What is decision automation? Why decision autonomy wins
Decision automation makes operational calls like price, allocation and credit without a person. Five beliefs about it, and why decision autonomy is the goal.
Decision automation is software that makes an operational business decision, such as a price, a stock allocation, a substitute product or a credit release, without a person making the call, using rules, data and models. Updated November 2026.
Ask a vendor for its automation rate and you get a number. In one evaluation this autumn, the number was 95 percent in year one. The buyer, an operations manager at a manufacturer, answered with a better question: what rate would you put in the contract?
That question is the gap between decision automation and decision autonomy. Automation asks which calls a system can make. Autonomy asks how many calls still wait for your team, and whether that number falls every month. The five beliefs below show where the two part ways.
What is decision automation in B2B commerce?
In B2B commerce, decision automation means the system settles the calls that sit inside every customer request, so the request moves without waiting for someone to decide. A single quote or order usually carries several of them.
- Price: which agreement applies, and whether a requested discount sits inside the band.
- Allocation: which warehouse or batch serves the line when stock is short.
- Substitution: whether an alternative part is acceptable to this customer.
- Credit: whether the order ships while the account is over its limit.
- Lead time: which delivery date the business is willing to promise.
One customer request, five decisions inside it
- Order for 40 spare parts, delivery next week
- Price
- Allocation
- Substitution
- Credit
- Lead time
Each one can hold the request until someone decides.
- ERP
- CRM
- Pricing
- Customer history
Most of these never appear on a process diagram. They live in the heads of the customer service team, which is why a request can look simple and still take a day. The Decision Model sorts them by confidence and stakes.
Is decision automation just a rules engine?
No. A rules engine automates the decisions someone was able to write down. Most decisions on a B2B request were never written down. Researchers Thomas Davenport and Jeanne Harris found the same limit in early expert systems: extracting the right knowledge from seasoned decision makers, and keeping it current, proved harder than anyone expected (MIT Sloan Management Review).
Rules work for the stable part: list prices, standard lead times, fixed approval limits. They break on the judgment calls an experienced colleague makes in seconds: this customer always accepts the substitute, that one never does; this region quotes the old price until the end of the quarter. The difference between following rules and making decisions is the core of automation versus autonomy. Decision autonomy starts from the other end: it learns the unwritten calls from the people who make them.
Does a higher automation rate mean fewer decisions for your team?
Not necessarily. An automation rate counts transactions that went through untouched. It does not count the decisions your team still made on the ones that did not.
A request that fails on one of six decisions still lands on a person, and that person often re-checks all six. So a team can report a rising automation rate while the hardest work stays exactly where it was. The rate that matters is the share of decisions the system owns. The Human Dependency Ratio measures exactly that: manual decisions per unit of revenue. Decision autonomy is judged on that number, not on the transaction count.
Should only low-stakes decisions be automated?
No. Stakes decide how much evidence a decision needs, not whether a system may make it.
A price at the edge of the band for a key account is high stakes, but often the call is clear: the contract, the history and the margin all point the same way. Holding it for a manager adds a day and no information. The better design executes it and attaches the full reasoning trail, so anyone can see what was read, what was checked and why. Two things make that safe: a measured confidence level per decision, and an audit trail that finance and compliance can read. That is how decision autonomy reaches the calls that matter most, not only the easy ones. Regulation points the same way: the EU AI Act requires high-risk AI systems to record events automatically over their lifetime (Article 12).
Stakes set the evidence a decision needs, not whether it is made
Prepare for an expert
Options laid out, every system already read
Execute with a reasoning trail
What was read, checked and why, open to finance
Ask, then keep the answer
The same question is not asked twice
Execute
No one waits, nothing to review
Does decision automation take your experts out of the process?
It changes their role. The experts stop being the queue every request waits in, and become the source the system asks when policy says it should. Deloitte’s 2026 Human Capital Trends research makes the same case: treat decision-making as a strategic discipline first, then design how people and AI share it (Deloitte Insights).
When a decision falls outside what the system knows, it does not guess and it does not drop the request into a shared inbox. It asks the right person, a sales manager, a credit controller, a product specialist or a supplier, with the options already laid out and every system already read. The answer resolves the request. More importantly, it is kept, so the same question does not need to be asked again next week. This is the mechanism behind decision autonomy: every expert answer makes the next request less dependent on one.
Do you have to write every policy before you start?
No. Policies can be proposed from decisions the business already makes, and approved by the people who own them.
Every executed request leaves a record of what was decided and why. When the same expert answer appears again and again, the system can propose it as a policy, and a person approves or rejects it. That turns the slowest part of a classic automation project, the requirements workshop, into something that happens continuously and in small steps. It is also where process intelligence comes from: execution data that shows where decisions still stall. Decision autonomy grows this way, one approved policy at a time.
How an expert answer becomes policy
- 01
Request executed
Every decision is recorded with its reasoning
- 02
Expert asked
Outside known policy, the right person gets the options
- 03
Answer repeats
The same answer shows up on similar requests
- 04
Policy proposed
The system drafts it from decisions already made
- 05
Owner approves
The next similar request needs no one
What is decision autonomy, and how is it different?
Decision autonomy is the share of decisions in a revenue flow that complete within policy without waiting for a person, and that grows over time because every expert answer becomes policy. Decision automation is a capability. Decision autonomy is an operating state.
| Decision automation | Decision autonomy | |
|---|---|---|
| Unit | One decision type, configured | Every decision on every request |
| When it does not know | Stops and hands over | Asks the right expert, then keeps the answer |
| High-stakes calls | Usually excluded | Executed with a reasoning trail, or prepared for an expert |
| Over time | Stays where it was configured | Grows as answers become policy |
| Measured by | Automation or touchless rate | Decisions still needing a person, per unit of revenue |
The practical test is simple. If revenue grows 20 percent and the number of decisions your team makes also grows 20 percent, you have automated tasks, not decisions. With decision autonomy, the second number falls while the first rises. That gap is what the autonomy rate tracks.
Go Autonomous built its execution layer for decision autonomy. It runs on top of existing ERP, CRM and pricing systems, across every channel requests arrive through. It executes the decisions inside quotes, orders, claims and price inquiries, asks experts when policy says so, and writes each answer back so the next request needs fewer of them. Customers run at 99% first-time-right on the requests it executes.
Frequently asked questions
What is decision automation?
Decision automation is software that makes an operational business decision, such as a price, an allocation or a credit release, without a person making the call, using rules, data and models.
What is an example of decision automation?
A system that receives an order for a part that is out of stock, checks whether this customer has accepted the substitute before, confirms the alternative and books it in the ERP, without a person reviewing the line.
What is the difference between decision automation and decision autonomy?
Decision automation executes decisions it was configured for and hands over the rest. Decision autonomy covers every decision on a request, asks experts when policy requires it, and grows as their answers become policy.
What is decision autonomy?
Decision autonomy is the share of decisions in a revenue flow that complete within policy without waiting for a person, and that grows over time as expert answers become policy.
Is decision automation the same as a rules engine?
No. A rules engine covers decisions that someone has written down. Most decisions on B2B requests are judgment calls that were never written down, so they need a system that can reason over data and learn from experts.
How do you measure decision automation?
Count the decisions that still need a person per unit of revenue, not only the share of transactions nobody touched. If that number falls while revenue grows, the system owns more of the decisions.
Sources
- Davenport, T. H. and Harris, J. G. (2005), Automated Decision Making Comes of Age, MIT Sloan Management Review
- Deloitte (2026), Decision-making with AI, Global Human Capital Trends, Deloitte Insights
- European Union (2024), Regulation (EU) 2024/1689 (AI Act), Article 12: Record-keeping
- Go Autonomous, Autonomous Commerce Blueprint: What is the difference between automation and autonomy?
To see how decisions, experts and policy fit into one operating model, read what Autonomous Commerce is.
