap.
Most organisations cannot name their own processes. Work is described by department, by system, or by project, and the process, the end-to-end thing the customer experiences, falls between those categories. Identifying means building the register: the list of processes that exist, who owns each, and how they connect. Dull, foundational, and usually missing.
Mapping records the designed sequence in readable steps: what happens, in what order, by whom, with what inputs and outputs. Done in business language, the map becomes usable by the front line rather than filed by the quality team. This is the artefact new joiners learn from and auditors read.
Modelling adds the decisions and branches, the roles, the conditions, the handoffs, in a structured form. A model can be reasoned about, compared against mined reality, and handed to downstream tools. It is the difference between a picture of a process and a specification of one.
| Verb | Produces | Primary reader | Failure if skipped |
|---|---|---|---|
| Identify | A process register with owners | Operations leadership | You improve processes you cannot name |
| Map | A readable procedure | Front-line staff, auditors | Knowledge stays tribal |
| Model | Structured process logic | Automation and analysis | Nothing downstream to build on |
Every process in Process Manager has a named owner. This single discipline does more for accuracy than any feature. A document with an accountable human attached stays honest; an ownerless diagram rots. Volge configures ownership before it configures anything else.
Between March 2025 and January 2026, Nintex removed most of the manual labour from mapping
Three AI features shipped inside eleven months have changed the economics of getting a map on the page, because each attacks the single most expensive part of process documentation: the human effort of writing it down. The old objection to mapping was always cost: it took specialists weeks. That objection is now much weaker. Exhibit 6 dates the releases; the paragraphs below explain what each one does.
The AI Process Generator entered beta in March 2025. It drafts a process map from a description, turning the terrifying blank page into an editable first draft. The draft is not the deliverable, a human owner still refines it, but starting from eighty per cent complete rather than zero changes who can participate and how fast.
Process Capture reached general availability in September 2025. It records the actual activity of a task as it is performed and auto-documents it, so the map is drawn from what someone did rather than from what they later remembered doing. This closes part of the memory gap that makes interview-based mapping unreliable, the map inherits the doer's real clicks, not their tidied recollection.
AI Model to Map reached general availability on 6 January 2026. It converts technical BPMN, the notation analysts and mining tools speak, into business-readable procedures that a process owner can actually use. This is the feature that bridges the mine back to the map, and it earns a full treatment in Part IV because it is the mechanism that closes the loop.
What changed: mapping went from specialist project to shared habit
The practical effect of readable maps plus AI drafting is that process documentation stops being a project and becomes a habit, which is the only state in which it stays accurate. A project ends; a habit persists. Exhibit 7 contrasts the old mode with the new one, and the difference is less about technology than about who is allowed to participate.
| Dimension | Traditional mapping | Nintex Process Manager |
|---|---|---|
| Author | Process specialists | Front-line owners, AI-assisted |
| Language | Formal notation | Plain business steps |
| Starting point | Blank page | AI-generated draft |
| Update cadence | Every few years | Continuous, in-flow |
| Where it lives | A shared drive, unread | The tool people work in |
| Accuracy over time | Decays | Maintained by owners |
The map is necessary and, on its own, insufficient. However current and readable, it remains a record of intention. It tells you how the work is meant to run, and, because owners keep it honest, it tells you that better than most organisations manage today. What it cannot tell you is whether the work actually runs that way in the millions of cases nobody watched.
That is not a criticism of Process Manager; it is a description of what a map is. A map of a city does not show you the traffic. For the traffic you need a different instrument, one that reads the record of what actually moved. In Nintex terms that instrument is Process Intelligence, a map with no mine is a plan with no feedback, and it is the subject of Part III.
Process Manager makes the designed process a living, owned, business-readable document, and recent AI features have cut the cost of producing it. That is the map. The next part reconstructs the territory.
Four capabilities
The map-vs-mine gap
10 · Where time is lost
11 · Map vs mine
Process Intelligence reconstructs the real process from data you already generate
Nintex Process Intelligence reconstructs how work actually runs, case by case, deviation by deviation, from the event logs your ERP and CRM systems already produce as a by-product of doing business. Nothing new needs to be recorded. Every time an order is raised, an approval clicked, or a record updated, the system stamps it with an identifier and a time. Process mining reads those stamps back into the story of the process, told by the process itself.
The method is worth understanding because its credibility rests on it. Enterprise systems log events: a case identifier, an activity, a timestamp, sometimes an actor. Process mining takes millions of those events and reassembles them into the paths cases actually followed, which steps, in which order, with what waits between them. The output is not an opinion about the process. It is the process, as recorded by the machines that ran it.
This is why mining beats interviews for factual questions. Nobody has to remember, estimate, or defend anything. The log does not round its cycle times, forget its reruns, or omit the awkward variant. It shows every path every case took, including the ones no one would have described in a workshop.
The honesty cuts both ways, which is the caveat we keep returning to. The log records what happened, not why, and it records only what the systems captured. Work done in email, on paper, or in a spreadsheet leaves no event to mine, so a process picture built only from system logs has its own blind spots. Good mining pairs the data with the people who can explain it. Volge treats that pairing as part of the method, not an optional extra.
Nintex positions Process Intelligence as a capability spanning four functions rather than a single feature; Exhibit 8 lists them. The distinction between mining and mapping inside that set is the hinge of the whole system.
| Capability | What it does |
|---|---|
| Process mining | Reconstructs actual process flow from system event logs |
| Task mining | Captures how work is done at the desktop, step by step |
| Process modeling | Represents the process logic for analysis |
| Process mapping | Renders the process in a readable form |
We describe Process Intelligence as a capability and category, not a confirmed standalone product line. That is Nintex's own framing (Nintex, 2 December 2024) and we have not embellished it. What matters commercially is the function, mining the real process, which Nintex delivers and Volge operates.
From event log to reconstructed process, in four moves
Process mining is easier to trust once you can see the mechanism. There is no black box worth being nervous about; there is a pipeline that reads timestamps and rebuilds sequences. The four moves below are what happens between your systems and the picture on the screen.
| Move | Question it answers | Typical finding |
|---|---|---|
| Extract | What did the systems record? | Millions of timestamped events per process |
| Reconstruct | What paths did cases follow? | Many more variants than anyone expected |
| Analyse | Where is time and money lost? | A few steps hold most of the delay |
| Compare | Where does reality break from design? | Deviations no one had flagged |
System logs miss work done between systems, the copy-paste, the manual reconciliation, the swivel-chair step where someone reads one screen and types into another. Task mining captures that desktop activity directly, so the picture includes the work that never touched a logged transaction. Together, process mining and task mining leave fewer blind spots than either alone.
Delay concentrates, and mining shows you exactly where
In most processes a handful of steps hold the large majority of the elapsed time, and the whole value of mining is that it names those steps instead of leaving you to guess. Waiting, not working, is usually the villain. A case spends far longer sitting in a queue than being actively handled, and the queues are invisible to a swimlane diagram that has no axis for time.
The pattern in Exhibit 10 is illustrative, and we have labelled it as such, but its shape is familiar to anyone who has mined a transactional process. The value-add work, the part a person is genuinely paid to do, is a thin slice. The rest is waiting, reworking, and shuffling data between systems. That composition is invisible in interviews, because people describe the work, not the waiting between it.
This changes where automation should point. Faced with a process like this, the instinct is to automate the value-add step, because that is the part everyone can see. The data says otherwise: the return is in the approval queue and the rework loop, where automation and process redesign remove waiting rather than speeding up work that was never the bottleneck.
It also disciplines the AI conversation. An agent set loose on the seven per cent of value-add work leaves ninety-three per cent of the elapsed time untouched. An agent, or a redesign, aimed at the approval wait and the rework loop attacks the actual cost. Mining is what tells you which is which, before you commit the budget.
Stop asking "which task can we automate?" and start asking "where does a case wait, and why?" The first question flatters the tools you already bought. The second finds the money. Mining answers the second.
The mine is only as complete as your digital exhaust. Work done on paper, over a call, or in a system the logs never reach stays invisible to it, which makes the mine a truer picture than the map, not the whole truth.
The gap between the map and the mine is the product
The single most valuable artefact Nintex can produce is not the map and not the mine but the difference between them: the gap where the designed process and the performed process disagree. That gap is not a failure to be tidied away. It is a costed, prioritised list of everything worth improving, automating, or deliberately leaving alone, and it is the reason to buy both instruments rather than either.
| The map says | The mine shows | The gap is worth |
|---|---|---|
| One approval path | Eleven live variants | Standardisation, then automation |
| Approval takes hours | Approval waits days | The largest cycle-time prize |
| Controls always applied | Controls skipped at scale | Audit and compliance exposure |
| Handoffs are clean | Cases loop back repeatedly | Rework elimination |
| Process is stable | Behaviour drifts by quarter | Continuous monitoring |
Neither instrument delivers this alone, which is the commercial heart of the argument. The map without the mine is a plan you cannot check. The mine without the map is a pile of reality with no baseline to judge it against, you can see that cases took eleven paths, but not which of the eleven you intended. Only the comparison produces a worklist, and only a worklist justifies a budget.
The gap is also self-prioritising, which C-level readers will appreciate. Each entry carries a size, the cycle time it wastes, the volume it touches, the control it bypasses. That turns process improvement from a matter of taste and politics into a matter of arithmetic. You work the gap in order of value, and you can say why each item sits where it does.
One direction of the gap remained awkward until recently. Mining speaks BPMN: technical notation that analysts read and process owners do not. So the mined reality, however valuable, tended to stay trapped in the analyst's tool, unread by the people who owned the work. The feature that unblocked this shipped in January 2026, and it is the subject of Part IV.
The variant explosion is the finding that surprises boards most
The first mined result that unsettles a leadership team is almost always the variant count: the discovery that a process everyone described as one path was, in the logs, dozens. This is not chaos, or not only chaos. Some variants are sensible local adaptations; some are quality gates; some are genuine mess. The point of mining is that you can finally tell which is which.
The "long tail" here bundles dozens of rare paths that individually carry little volume but collectively resist automation and hide risk.
Read Exhibit 12 the way an operator would. The designed path carries a comfortable-looking majority, which is why interview-based mapping feels adequate, most of the time, most cases really do follow it. The trouble is the rest. The top handful of variants carry real volume, and the long tail, though thin per path, collectively defeats any automation built for the main path alone.
The strategic move is not to eliminate variants on sight. Some earn their place. The move is to sort them: standardise the ones that are needless divergence, protect the ones that are deliberate, and automate the main path only once it is stable enough to be worth the build. Mining supplies the sort; judgement supplies the decision.
This is where the map earns its keep a second time. Compared against the mine, it tells you which variants were intended and which crept in: the difference between a designed branch and an undocumented workaround. Without the map, every variant looks the same. With it, the mess sorts itself into signal.
A high variant count is not proof of a broken organisation. It is proof that you can now see the organisation. The count was always that high; you simply could not measure it. Measurement is the first improvement.
You now have reality; the problem is who can read it
At the end of mining you hold something most organisations have never had: a factual, complete, case-level account of how the work truly runs. The remaining obstacle is one of translation. That account is expressed in BPMN, and BPMN is a language of analysts, not of the process owners who must act on what it says.
This is a real barrier, not a cosmetic one. A finding that only the analytics team can read is a finding that stalls. The process owner in finance or procurement, the person with the authority and the context to change the process, cannot act on a diagram they cannot interpret. So the mined truth sits in a specialist tool, admired and unused, while the map in Process Manager stays comfortably out of date.
The whole system Nintex has assembled only pays off when the two directions connect: design flowing into reality for comparison, and reality flowing back into design for action. For years the second direction was the weak link. Mining could expose the truth, but returning that truth to the business in a form it could use required manual translation that rarely happened.
That link closed on 6 January 2026, when AI Model to Map reached general availability and turned technical BPMN into business-readable procedures automatically. It is the mechanism that makes the map-and-mine argument operational rather than merely elegant, and it is where Part IV begins.
Process Intelligence reconstructs the real process from data you already keep, exposes where time is lost and how many variants exist, and, set against the map, produces the costed gap that is the actual product. The one missing piece was translation. Part IV supplies it.
Insight to action
Proof & ROI
14 · Proof points
15 · The return
AI Model to Map turns the mined truth into something the business can act on
AI Model to Map is the feature that makes the whole map-and-mine system work in practice, because it carries the mined reality across the one barrier that used to stop it: language. Generally available since 6 January 2026, it converts technical BPMN, the notation mining produces, into business-readable procedures that a process owner can read, question, and change. The mine flows back into the map, automatically, and the loop closes.
Recall the impasse from Part III. Mining produces truth in a technical dialect. The people empowered to act on that truth do not speak the dialect. The translation between them was, until this year, a manual job that competed with everything else on an analyst's desk and usually lost. So insight accumulated faster than anyone could convert it into action.
Model to Map removes the translation cost. The mined process, expressed in BPMN, becomes a procedure in plain business steps inside Process Manager: the same tool the front line already uses to own its maps. What was locked in the analyst's screen becomes a document the process owner can read on Monday and change by Friday. The bottleneck between knowing and doing is the one being automated.
This is why we held the feature back from Part II despite it being a Process Manager capability. Its significance is not that it drafts a map; the AI Process Generator already does that. Its significance is directional. It runs mine-to-map, closing the return path that turns a diagnostic exercise into an operating loop. Chris Ellis of Nintex called the broader effort "a major step forward in making process intelligence accessible, actionable, and scalable," and the accessibility he names is exactly this: reality, rendered readable.
With the loop closed, the sequence from Part I finally runs end to end without a manual gap in the middle. Discover, map, mine, compare, act, and then map again, because acting changes the process and the cycle repeats. Exhibit 13 draws the closed loop.
AI Model to Map has been generally available only since January 2026. We describe it on Nintex's framing and our own reading, not on a track record across dozens of enterprises. Automated translation also smooths, a mislabelled step reads as cleanly as a correct one, so treat the output as a strong draft an owner still checks, not an oracle.
The output of "owners act" becomes the next designed map, and the loop runs again, process understanding as a habit, not a project.
A readable process is only useful if it drives the daily work: DocGen and Forms make it do so
Understanding a process pays nothing until it changes what happens on the ground, and two Nintex enhancers carry the improved process from the diagram into the day-to-day work. Nintex DocGen generates documents from process data. Nintex Forms captures data into the process. Named here as enhancers rather than headline products, they are what stop an improved map from staying a map.
Forms captures the data a process needs at the point the process needs it, in a structured shape rather than a free-text email. Clean structured input is what makes downstream automation possible; garbage entering a well-designed process still produces garbage. Forms is the front door that keeps the process fed with data it can actually use.
DocGen generates the documents a process produces, contracts, quotes, statements, letters, directly from process data, without the copy-paste step that introduces errors and delay. The process becomes the source of the document rather than a prelude to someone assembling it by hand. That removes a whole class of the manual data entry Exhibit 10 flagged as pure waste.
Neither enhancer is the reason to buy. We are deliberately not over-featuring them, because the argument of this book is about seeing the process, not about any single tool for running it. But they matter to the C-level reader for a specific reason: they are how the improvement you identified in the gap actually reaches the people doing the work. Insight that never touches the daily task is a report, not a result.
Placed in the loop, Forms sits at the process entrance and DocGen at its exit, with the improved, mined-and-mapped process in between. Data enters clean, moves through a process you can now see and have deliberately shaped, and leaves as a finished document. The understanding from Parts II and III becomes operational rather than analytical.
This is the honest end of the "insight to action" promise. There is no single button marked "act". There is a readable process, a clean way to feed it, and an automatic way to produce what it owes the customer, and, above all, the diagnosis that told you which part was worth the effort.
DocGen and Forms are enhancers, not the headline. They turn an improved process into improved work: Forms feeds it clean data, DocGen produces its documents. Buy them for what they do to a process you already understand, not as substitutes for understanding it.
The evidence that this is a mature platform, not a promising demo
Nintex is a widely deployed, independently rated platform rather than an emerging bet, and the numbers say so across scale, satisfaction, and named customers. More than 7,000 organisations across over 100 countries run on it (Nintex, 24 September 2025). The customer list includes L'Oréal, Panasonic, Virgin Atlantic, Sonos, and CARFAX. This is the part of the argument that de-risks the purchase.
| Proof point | Figure | Source · date |
|---|---|---|
| Deployed scale | 7,000+ organisations, 100+ countries | Nintex, 24 Sep 2025 |
| G2 rating | 4.3 / 5 across 1,322 reviews | G2 |
| TrustRadius rating | 8.0 / 10 across 530 reviews | TrustRadius |
| Recognition | "Top Rated", seven consecutive years | TrustRadius |
| Named customers | L'Oréal, Panasonic, Virgin Atlantic, Sonos, CARFAX | Nintex |
Read the ratings for what they are: durable third-party signal from large review bases. A 4.3 out of 5 across 1,322 G2 reviews and an 8.0 out of 10 across 530 on TrustRadius are not the scores of a product that frustrates its users, and "Top Rated" for seven straight years is a statement about consistency rather than a single good quarter. These are noisy measures, but the sample sizes make the direction reliable.
The named customers matter for a different reason. L'Oréal, Panasonic, Virgin Atlantic, Sonos, and CARFAX are large, demanding organisations with the option to build or buy anything. Their presence is evidence that the platform survives contact with complex, regulated, high-volume operations: the conditions under which a lightweight tool would fail.
We should be equally clear about what we are not claiming. We do not assert that Nintex is a Gartner Magic Quadrant leader or a Forrester Wave leader, because we have not sourced those designations and will not imply them. The proof above stands on its own terms, which is the only kind of proof worth putting in front of a board.
The economics: a documented return, a fast payback, and a small council that proves the point
The financial case rests on a documented study and a concrete customer result, both attributed and both dated so you can weigh them yourself. Forrester's Total Economic Impact study of Nintex Workflow found a 213 per cent return on investment with a nine-month payback (Forrester, circa 2022, an older vintage, and we flag it as such). Port Stephens Council saved more than 35,000 Australian dollars a year using Process Manager (Nintex, 2 December 2024).
ROI bar shown to scale against a 100% reference; payback bar shown against a 24-month reference. Forrester figures are circa 2022 and describe Nintex Workflow specifically.
Treat the Forrester figures with the care their age deserves. A 213 per cent ROI and a nine-month payback are strong numbers, but the study is roughly four years old and concerns Nintex Workflow rather than the mapping and mining capabilities that are this book's subject. We include it as directional evidence that Nintex automation pays back quickly, not as a forecast of your specific result. The vintage is on the label for a reason.
The Port Stephens Council figure is smaller, more recent, and in some ways more useful. A single local council saving more than 35,000 Australian dollars a year with Process Manager (Nintex, 2 December 2024) is a modest, concrete, checkable result from exactly the mapping capability we have been describing. It scales in the imagination: if a council of that size finds that much in mapping alone, a multinational's landscape holds a great deal more.
The honest summary is that the returns are real and the exact number is yours to discover. What the map-and-mine approach changes is the confidence with which you pursue it. Instead of automating on faith and hoping for a Forrester-shaped outcome, you mine the gap, size each opportunity, and spend against evidence. The ROI stops being a promise and becomes a calculation.
The 213% / nine-month figure is real but dated (circa 2022) and specific to Nintex Workflow. The AUD $35,000+/year council result is recent (2 December 2024) and specific to Process Manager. Neither is a guarantee for your organisation. Both are reasons to run the calculation on your own processes: which is what Volge does first.
The whole argument, on one page
For the reader who arrived at this page first, and some will, here is the argument in full, compressed. Each line is defended somewhere in the preceding pages.
The technology case is made, and it is the easy half. A costed worklist changes nothing until someone is mandated to act on it; tools reveal the gap, but they cannot make an organisation close it. Who runs discovery, who reads the gap, and above all who owns the fix are the questions Part V turns to.
Objections answered
The independent case
Glossary · sources
Reader's checklist
What to ask before you buy: a short interrogation for the C-suite
Before committing to any process-understanding programme, a C-level buyer should be able to answer six questions, and if the answers are vague the programme is not ready. The questions are deliberately blunt. They separate a real diagnostic exercise from a tooling purchase dressed up as one, and any honest buyer should ask them before committing spend.
If the answer is a hunch rather than a number, start with mining, not automation. You are about to spend against a guess.
If nobody has compared the map to the mine, the honest answer is "we do not know", and that is the first thing to fix.
Ownerless processes rot. If you cannot name the owner, you cannot hold the accuracy, and no tool fixes an accountability gap.
An agent inherits your process picture. If the picture is the designed map, the agent inherits its blind spots. Mine first.
If the process lives in a notation only analysts read, insight will stall at the analyst's desk. Business-readable is not a nicety; it is the difference between knowing and doing.
Software bought and left unoperated is shelfware. Decide up front whether you will run this capability yourself or have it delivered by a partner.
A programme that can answer these six with numbers and names is ready to spend. One that answers with intentions and hopes should start with a diagnostic, which costs little and prevents the expensive mistake of automating the map.
The four objections a sceptical board raises, and the honest reply to each
A well-run board does not buy on enthusiasm, and the four objections below are the ones we meet most often, each reasonable, each answerable, none dismissed with a slogan. We set them out because an analysis that ducks the hard questions is not worth reading, and because the answers are, in every case, stronger than the objection.
You almost certainly do, and they are almost certainly out of date in the places that cost money. The question is not whether you have maps but whether you have measured them against reality. Most maps are sincere and wrong; mining tells you by how much. If the comparison shows a small gap, you have spent little to confirm good news. It rarely does.
Process mining reads event logs your systems already keep; it records nothing new. Task mining does observe the desktop, and it is optional, scoped, and deployed only where the logged systems leave a gap. We are candid that task mining touches how people work, which is exactly why we scope it narrowly and explain it plainly rather than switching it on everywhere by default.
The Forrester figure is, and we flagged it as circa 2022 and specific to Nintex Workflow rather than to mapping and mining. We do not lean on it as a forecast. The recent, concrete number, Port Stephens Council saving more than AUD $35,000 a year with Process Manager (Nintex, 2 December 2024), is smaller and more honest, and your own diagnostic produces a figure specific to you.
You can, and the licence is the easy part. The value leaks in the work around it: connecting event logs, sorting real deviations from sensible adaptations, keeping owners engaged, and turning a one-off diagnostic into an operating habit. That is judgement and persistence, not a login. It is what Volge delivers, and it is where programmes bought as software alone tend to stall.
Each objection assumes the purchase is the software. It is not. The purchase is a truthful, maintained understanding of how your work runs, and the discipline to act on it. The software is a means; the understanding is the thing you are paying for.
One process, end to end: how a supplier-onboarding review actually runs
A single worked example shows the whole method more clearly than any amount of description, so here is supplier onboarding, a process every large organisation runs, and one that reliably hides money. The figures below are illustrative, drawn to match the shape mining commonly reveals rather than any one client: a walkthrough, not a benchmark.
The documented process is tidy: request, credit check, compliance review, approval, and activation. Interviewed, the team confirms it. On the map, a supplier is live in about fifteen working days, and everyone believes it.
Mined from the procurement and finance logs, the picture changes. The median case takes twenty-six days, not fifteen. Compliance review carries most of the extra: cases sit in a review queue for days before anyone opens them. And there are not five steps but nine distinct paths, because two regions added their own checks that never reached the manual.
| Measure | Map (as designed) | Mine (as performed) |
|---|---|---|
| Median cycle time | ≈ 15 working days | ≈ 26 working days |
| Distinct paths | 1 | 9 |
| Largest delay | None assumed | Compliance review queue |
| Control adherence | Assumed 100% | Credit check skipped in ~1 case in 12 |
The gap sorts itself. The compliance queue is the cycle-time prize, and the fix is a redesign plus a reminder, not a bot. The skipped credit check is a control exposure to close before it becomes an audit finding. Two of the nine paths are sensible regional checks worth keeping; the rest are drift to standardise. Only then, on the standardised main path, does automation earn its build.
AI Model to Map returns the mined, standardised process to Process Manager as a readable procedure the onboarding owner maintains. Forms captures the supplier's data cleanly at the front; DocGen produces the onboarding pack at the end. The next mining run checks whether the fixes held. They usually do, until the process drifts again: which is why the loop runs continuously rather than once.
Build, buy, or have it delivered: an honest comparison
There are three ways to acquire process-understanding capability, and each suits a different organisation, so we set them side by side rather than pretending only one is defensible. Build it yourself. Buy the software and staff it. Or have it licensed and delivered through Volge. We sell the third, and we will still tell you when one of the others fits you better.
| Dimension | Build in-house | Buy software, self-run | Delivered through Volge |
|---|---|---|---|
| Time to first insight | Slow: months to tooling | Medium: licence then learn | Fast: diagnostic in weeks |
| Where value leaks | Reinventing mining | Log connection, gap reading | Managed as part of delivery |
| Ongoing burden | You own it all | You staff and operate it | Single accountable partner |
| Best for | Firms with deep in-house data science | Firms with spare process capacity | Firms wanting the outcome, not the project |
Building it yourself makes sense for a small number of organisations: those with mature data-science teams who would rather own the whole stack. For most, reconstructing what Nintex already ships is an expensive way to arrive at the same place two years later. We would talk you out of it unless you have a specific reason.
Buying the software and running it yourself is the middle path, and it works when you have spare process capacity and the patience to climb the learning curve on log connection and gap reading. The risk is familiar: the licence gets bought, the diagnostic never quite happens, and the tool becomes shelfware. Most stalled programmes we are asked to rescue started here.
Delivery through Volge suits the organisation that wants the outcome rather than the project, a truthful process picture, a worked gap, and a running loop, with one partner accountable end to end. It is what we do, and we have priced the first step low enough that choosing it is a small decision, not a large one.
What independent research finds about automating blind
The argument of this book, see the process before you automate it, is not Volge's alone. Independent analysts, working from their own samples, report the same pattern: programmes that skip process truth stall, and the ones that scale are the exception.
| Finding | Figure | Source |
|---|---|---|
| Firms that have scaled RPA beyond 50 robots | 3% | Deloitte Global RPA Survey, 2018 |
| Digital transformations that fall short of objectives | 70% | BCG, Flipping the Odds, 2020 |
| GenAI projects to be abandoned after proof of concept by end-2025 | 30% | Gartner, 29 Jul 2024 |
| Enterprise GenAI pilots with no measurable P&L impact | ~95% | MIT NANDA, Aug 2025 |
| Executives calling their GenAI rollouts “mature” | 1% | McKinsey, State of AI, Mar 2025 |
Read down the column and one story repeats: the tooling scales faster than the understanding beneath it, and the gap is where the money leaks. Deloitte's 3% is the share of firms that got RPA past a pilot fleet; the rest stalled. The most common reason, across these studies, is that the process the software was pointed at was not the process the business actually ran.
The 2025 AI figures sharpen the point. When Gartner expects a third of GenAI projects abandoned after proof of concept, and an MIT group finds almost no enterprise pilots moving the P&L, the binding constraint is not model quality. It is that agents are pointed at operations no one has mapped against reality: the readiness layer this book calls process intelligence.
A serious argument should survive contact with sources that have no stake in it. Every figure above comes from an analyst or academic group independent of both the vendor and its partner.
Glossary
Plain definitions of the terms used in this book, in the sense we have used them.
- Process as designed (the map)
- How a process is meant to run, as documented. Held in Nintex Process Manager.
- Process as performed (the mine)
- How a process actually ran, reconstructed from data. Surfaced by Nintex Process Intelligence.
- Nintex Process Manager
- Business process mapping and modelling. Formerly Promapp; rebranded 2022. Tagline: identify, map, and model your processes.
- Nintex Process Intelligence
- Process mining and related capability, mining, task mining, modeling, mapping. A category and positioning, not a confirmed standalone SKU (Nintex, 2 Dec 2024).
- Process mining
- Reconstructing the real sequence and timing of a process from event logs.
- Task mining
- Capturing how work is performed at the desktop, filling the gap between logged systems.
- Event log
- The timestamped record enterprise systems keep of what happened: the raw material of mining.
- BPMN
- Business Process Model and Notation, the technical notation mining produces, which AI Model to Map renders business-readable.
- AI Process Generator
- Drafts a process map from a description. Beta, March 2025.
- Process Capture
- AI activity-recording that auto-documents a task as performed. GA September 2025.
- AI Model to Map
- Converts technical BPMN into business-readable procedures. GA 6 January 2026. Bridges mine to map.
- Nintex DocGen
- Generates documents from process data. An enhancer.
- Nintex Forms
- Captures structured data into a process. An enhancer.
- The gap
- The difference between the designed map and the mined reality, variants, bottlenecks, deviations, and the costed worklist it forms.
- Variant
- A distinct path cases actually take through a process, often undocumented.
- IPA
- Intelligent process automation: the market category sized in Exhibit 2.
Methodology & editorial standards
This book follows a small set of rules designed to keep a disclosed-interest analysis honest, and we state them so you can hold us to them. Every external figure is attributed and dated. Vendor positioning is labelled as positioning. Market sizing is labelled as estimate. Illustrative figures are labelled as illustrative. Our commercial interest is disclosed wherever it bears on the argument.
Where a number comes from Nintex, we say so and give the date. Where it comes from a third party, Market.us, Forrester, G2, TrustRadius, we name the source. The Forrester ROI figure is flagged as circa 2022 and specific to Nintex Workflow, not to the mapping and mining capabilities that are this book's subject. The market-size figure is a Market.us estimate, and the interpolated years in Exhibit 2 are our own arithmetic at the stated growth rate, marked as such.
Exhibits 10 and 12, where time is lost, and how variants distribute, are illustrative composites, not measurements of a named client. They show the shape mining typically reveals, and we have labelled them so no reader mistakes a representative pattern for a benchmark. Your own numbers are the only ones that matter, and finding them is the first thing a Volge engagement does.
We make no claim that Nintex is a Gartner Magic Quadrant leader or a Forrester Wave leader, because we have not sourced those designations for Nintex specifically. We do note that process mining is an analyst-recognised category: Gartner has published a Magic Quadrant for Process Mining Platforms (April 2025). We describe Process Intelligence as a capability and category rather than a confirmed standalone product, following Nintex's own framing.
Volge is a Nintex delivery partner and benefits if you buy. We have handled that conflict by sourcing every external claim independently, including the analyst figures on page 31, which come from Deloitte, BCG, Gartner, MIT and McKinsey, none of whom have a stake in this book, labelling our estimates, and keeping the first step small.
If any figure here has aged or any attribution is wrong, we would rather know. A field guide is only useful while it is accurate, and this is a first edition.
Sources
Every external claim in this book, with source and date. Listed in order of first appearance.
- Market.us (2024): Intelligent process automation market estimated at US$16.81bn (2024), projected US$61.23bn by 2034 at 13.8% CAGR. Labelled throughout as an estimate.
- Nintex, Niranjan Vijayaragavan, CPTO: Forbes Technology Council (3 February 2026): "The AI entry point no one is talking about: process intelligence."
- Nintex (2022): Rebrand of Promapp to Nintex Process Manager confirmed.
- Nintex: AI Process Generator (beta, March 2025): Drafts a process map from a description.
- Nintex: Process Capture (GA, September 2025): AI activity-recording auto-documentation.
- Nintex: AI Model to Map (GA, 6 January 2026): Converts technical BPMN into business-readable procedures.
- Nintex, Chris Ellis, Director of Solution Engineering: "A major step forward in making process intelligence accessible, actionable, and scalable."
- Nintex (2 December 2024): Process Intelligence capability spanning process mining, task mining, process modeling, and process mapping; Port Stephens Council saved AUD $35,000+ per year with Process Manager.
- Nintex (24 September 2025): 7,000+ organisations across 100+ countries.
- G2: 4.3 / 5 across 1,322 reviews.
- TrustRadius: 8.0 / 10 across 530 reviews; "Top Rated" seven consecutive years.
- Nintex: Named customers include L'Oréal, Panasonic, Virgin Atlantic, Sonos, and CARFAX.
- Forrester: Total Economic Impact of Nintex Workflow (circa 2022): 213% ROI, nine-month payback. Vintage flagged; specific to Nintex Workflow.
Deloitte Global RPA Survey (2018, n=424): 3% scaled RPA beyond 50 robots. BCG, Flipping the Odds (2020): 70% of digital transformations fall short. Gartner (29 Jul 2024): ≥30% of GenAI projects abandoned after PoC by end-2025. MIT NANDA, The GenAI Divide (Aug 2025): ~95% of enterprise GenAI pilots show no measurable P&L impact. McKinsey, State of AI (12 Mar 2025): 1% call GenAI rollouts "mature". Gartner MQ for Process Mining Platforms (15 Apr 2025). None have a stake in this book.
Volge-analysis exhibits and the illustrative composites (labelled inline) are our own; interpolated market years are Volge arithmetic at the stated CAGR.
The reader's checklist
Tear-out version. Six questions from the operator's read (page 27), on a single page you can take into a meeting.
| Ask | If the answer is vague |
|---|---|
| Which processes cost us most? | Mine before you automate |
| How far is the map from reality? | Run the comparison first |
| Who owns each process, by name? | Assign ownership before tooling |
| Is our AI target picture true? | The agent inherits your blind spots |
| Can operators read the process? | Insight will stall at the analyst's desk |
| Who holds and operates the licence? | Decide, or it becomes shelfware |
What you need in place to map and mine
Process intelligence is the lightest of the three layers to buy, but not free of prerequisites. Two lists follow: the Nintex elements to licence, and the non-Nintex pieces that must already be in place for a map-and-mine engagement to produce anything. Tiers and figures are indicative for 2026; confirm current SKUs with Nintex.
| Nintex Process Manager (Promapp) | The mapping and operating-procedure library. Platform subscription, priced by user. |
| Nintex Process Intelligence | Mines system event logs to reconstruct the real process. Add-on module, priced by data volume. |
| Process Capture | Captures steps into draft maps. Bundled with Process Manager or licensed as an add-on. |
| AI Model to Map (2026) | Turns mined event data into a readable map. A platform AI feature; check tier eligibility. |
| Source-system event logs | ERP, CRM, or ITSM logs, or their exports. You cannot mine what you cannot extract. |
| Identity / single sign-on | Microsoft Entra ID or a SAML provider, for user access. |
| Data-governance clearance | Sign-off to pull operational logs, with privacy and retention rules agreed. |
| Process owners' time | Named owners to validate the mined maps against how work really runs. |
Nothing here needs a server or a data-centre. The gating item is not a licence, it is access to the logs, and the owners' time to read what the mine finds.
One accountable team, from strategy to shipped software
Volge delivers process intelligence on the Nintex platform through one team and one arc. Four roles hold the work from the first workshop to operational handover; nothing is handed off between phases.
One process-intelligence approach, delivered tool-based or manual
See the work before you change it. The five phases below are how Volge runs process intelligence on the Nintex platform; only the delivery changes.