Adaptive Process Orchestration (APO) is the software category that Forrester defined in September 2025: automation platforms that combine AI agents and non-deterministic flows with traditional deterministic flows. Forrester expects fewer than 15 % of companies to turn on the agentic features of their automation suites.
What Adaptive Process Orchestration (APO) is
Adaptive Process Orchestration is the name Forrester gives to an automation platform that uses AI agents and non-deterministic control flows, in addition to traditional deterministic flows, to fulfil business goals, perform complex tasks and make autonomous decisions. The definition appears in the announcement that analyst Bernhard Schaffrik published on 4 September 2025, where he presents it as a new software category.
In business terms: a process has stretches where the path is written down (“if the amount is over €3,000, management approves”) and stretches where what is fixed is the outcome and the limits (“read this email, work out what the customer is asking for and prepare the reply”). Until now, the first was done in a process engine and the second, when it was done at all, in a separate AI tool. APO describes platforms that run both in the same process.
According to Forrester, these platforms focus on long-running or complex processes that span several systems, and they build on the automation components that already exist: integrations, workflow engines and external automations. They do not start from scratch or ask you to throw away what works.
Why it is appearing now
APO is appearing because the automation tools of the last decade were designed for fixed paths, and AI agents are not fixed. Forrester says so plainly in its announcement: deterministic workflow engines, RPA bots and digital process automation (DPA) tools are not powerful enough for the complexity that autonomous operations demand.
But the same firm adds three caveats, and all three matter to buyers:
- The deterministic engine is not going away. In the November 2024 report that opened this line of analysis, Forrester stated that the deterministic process engine would remain indispensable for the following three years.
- Real adoption is slow. In its predictions for 2026 (3 November 2025), Forrester estimated that fewer than 15 % of companies will turn on the agentic features of their intelligent automation suites.
- Knowing the process is what saves projects. The same publication expects process intelligence to rescue 30 % of failed AI projects.
Taken together, the picture calls for caution: the direction is clear, but almost nobody has got there. Anyone who presents APO as something everybody already has in production has not read the source. We discuss why so many projects stall at the pilot stage in how to take an AI pilot to production, and how to get to know the real process in process mining.
The five capabilities Forrester asks of an APO platform
Forrester lists five capabilities a platform must add to manage processes with agents. The table translates them into what they mean in practice and the question you can ask in a demo:
| Capability (Forrester) | What it means | Question for the demo |
|---|---|---|
| Managing model options and constraints | Choosing which AI model each step uses and within what limits | Can I switch models without rebuilding the process? Where does it run? |
| Processing content and formats | Reading documents, emails and images and turning them into process data | Does it read my real documents or does it need a separate tool? |
| Creating AI agents | Defining an agent with its goal, its tools and its permissions | Who can create an agent, and what can that agent touch? |
| Agentic orchestration | Coordinating agents, people and systems within the same process | Show me a process where an agent and a person hand work to each other |
| Governing and protecting data and intellectual property | Controlling which data the AI sees and keeping a record of what it decides | What is recorded for each agent decision, and who can look it up? |
None of the five is new on its own. What is new is requiring them together and inside the process, not in five different products. The fifth carries the most weight in Europe: the inventory of systems required by the EU AI Act is fed by that record, as we explain in keeping a register of AI systems.
APO versus BPM, RPA, iPaaS and BOAT
APO does not replace the earlier categories: it brings them together and adds the steps where an agent decides. This comparison is our own, based on the public definitions of each term:
| Term | What it automates | How the path is decided | Relationship with APO |
|---|---|---|---|
| BPM / DPA | Processes with human tasks, rules and deadlines | Deterministic: the flow is designed in advance | It is the foundation. APO adds non-deterministic steps to it |
| RPA | Repetitive tasks in applications without an API | Deterministic: the bot repeats recorded clicks | One more resource the process can call on |
| iPaaS | Data exchange between applications | Deterministic: trigger and actions | The integration layer that processes and agents act on |
| BOAT (Gartner) | All of the above in one platform, with agent orchestration | Both | Another analyst firm’s label for a similar convergence |
| APO (Forrester) | Long-running or complex processes with AI agents | Both, in the same process | — |
If you have read our guide to BOAT, Gartner’s category, the sense of familiarity is justified. Two analyst firms have reached the same conclusion by different routes: Gartner stresses the consolidation of tools and Forrester the mix of fixed and adaptive flows. For the difference between RPA and BPM, which still holds, see low-code BPM versus RPA.
Forrester announced a market overview report (Landscape) for early in the second quarter of 2026 and a Forrester Wave evaluation for later in 2026. If a vendor shows you its position in either of them, ask for the date of the report and exactly what was evaluated.
Fixed flow, agent or person: what goes in each step
The rule of thumb is simple: a fixed flow where there is a rule, an agent where there is messy information and a person where there is judgement or risk. The most expensive mistake is having an agent decide something a rule already handled well.
| Type of step | Example | Who does it best |
|---|---|---|
| Business rule or legal obligation | Approval by amount, response deadline, dual sign-off | Deterministic flow |
| Reading or classifying unstructured information | Extracting data from an invoice, classifying an email, summarising a case file | AI agent, with a confidence threshold |
| Preparing a draft | A reply to a customer, a contract clause | AI agent, with human review |
| Deciding with consequences | Turning down a request, accepting an exception, offboarding a supplier | A person, with the information prepared |
| Moving data between systems | Creating the entry in the ERP, updating the CRM | Connector or API |
| Operating an application without an API | Entering data in a legacy portal | RPA |
A process in a small or mid-sized company rarely has more than two or three steps where an agent truly adds value. Identifying them is more useful than choosing a platform by its acronym. In agentic orchestration versus traditional BPM we explain how the two models coexist.
Eight questions for a vendor that claims to be APO
The label guarantees nothing: what you need to check is whether the AI steps live inside the process or are bolted onto it. These questions reveal it in a demo:
- Is the agent a step in the process, with defined input and output, or a separate tool that the process calls?
- Can I set a confidence threshold below which the step goes to a person?
- What is recorded for each agent run: input data, result, model and who validated it?
- Whose permissions does the agent act with: its own or those of the user it works on behalf of?
- Where do the models run, and where is the data they see stored?
- Can I swap an agent step for a rule, or the other way round, without rebuilding the process?
- Is the process modelled in a standard (BPMN) that I can export?
- How is AI usage charged: per user, per run or by model consumption?
Questions 2 to 4 are the ones that separate a governed agent from an experiment. We develop them in governed AI agents with BPM and, from the security side, in AI agent security in business processes.
How Dokuflex solves it: the agent is a step in the process
Dokuflex, the low-code BPM platform with AI, does not appear in Forrester’s reports on APO, which evaluate global vendors. What it shares with the category is the central idea: fixed steps and AI steps are run and governed in the same process.
- Processes in BPMN with human tasks, rules and deadlines in Dokuflex low-code BPM, built by the business team itself and exportable in a standard format.
- AI agents as process steps, with a confidence threshold, human validation and a record of every decision, as described on the AI agents page. Each agent acts with explicit permissions, defined by domain.
- AI document reading inside the same case file, with no intermediate tool.
- Integrations with ERP, CRM and payroll through connectors and APIs in the low-code iPaaS layer.
- Data processed in European data centres by default and an audit trail of every step, human or AI: who, what, when and with which data.
What Dokuflex does not do is decide for you which steps deserve an agent. That comes from looking at the real process, and it is exactly what we do in the first session. Bring a process and we will review it in 30 minutes.
Frequently asked questions
What is Adaptive Process Orchestration (APO)?
It is the software category Forrester defined in September 2025: an automation platform that uses AI agents and non-deterministic control flows, in addition to traditional deterministic flows, to fulfil business goals, perform complex tasks and make autonomous decisions. It focuses on long-running or complex processes that span several systems.
What is the difference between APO and BPM?
Classic BPM runs deterministic flows: the steps and branches are designed in advance. APO adds non-deterministic steps to that engine, where an AI agent decides how to reach an outcome within set limits. It does not replace it: Forrester stated in 2024 that the deterministic process engine would remain indispensable for the following three years.
Are APO and BOAT the same thing?
They are labels from two different analyst firms for a similar convergence. BOAT is Gartner’s name for platforms that unify process automation, RPA, integration and document reading with agent orchestration. APO is Forrester’s name and stresses the combination of deterministic and non-deterministic flows. For a buyer, the questions to ask are almost the same.
What is a non-deterministic flow?
It is a stretch of the process where the path is not written in advance: an AI agent is given a goal, some tools and some limits, and the agent decides the steps. Two runs with the same input may follow different paths. That is why it needs a confidence threshold, human validation and a record of every decision.
Does an SME need an APO platform?
It does not need to buy a category. It needs a place where its processes run with rules, deadlines and an audit trail, and the ability to add an AI agent to the steps where it helps (reading documents, classifying, drafting) without losing that control. If the platform it already uses allows this, it meets the idea behind APO even if it appears in no report.
How many companies already use agentic features in their automation?
Few. In its predictions for 2026, published on 3 November 2025, Forrester estimated that fewer than 15 % of companies will turn on the agentic features of their intelligent automation suites. The same publication expects process intelligence to rescue 30 % of failed AI projects.
Sources
Consulted on 13 November 2026. We cite Forrester’s open publications; the full reports are paid content.
- Forrester, “Announcing The Evaluation Of The Adaptive Process Orchestration Market” (Bernhard Schaffrik, 4 September 2025): definition of APO, characteristics and report schedule.
- Forrester, “Predictions 2026: Automation At The Crossroads” (Leslie Joseph, 3 November 2025): fewer than 15 % will turn on agentic features; process intelligence will rescue 30 % of failed AI projects.
- Forrester, “Reimagining Automation: The Need For Agentic Process Management” (Craig Le Clair, 13 November 2024; public summary): five capabilities and the continued role of the deterministic engine.
- Process Excellence Network (27 November 2025): coverage of the prediction on process intelligence.
Fewer acronyms and one of your processes up and running, with AI where it helps
Bring the process that generates the most email and spreadsheets in your company. In 30 minutes we map it with you and mark which steps belong to a rule, which to an agent and which to a person. Or create a free account and get started on your own.