Our own score
We scored 15 out of 25
We built an instrument that scores how ready a business is for AI agents, and ran it on ourselves before running it on anybody else. Every point below cites something you can check from your own browser. We score well on the two domains a visitor can see and poorly on the three that have to be built, and those three are what we are building.
Reading the number
The total hides the shape, and the shape is the useful part
Fifteen out of twenty-five sounds like a middling result. Two strong domains and three weak ones is a plan. Every business we have scored so far has this shape, because Discovery and Data are what a website already does by accident and the other three have to be decided on purpose.
The verdict band for 15 to 19 is Understandable, which means an agent can work out what we sell and what it costs without a person in the loop, and still cannot transact. We sit at the bottom of that band, one point above the Discoverable band we opened in on 14 August 2026. The instrument caps any organization at Discoverable while Discovery scores below 3, and caps it at Understandable while Governance is below 3 and Transactions is 3 or higher. Neither cap applies to us yet, so 15 is the raw total rather than a constrained one. The second cap is the one to watch: the moment we make anything transactable, Governance at 2 becomes the ceiling on the whole score.
The scorecard
Five domains, with the evidence and the gap
Each card carries what earned the score and what is holding it down. The second half is the part worth reading, and it is the part most published assessments leave out.
- Discover 4 of 5
Discovery
An AI system can find this site, read every page of it as plain text, and see a stated position on what it may do with what it reads.
- A crawler policy answering search, AI grounding and AI training separately rather than as one blanket yes or no
- A hand-written index of the site built for models to read
- A generated file carrying every indexable page as one plain-text document
- A markdown version of all 190 pages, each at its own address
- Our organization record naming the four profiles that are the same company, so nothing has to be inferred
What holds it there. Freshness. No marketing page publishes a modified date and nothing watches for staleness, so a reader has no way to tell a page revised last week from one untouched for a year. Level 5 is an operating commitment rather than a build, and this is the half of it we have not made.
- Understand 4 of 5
Data
An agent can read what we sell and what it costs, and the machine-readable copy agrees with the page it sits on.
- Service and Offer schema across nine engagement pages, carrying real prices rather than a contact-us placeholder
- Every published price generated from the page it appears on, so the two cannot disagree
- Nineteen price points checked against the visible text with zero drift
What holds it there. No data contract states what a field means, how often it changes or who owns it. Nothing watches for staleness, so the first person to notice a stale price would be a reader.
- Access 3 of 5
Interfaces
An agent can ask this site what we sell and what it costs, and get data back rather than a page to parse.
- Four documented endpoints with a published specification that matches their behaviour
- One of them reads: the engagement catalogue, every tier and every price band, at a single address
- The catalogue is generated from the same structured data the pages publish, so an answer given to a machine cannot differ from the figure a person is shown
What holds it there. One read endpoint is one question answered. Level 4 wants the interfaces to cover what a buyer actually needs to establish, and availability, fit and scope are all still human conversations.
- Act 2 of 5
Transactions
Every path to getting work started runs through a person or an external calendar.
- A working enquiry path and a booking route that a human can complete without friction
What holds it there. No quote request, eligibility check or availability check can be called. An agent acting for a buyer has to hand back to its user at exactly the point the buying decision gets made.
- Govern 2 of 5
Governance
We have a written position on crawling and no controls over agents acting.
- A published preference on AI use, with enforcement at the network edge rather than on the honour system alone
- A rate limit on every form endpoint, enforced at the edge and failing open so an outage in the limiter cannot take the contact form down with it
What holds it there. The rate limit is a real bound and the ladder still refuses to credit it here, which is worth saying out loud because it happened to us. It was built to stop a script filling the CRM, so it bounds volume from an address rather than authority for an agent. A control that exists as a side effect of another decision scores 2 on this ladder whatever it protects. What is genuinely absent: identity for an agent, approval thresholds, spend limits, an audit log of agent actions, and a revocation path anybody has tested. We also said this score had to rise with the read endpoint and it did not, for a reason we will defend: the catalogue is public information behind no credential, so a key on it would guard nothing and would lock out the readers it was written for. The rule holds for the endpoint after this one. Anything that returns per-client data, quotes a figure we would honour, or costs us money to answer does not ship until the controls are built.
Check it yourself
Four of the five domains are scoreable without asking us anything
That is a design decision in the instrument rather than a convenience. Evidence anyone can verify from outside survives an argument in a boardroom, and it means we can score a prospect before the first call. It also means a competitor can score us, which is the reason this page exists.
Governance is the exception. Nothing about approval thresholds, audit trails or revocation can be verified from outside an organization, so that score is the one we are asking you to take on our word. When we score a client, Governance evidence is recorded as attested rather than verified for the same reason.
What has moved
The first run scored 14, and here is what changed since
A published score is only worth reading if it can go up, and only worth trusting if the thing that moved it is linked. Each row below names the domain, the movement and the date. Rows are added and never edited.
What happens next
Four things, ordered by how much score each one moves
This is the same sequencing we would give a client at 15 out of 25. The cheapest point goes first, and the governance work goes before the endpoint that needs it rather than after.
-
A modified date on every page, and something watching it
Not the date alone. Level 5 on this ladder is an operating commitment: staleness has to raise an alert to a named owner rather than wait for a reader to notice. This is the cheapest point on the board and the one we have put off longest. Takes Discovery from 4 to 5.
-
Agent identity and an audit trail
Registration, scope, thresholds and a revocation path, tested rather than assumed. It has to be built before the second endpoint rather than after it, because opening access before controlling it is the mistake we tell clients to avoid. Takes Governance from 2 to 4.
-
A callable eligibility check
Enough for an agent to establish whether an engagement fits its user before a person is involved. This is the one that needs the controls above it first, and it is the expensive one. Takes Transactions from 2 to 3.
-
A data contract on the catalogue
What each field means, how often it changes, who owns it, and something that notices when a price goes stale rather than waiting for a reader to. Takes Data from 4 to 5.
Come back and check. This page carries the date of the first run and the date of the last rescore, and both move with the number. A published score that quietly stops being true is the failure this whole page is about.
Why publish this
An instrument that scores its author 25 out of 25 is a brochure
We ran this on ourselves first because it was the only honest way to find out whether the scoring discriminates. A ladder that hands full marks to the company that wrote it measures nothing. 14 told us the instrument works, and it told us which three domains we had been describing rather than doing.
It also settles a question every buyer of this work should ask. Anyone can write about agent readiness. Publishing a score against your own instrument, with the evidence linked and a date on it, costs something to be wrong about.
Find out what your own score is
Four of the five domains we can score from the outside before you have engaged us. Bring your website and whatever agent access your team has already granted, and we will show you the point where a machine stops.