TL;DR
- The valuable part of an expert GTM tool is the decision a buyer can act on. Data access makes that decision possible; the policy, interpretation and explanation make it useful.
- Our worked example qualifies SaaS businesses adding implementation capacity for UK customers. A hiring signal starts the research, while the evidence determines whether the account belongs in the queue.
- A callable tool needs a narrow promise: defined inputs, traceable reasons, useful uncertainty and an explicit next action. The caller retains control of the campaign.
- Treg provides documented routes for tool access and a feature-gated Hub model for recipes and scripts. Listing approval and maker balance credits affect how an expert product reaches buyers and funds its operation.
- The economics depend on accepted decisions, support and maintenance. A small per-call fee can look attractive until the first buyer disputes the answer.
The judgement worth packaging
A research agent can find a hiring page, summarise the company and return a contact. Someone still has to decide whether that combination deserves a salesperson's attention.
That is the part of the expert-tool idea that interested us. Experienced GTM operators make distinctions that are easy to lose inside a broad research prompt: a replacement hire looks different from additional capacity; a relevant title can belong to the wrong subsidiary; an interesting account can still have no suitable contact.
The question became what a buyer should receive if those distinctions were packaged into a callable tool. A long research report would leave much of the work with the buyer. A bare score would hide the reasoning. The useful product sits between them: a bounded decision, the evidence behind it and a clear route for handling the unresolved part.
For this walkthrough, we designed that product around SaaS businesses adding implementation capacity for UK customers. The account and responses below are constructed examples, and the prices are modelling assumptions. The design uses Treg documentation checked on 7 October 2026; no Forma Nôrden tool has been launched or benchmarked.
The GTM AI agent guide covers the broader caller. Here, the specialist contribution is the account decision that lets that caller continue with its job.
Give the tool one job
The starting brief contains 500 SaaS accounts. The commercial play concerns customer implementation: a service provider wants to identify businesses whose delivery capacity is expanding, then decide where its offer might be relevant.
“Is this company growing?” is too broad to be a useful product promise. The tool needs to establish three things: the correct SaaS operating business, current customer-implementation work and additional capacity serving UK customers. The ABM account scoring template gives those conditions somewhere explicit to live before they become code.
Our first policy uses a vacancy posted within 45 days and checked open within seven, plus expansion evidence from the previous 90 days. Those windows can change for another play. What matters is that the caller knows which policy produced the answer and which date the assessment used.
The outcome has three states. qualify means the conditions are supported. exclude means the evidence establishes a mismatch, such as expansion restricted to US delivery. review identifies a material question the available evidence cannot resolve.
That third state needs care. A limited search that finds no vacancy does not establish that a business is standing still. Equally, “needs more research” on every difficult account would create a manual queue of little value. A useful review identifies the missing condition, explains what was checked and tells the buyer what could resolve it.
Account ownership and campaign activation remain with the caller. The tool's promise ends at qualification and a handoff for review. Contact finding, verification and sending have different acceptance rules and costs; folding all of them into the first product would make its value harder to assess.
Working through the first account
Account 041 is Alder Implementation Cloud, a fictional SaaS business represented by the reserved domain example.com. A vacancy for an implementation manager appears on its careers page. At first glance, it looks like a plausible prospect.
The first useful question is whether the vacancy belongs to the same operating business as the account. A parent group's careers site might recruit for several subsidiaries. A software consultancy might use the same job title while implementing other vendors' products. Company identity and business model need to survive that check before the hiring language means much.
For Alder, the product and careers pages describe the same SaaS business. The role involves onboarding customers, coordinating integrations and managing deployment handovers. That supports customer implementation rather than internal IT work.
The next question is additional capacity. A vacancy on its own could replace someone who left. In this example, the role was posted on 14 September, and a separate company statement dated 18 September says two implementation positions are being added for UK deployments. Both sit inside the policy windows on the 7 October evaluation date, and the vacancy's open status is checked that day.
Those details support qualification. They still leave the buying budget unknown. Alder might intend to handle the entire expansion through its new employees. The account deserves examination for this play, but the record gives the salesperson a reason to investigate rather than a claim that a purchase is imminent.
The alternative outcomes show where the specialist judgement does useful work:
| Evidence on the account | Outcome | What changes the decision |
|---|---|---|
| Fresh customer-implementation vacancy and explicit added UK capacity | Qualify | The policy's material conditions are supported |
| Fresh role explicitly replacing a departing employee | Exclude from this play | Replacement does not establish additional capacity |
| Fresh vacancy with no explanation of capacity | Review | Hiring is supported; expansion remains unresolved |
| An evergreen recruitment page with no original posting date | Review | A current expansion event cannot be dated |
| Expansion explicitly limited to US delivery | Exclude | The geography falls outside the policy |
| Current employer sources disagree about the role's territory | Review | A decisive conflict needs resolution |
Source choice also affects the answer. An employer posting is stronger evidence of its current role than an aggregator's copied description. Two job boards carrying that same posting remain one underlying observation. If the employer's own current sources disagree, the conflict belongs in the answer; collecting more copies does not settle it.
For the eventual handoff, Maya Rowan, Alder's fictional Head of Implementation, is a plausible person to investigate. Current remit matters more than seniority alone. A relevant person without a verified address remains a research result, while a verified address for an unrelated person does not become a suitable contact. The account decision survives independently of either contact outcome.
Turn the reasoning into an answer
Once the account was worked through, the interface became easier to define. The caller supplies a domain, policy version, evaluation date and cost ceiling. The tool returns the decision with enough evidence to inspect it.
This is the request shape for our design, rather than a built-in Treg endpoint:
{
"request_id": "fixture-041",
"account_domain": "example.com",
"policy": "[email protected]",
"evaluation_date": "2026-10-07",
"contact_scope": "not_requested",
"max_provider_cost_usd": 0.12,
"max_maker_fee_usd": 0.40,
"max_total_cost_usd": 0.52
}
The policy version fixes the evidence windows and geography. An unsupported version or malformed domain should fail before paid research begins. Changing a threshold creates a different assessment configuration, so a cached result cannot silently stand in for the new request.
For Alder, the essential answer is short enough for another agent to consume:
{
"request_id": "fixture-041",
"result_kind": "illustrative_fixture",
"account": "Alder Implementation Cloud",
"resolved_domain": "example.com",
"policy": "[email protected]",
"schema_version": "1.0.0",
"decision": "qualify",
"reason": "Current customer-implementation hiring adds UK capacity",
"evidence": [
{
"supports": "customer_implementation_vacancy",
"url": "https://example.com/careers/implementation-manager",
"event_date": "2026-09-14",
"checked_date": "2026-10-07"
},
{
"supports": "additional_uk_capacity",
"url": "https://example.com/news/uk-delivery-expansion",
"event_date": "2026-09-18",
"checked_date": "2026-10-07"
}
],
"unknowns": ["External-service buying budget"],
"next_action": "account_owner_review"
}
The full stored record also needs the identity evidence, permitted source excerpts, fetch references, implementation revision and settled charges. Those fields let an operator investigate a dispute without asking the model to reconstruct its reasoning from memory.
Unknown budget and unknown expansion deserve different treatment. Budget lies outside this qualification promise. Expansion is one of its conditions. Keeping the uncertainty attached to the relevant field gives the caller a usable answer; a general “medium confidence” label would conceal the distinction.
Our preference is for a salesperson to be able to read the decisive evidence beside the reason. When that person disagrees, the correction should identify a rule or observation. “Bad lead” is understandable feedback, but it does not tell the product owner whether the fault was identity, freshness, geography or interpretation.
Where Treg enters the product
Treg supplies access to the services underneath this assessment. Its API documentation describes registered tools and server-held credentials. An assessment service can be registered for agents to call while its owner retains the policy and runtime.
The Hub is a separate product path. Treg's CLI reference documents recipes and sandbox scripts behind TREG_HUB_ENABLED. For this assessment, a script suits the branching: identity can fail early, evidence can conflict, and some accounts should stop before further spending. Access to that path and approval for a listing remain platform dependencies.
Underlying calls still need to be chosen deliberately. Ordinary provider relays retain native interfaces. Supported routed capabilities offer their own normalised inputs, selection and waterfall behaviour. The Treg protocol explains that distinction, including filter reporting and route cost controls.
A routed response is useful when it contains the fields the decision requires. A native provider call is useful when a decisive detail would otherwise disappear. Neither removes the need to retain the source or interpret what it establishes about the account.
The assessment also owns its overall spending rule. A ceiling on one routed call cannot cap a job that makes several calls. Settled costs and outstanding reservations both count towards the allowance, and an asynchronous pending response needs its saved reference followed through rather than a duplicate request.
The same discipline applies at the product boundary. If the caller loses the HTTP response and retries, the request ID and assessment parameters should reconnect it to the existing work. A deliberate reassessment with a new date is different from recovering an interrupted response. The coding-agent architecture guide covers how the surrounding application keeps that state.
These are part of the product's value. The buyer should be purchasing an assessment it can use and recover, rather than inheriting the job of coordinating every provider call.
What a useful decision costs
The pricing question starts with what has actually been delivered. A supported exclusion can save the buyer a pointless enrichment and outreach sequence. A completed assessment that identifies a genuine evidence conflict can also be useful. A timeout leaving the decisive research unfinished delivers much less.
For our design, a complete qualification or exclusion carries the full maker fee. A completed evidence assessment ending in review carries half. A technical failure or an identity-only fragment carries no maker fee. That is a proposed product policy, with the chargeable work explained before the call.
Treg documents fixed maker fees for recipes and capped script-declared charges for scripts, with upstream metered fees charged to the caller on top. Failed Hub runs charge nothing. Maker earnings land as balance credits, and listing approval also applies to later listed versions and price changes. Those platform rules are distinct from the decision about what our assessment promises. Treg CLI reference
The following batch uses US-dollar assumptions for an account-only tool. It includes no mailbox finding or verification:
| Batch outcome | Assumed charge | Total |
|---|---|---|
| 400 complete qualifications or exclusions | $0.12 provider cost + $0.40 maker fee | $208.00 |
| 60 completed assessments ending in review | $0.12 provider cost + $0.20 maker fee | $19.20 |
| 40 technical failures | No charge in this model | $0.00 |
| 500 requested accounts | Blended tool charge | $227.20 |
Suppose the 400 complete decisions comprise 250 qualifications and 150 exclusions. If 200 qualified accounts survive the buyer's review, the batch charge is $1.136 per accepted account: $227.20 divided by 200. The buyer still needs to budget for its review time and any later contact work.
This matters when comparing tools. A fee per request is a convenient invoice unit, but accepted decisions are closer to the buyer's useful output. The 150 exclusions also contributed value if they prevented unsuitable accounts from consuming the rest of the campaign budget.
The maker side is less comfortable. In this scenario, 400 full fees and 60 half fees produce $172 in balance credits. Three support hours at $60 cost $180 in cash. Even treating each credit dollar as fully useful for future platform spending leaves an $8 gap before other maintenance costs; it does not create cash available to pay those costs.
Cash payout terms therefore need confirmation before this becomes a revenue plan. So do listing access, the platform's actual charging behaviour for partial outcomes, upstream data rights and the support burden. A failed Hub run's charging rule should not be stretched into a promise about refunding separately incurred provider bills.
For us, this is the commercial test of the expert-tool concept. The judgement needs to be reusable enough that another buyer's call adds relatively little work. A product whose fee is consumed by explaining every answer needs a better contract, a different price or a narrower market.
Keeping the judgement useful
The first policy is deliberately strict about additional capacity. That makes its answers easier to explain, but it can miss companies expanding without a direct statement. Whether that trade-off is commercially acceptable depends on the buyer's offer and the cost of a mistaken inclusion.
A small evaluation set should therefore contain ordinary vacancies, replacement hires, evergreen pages, ambiguous subsidiaries, wrong geographies and conflicting evidence alongside straightforward qualifications. An experienced operator can label those cases against the written policy, then inspect disagreements by cause. Treg's documented publish checks can help validate a Hub artefact, but a handful of execution cases does not establish business accuracy.
The useful measures are qualification errors, unresolved reviews, evidence retention and the time a buyer spends resolving exceptions. Those results would tell us whether to widen the policy or improve the research. They are evaluation targets for this design, rather than measured performance claims.
Versioning makes those changes manageable. A policy revision changes what qualifies. An implementation revision can fix extraction while preserving the rule. A schema revision changes the caller's interface. The response needs to identify each so that a buyer can explain why last month's account outcome differs from today's.
Freshness also limits reuse. An assessment tied to a 7 October evaluation date cannot stay current indefinitely just because its source URL still opens. A later call needs the appropriate evidence checks again, with its new evaluation date and costs visible.
Support questions then become product evidence. If buyers repeatedly dispute UK service coverage, that condition needs a clearer definition or a better source. If they repeatedly mistake qualification for purchase intent, the promise and output need correction. Maintenance includes those decisions as well as keeping API adapters working.
How the tool earns another call
Distribution starts with a recognisable job. “Assess SaaS implementation expansion serving UK customers” gives an agent and its operator something concrete to request. “Expert GTM intelligence” makes them work out what the product actually does.
A useful listing would show a supported account, an exclusion and an unresolved review, with the relevant evidence beside each. It would state the input requirements, policy windows, fee ceiling and output boundary. A buyer can then decide whether the judgement fits its campaign before committing a large batch.
The open-source GTM tools guide covers another distribution route: exposing useful building blocks where technical teams can inspect and adopt them. A demonstration of the decision contract can create interest even while access to a marketplace listing is being resolved.
Our strongest reason to build this product would be that the same judgement already proves useful in repeated work. Packaging it forces the definition to become clearer, and Treg gives that definition a route into an agent's workflow. The commercial promise depends on the resulting decisions being good enough that buyers come back with another account list.
FAQ
Can an existing assessment API be connected to Treg?
Treg documents registered HTTP tools and server-side credential handling. That provides a tool-access route. Hub recipes and scripts follow a separate, feature-gated publication model with listing approval, so registration should not be treated as a marketplace listing.
What belongs in the expert tool instead of the caller's prompt?
The reusable policy, source interpretation and response contract belong together. The caller supplies campaign context and acts on the answer. If every caller has to redefine what expansion means and resolve the same evidence conflicts, much of the expert value is still outside the product.
Should an uncertain account incur a fee?
A completed assessment can identify a useful unresolved condition. Its fee should reflect the work promised and delivered, with that treatment disclosed before the call. Technical failure and a completed review are different outcomes; this design charges half the maker fee for the latter.
Are maker balance credits equivalent to cash revenue?
The CLI reference describes credits to the maker's balance. Their usefulness depends on future platform spending. Cash withdrawal or a separate partner-payment arrangement needs confirmed terms before it belongs in a cash-revenue forecast.
What would justify widening the qualification policy?
Repeated missed accounts, buyer feedback and labelled examples can show that the current expansion rule is too restrictive. A broader evidence tier should have its own definition and evaluation, then ship as a visible policy change so callers can choose it deliberately.





