Why a refusal costs 40% of a governed action

Every agent action that reaches elan1 Govern ends in one of three outcomes. It is allowed, it waits for a person you have named, or it is refused. Each outcome is written to the same record. The question we had to answer when we set prices was simple to state and awkward to settle: what should a customer pay when the answer is no?

An earlier post on this site suggested that a refusal should cost nothing. That was wrong, and this note replaces it. The old address now redirects here.

A refusal is not a shortcut

It is tempting to think of a refusal as the cheap path: nothing was posted, nothing was paid, nothing changed in the ledger. In practice a refusal does exactly the same work as an approval up to the moment of decision.

1. 01The agent's identity is confirmed and its request is read in full.
2. 02Your rules are evaluated against the request, in order, with the data they need.
3. 03If a person must decide, the request waits for them, for up to 24 hours. An unanswered request expires and is refused.
4. 04The outcome, the rule that produced it and the person who decided are written to the record, which your auditor can check offline.

Only the last step differs: an allowed action is carried out with credentials the agent never sees, and a refused one is not. Everything that makes the decision trustworthy has already happened.

Why not charge nothing?

Because pricing shapes behaviour, including ours. If we earned only when an action went through, every rule we helped you write and every default we shipped would carry a quiet incentive to say yes. Over thousands of decisions a day, quiet incentives become policy.

A free refusal also hides a real cost from you. Refusals take compute, storage for the record and, often, a person's time. When those costs are invisible, nobody asks why an agent keeps proposing actions that your rules will never allow. That question is worth asking: it usually points to a badly scoped agent or a rule that needs a person to look at it again.

Why not charge the full price?

Because a refusal does not carry the work through to your system of record. There is no posting, no reconciliation of the result and no follow-up record of the change taking effect. Charging the full amount would make refusals expensive enough that a buyer might loosen rules to save money, which is the opposite of the point.

Forty per cent is where we landed after costing each step. It is high enough to say that a no is real work, and low enough that nobody is tempted to trade safety for a smaller invoice.

What we publish so you can check us

A price is only answerable if you can see what drives it. So every invoice shows refusal rates beside cost, by agent and by rule. If an agent's refusal rate climbs, you see it in the same place you see the bill, and you can decide whether the agent or the rule needs changing.

• Governed actions and refusals, counted separately, for every billing period.
• Refusal rate by agent, so a misbehaving agent is visible before it becomes expensive.
• Refusal rate by rule, so a rule that refuses almost everything gets a second look.
• Requests that expired unanswered, so you can see where approvers are overloaded.

The principle underneath

We say that elan1 is intelligence you can answer for. That has to apply to our commercial model as well as our software. A governance product that profits only from yes is not neutral, and a buyer should not have to trust that it behaves as if it were.

The full price list, including how governed actions are counted, is on the pricing page. If something there does not match this note, tell us and we will fix whichever one is wrong.