An assessment of the inbound lead workflow against six design criteria, the instrumentation the growth model requires, and the sequence in which it can be built.
The current inbound workflow is not broken. It does what it was designed to do: validate inbound enquiries and distribute them quickly across direct sales, resellers and a third exit.
The 10x growth target now requires it to do more. The system has to recognise the account behind each enquiry, estimate that account’s expansion potential, select the go-to-market motion that fits it, and learn which routing decisions produce activated and expanding units. The migration to Salesforce creates the opportunity to establish that broader foundation before the current process is transferred into a new system.
The principal constraint is not the automation within the current workflow. It is that the workflow operates on contacts and initial demand, while the growth model operates on accounts and their expansion over time.
If that thesis holds, improving the routing engine alone makes the current process faster without necessarily improving account selection, activation or expansion. The data and measurement model would then have to be established before the workflow is redesigned.
We reconstructed the inbound process from the workflow shared on 28 August and placed it within the full customer journey, from awareness through activation and expansion. We then:
This is therefore not an assessment of whether the current workflow was designed well. It is an assessment of whether its foundations are sufficient for the growth system now required.
The workflow already performs enrichment and validation ahead of human involvement. The opportunity is to build on that foundation in six areas. Sections 3 and 4 set them out in full; Section 6 places them in order.
These are not six isolated process changes. Together they establish a system in which every account can be located, measured, assigned and improved across the complete customer journey.
This section sets out the reference model used in the remainder of the document. A go-to-market operating model comprises multiple workflows. Each spans a range of stages in the customer journey; adjacent workflows overlap; and collectively they cover the journey from first awareness through to expansion, with an advocacy-referral loop returning named accounts to WF2.
The bowtie below represents the customer journey in seven stages. The five bands beneath it represent the workflows operating across those stages. The overlap between adjacent bands is the interval in which responsibility for an account transfers from one workflow to the next.
Table 1. The five workflows, their function and output.
| Workflow | Output | |
|---|---|---|
| WF1 | Demand DevelopmentTargets accounts against the addressable market and assigns them to named segments. Nurtures accounts not currently in market. A monitoring agent detects propensity change and initiates outbound on signal. | An account with recorded intent. |
| WF2 | Lead DistributionResolves the contact to an account, then assigns it to a go-to-market motion and an owner. | An account assigned to a motion, with an owner and a response time. |
| WF3 | Deal ExecutionRuns the assigned motion. Allowable cost of sale is set against 24-month value rather than initial order size. | A mutual commitment: order placed, provisioning committed. |
| WF4 | Onboarding & ActivationShips, installs and connects to first successful use. Establishes the 12-month expansion plan. | Units in service and an expansion plan on record. |
| WF5 | ExpansionAdds units, sites and use cases from activation onward, measured against the 24-month plan. | Connected units, and referrals returning to WF2. |
The workflow shared on 28 August corresponds to WF2, Lead Distribution. Section 2 assesses it within that scope.
The remaining four workflows — WF1, WF3, WF4 and WF5 — fall outside the scope of that map and are addressed here only at the points where Lead Distribution interfaces with them. Specification of those four workflows is a separate exercise, and the 24-month target depends on all five.
This section is a record of the workflow as it currently operates, transcribed from the process map of 28 August. It states what the workflow does; it does not assess it. Sections 3 and 4 do that.
The workflow executes approximately fifteen automated checks — enrichment, region availability, scalability tier, product and vertical overrides, spam filtering and prior-account matching — prior to human involvement. Elapsed time from submission to Agent Qualified is under one hour. Classification is performed on a composite scalability tier derived from kit count, location count and growth evidence rather than on order size alone. Three destinations are defined, each with an assigned owner.
Six requirements of the target design, against which the workflow is assessed in Section 4. Each is stated in terms a design either meets or does not.
The properties the design depends on — vertical, expansion value, elapsed time since activation — belong to an account rather than to a person. The workflow resolves each record to an account before any routing decision, including subsidiaries, prior enquiries, reseller-managed accounts and installed units. An existing customer requesting an additional site is then handled as expansion, not as new demand.
The addressable market is 4,130 named companies across thirteen verticals. Assignment proceeds in two steps: whether a record maps to one of these accounts, and if so which vertical, which determines the receiving specialist. Vertical carries the buyer’s language, deployment pattern and decision cycle, and is therefore the smallest unit at which a motion can be designed. A record outside the list that demonstrates comparable scale enters through a defined exception and is added to it, so that the list operates as a prior rather than as a boundary.
Table 2. Addressable market by vertical, ordered by value per company.
| Vertical | Companies | Units per company | ACV per unit | Value per company | TAM |
|---|---|---|---|---|---|
| Telecom Backhaul | 80 | 37,500 | $6,000 | $123.8M | $9.90bn |
| Agriculture | 150 | 13,333 | $12,000 | $88.0M | $13.20bn |
| Retail & Hospitality | 100 | 27,000 | $5,000 | $74.2M | $7.40bn |
| Transport & Logistics | 300 | 16,667 | $6,000 | $55.0M | $16.50bn |
| Maritime | 100 | 1,100 | $36,000 | $31.7M | $3.20bn |
| Construction | 600 | 8,333 | $6,000 | $27.5M | $16.50bn |
| Energy Oil & Gas | 400 | 1,250 | $24,000 | $24.0M | $9.60bn |
| Multi-Site Enterprise | 1,000 | 5,000 | $6,000 | $16.5M | $16.50bn |
| Healthcare | 300 | 667 | $15,000 | $5.5M | $1.65bn |
| Education | 200 | 100 | $60,000 | $3.3M | $0.66bn |
| Energy Ocean Rigs | 400 | 25 | $120,000 | $2.4M | $0.96bn |
| Finance | 400 | 110 | $15,000 | $0.9M | $0.36bn |
| Media / Broadcast | 100 | 100 | $6,000 | $0.3M | $0.03bn |
| Total | 4,130 | — | — | — | $96.5bn |
Value per company spans $0.3M to $123.8M, a 375× range; ACV per unit spans 24×. The two highlighted rows invert: the vertical with the highest ACV per unit ranks near the bottom on value per company, and one of the lowest ranks first. Assignment on ACV alone reverses the result. Verticals hold between 80 and 1,000 companies each, so no target market is a long-tail market and all 4,130 accounts can be enumerated in advance.
Three values are distinguished and not interchanged. The initial order is what is requested now. The expected 24-month value is what the account is forecast to produce — for a two-kit order at plan, approximately $41,000 rather than $12,000, the same 3.4× multiple that applies to a $180,000 order. The eligible value is the deployment ceiling of the account’s estate, which is a bound and not a forecast.
The design routes on expected value relative to cost to serve, informed by the ceiling but never substituting it. The multiple is derived from the enterprise market model and is restated once account-level unit data is available; cost to serve is established per motion. On that ratio a small initial order with substantial expected value can justify human coverage, and a large one without it cannot.
Time is the variable with compounding effect: a shorter cycle produces more cycles within a period. The design sets a response time per vertical against the observed decision cycle of that buyer, rather than applying a uniform service level to every record.
Table 3. Effect of sustained velocity improvement on the run rate after eight quarters.
| Velocity improvement | Per quarter | Run rate entering year three |
|---|---|---|
| Modest | +5% | ×1.48 |
| Sustained | +10% | ×2.14 |
| Aggressive | +15% | ×3.06 |
Conversion compounds across stages; velocity compounds across periods. A ten percent per-quarter improvement approximately doubles the run rate by the end of eight quarters at constant headcount. The figures are end-of-period rates rather than cumulative output. Within WF2, elapsed time is the clearest constraint on that rate.
A record leaving the workflow is assigned one of a defined set of dispositions — direct, channel, self-serve, nurture or out of scope — with a recorded reason and an owner. Outcomes requiring different subsequent handling are not aggregated into a single path, since the proportion of that volume representing addressable demand then cannot be determined.
A routing decision is a prediction of an account’s 24-month value. The design records the prediction and compares it against the realised outcome. The first testable checkpoint is expansion at 90 days: reaching plan requires approximately 5.3% additional units per month, so an account with no expansion at that point stands some 24 units below plan, and the rate required across the remaining 21 months rises to 6.1%. Absent this comparison, classification accuracy does not improve with volume.
The workflow as built, tested against each criterion in turn. One observation per criterion; the reference is given at the right of each heading.
A form submission is processed as a contact record. The quantity the model requires to increase is units per account. The prior-account check exists within the workflow, positioned tenth of eleven steps, inside the routing script.
The consequence is that an existing customer submitting a request for an additional site is classified as a net-new lead and routed into acquisition, and does not enter the expansion workflow. Resolving the contact to an account — including subsidiaries, prior enquiries, reseller-managed accounts and installed units — is a reordering of steps already present in the workflow.
A Vertical Override sits alongside a Product Override on the enrichment agent, as an exception applied to a decision otherwise made on kit count, region availability and validity. The routing record itself (Annex B) carries service fit and business size; it carries neither a vertical nor a match against the named-account list.
Records therefore reach a general representative pool rather than the specialist for that vertical, and the 4,130 named companies are not tested at the point of entry — although, being enumerable in advance, they can be resolved by list lookup rather than by inference. The vertical field is already carried on the enrichment agent, so the change is one of position in the sequence rather than of new capability.
The composite tier scores kits, locations and growth evidence at the point of enquiry — measures of what is being requested now. Requested scale and account potential are weakly correlated, so the two produce different assignments:
Table 4. Requested scale against the eligible ceiling of the account.
| Account | Kits requested | Eligible units behind it | Eligible value (ceiling) |
|---|---|---|---|
| Construction group | 2 | ~8,300 | ~$27.5M |
| Regional retailer | 15 | ~27,000 | ~$74.2M |
| Single-site finance office | 12 | ~110 | ~$0.9M |
| Small advisory firm | 2 | 2 | ~$12K |
The final column is a ceiling, not a forecast. A two-kit request may sit in front of approximately $27.5M in eligible deployment value and approximately $41,000 of expected 24-month value; a twelve-kit request may sit in front of under $1M on either measure. Neither expected value nor cost to serve is computed at the point of decision, so the ratio the criterion routes on is not available to the routing agent.
A record clears fifteen checks and reaches Agent Qualified in under one hour, then waits approximately one day before contact with a representative. Representatives are notified immediately. The interval is therefore attributable to the requirement for manual review on every direct record, rather than to processing or to notification.
The interval is uniform across records: the same review is applied whether the buyer decides in a week or in two quarters. Further optimisation of the qualification engine does not reduce it, as the constraint is downstream of that engine.
Eight outcomes are folded into one exit on the map: invalid, no regional coverage, no current need, low potential, self-serve fit, reseller fit, competitor and duplicate. Each requires different subsequent handling, and none is separated by a recorded reason.
What is established is the volume below the ten-kit line. What is not established is where it goes: the proportion routed to a reseller, to self-serve, to nurture or out of scope cannot be determined from the material provided, and no reactivation signal can be specified for a population that is not classified. The uncertainty is itself the argument for reason codes.
Two related points. The remaining 37% is approximately 28 to 37 records per seller per week across the current 40 sellers. A checkout limited to one kit per transaction with payment in advance does not yet constitute a self-serve destination, so one of the eight outcomes has no viable receiver. A correction for the threshold analysis: 63% below ten kits implies 37% above it, not 47%.
No line on the map runs backwards from an outcome to a decision. Won and lost deals send nothing upstream, nurture branches are terminal boxes, and once a record passes to a reseller through Euler, stage, conversion and timing data do not return.
A completed handoff carries acceptance, first contact, progression, outcome and reason code, with automatic return when an SLA expires. The current handoff carries none of these, so the volume routed to the reseller cannot be compared against the volume retained.
Four stage definitions on the current map are also to be resolved; they are set out in Annex C.
The migration to Salesforce sets the instrumentation for the period that follows. Fields not specified while the data model is being standardised are added afterwards, against live records and at higher cost. Instrumentation resolves into three layers, each a precondition for the one above it.
The seven stages of the bowtie, carrying an entity and a segment. A record is an account, at a stage, within a vertical. All three dimensions are required: without the entity the unit of work is a person rather than an account; without the segment, conversion can be computed in aggregate but not by vertical, which is the comparison on which assignment depends.
Entry and exit criteria per stage belong to this layer rather than to documentation about it. A definition that does not hold propagates into every measure built on it, so the four definitions listed in Annex C are resolved at this layer and not downstream. The model is also what allows a position to be determined rather than inferred, by a representative or by an agent; inference applied at volume produces systematic error.
Volume, conversion, velocity and efficiency, applied at every stage and to every motion including the channel (reseller) path. Velocity is elapsed time and efficiency is cost to serve; the two are stated in those terms because they are the variables the design acts on. None of the four is instrumented on the workflow as mapped, so the table below states what is required rather than what is present.
Table 5. Metric families required at each stage.
| Family | Measures | Examples |
|---|---|---|
| Volume | count | Leads, matched accounts, accepted, qualified, commitments, activated accounts, connected units |
| Conversion | yield | Form-to-engagement, engagement-to-qualified, qualified-to-commit, commit-to-activation, activation-to-expansion |
| Velocity | time | Submission-to-route, route-to-acceptance, acceptance-to-first-contact, commit-to-activation, activation-to-additional-units |
| Efficiency | cost | Cost per qualified account, per committed account, per activated unit, per incremental connected unit |
A routing decision is recorded with its prediction and compared against the realised outcome. That closure is the difference between a measurement system and telemetry: it produces a labelled record of decision and result, which is the data on which any agent placed in the routing path improves. Absent it, the four families describe the system without providing a basis for changing it.
With the first two layers in place, workflows are modules over a shared model. Each can be specified, changed and retired independently, and the effect of a change is read on the same measures as every other workflow. Sequencing follows from this: a workflow built before the model beneath it encodes its own implicit model, and implicit models held by adjacent workflows do not agree.
Throughput of a system is set at its constraint, so a workflow improved away from the constraint adds cost without adding output. The layered arrangement makes the constraint locatable, which is what allows optimisation to be directed at the system rather than at a component of it.
The six observations in Section 4 originate at the lower two layers: O1, O2 and O5 in the data model, O3, O4 and O6 in the measurement model. Each is ultimately executed as a change to the workflow, and none can be executed before the layer beneath it is specified. The workflow shared on 28 August is a reasonable construction on foundations that lay outside its original scope, which is why the sequence runs model first and workflow second.
With the three layers in place the business unit is managed as a production system: throughput and conversion against velocity and efficiency. Capacity, yield and cycle time become properties of the revenue system that are measured and planned against rather than inferred from pipeline — the terms in which this organisation already manages manufacturing.
Table 6. The three goals of a factory, applied to recurring revenue.
| Factory goal | In a revenue factory | What it requires |
|---|---|---|
| Increase production | Achieve growth | Processes that scale without proportional headcount — scalability |
| Improve efficiency | Lower cost to serve | Unit economics established per motion — sustainability |
| Enhance quality | Deliver the promised outcome | Outcome measured and returned to the decision — durability |
Each go-to-market motion is a production line carrying its own growth metric and cost structure. Direct, channel and self-serve are three lines rather than one path with three exits, and the four families are what give each line its own economics.
An autonomous system performs in proportion to how completely its environment is specified. Where position, segment and outcome are recorded, an agent acts, observes the result and adjusts, and each cycle raises the accuracy of the next. That is a growth loop: the compounding shown in Table 3, applied to the quality of the decision rather than to elapsed time. Where they are not recorded, automation accelerates the existing decision logic without improving it.
We know the system is working when, on the day it goes live, we can answer for every account and on every path: which stage it is in, which vertical it belongs to, how long it has been there, and what it cost to get there.
That is what we are building toward — a growth system that answers this on day one.
Ordered by layer, and within each layer by what is time-bound. Items 1 to 3 sit in the data model, 4 and 5 in the measurement model, 6 in the workflow. Only the first is bounded by a date set outside this work.
Items 2, 3 and 6 reposition and define components that are already present, and require no new engineering. Items 4 and 5 are specification work. Item 1 is governed by the migration schedule rather than by this analysis, which is what places it first.
Three questions that determine scope and cannot be resolved from the material provided.
Stages, owners, CRM states and the checks attached to each step, transcribed from the source map of 28 August.
Table A1. Stages, owners and CRM states, as documented on the source map.
| Stage | Owner | CRM state | Definition as documented |
|---|---|---|---|
| Education | — | — | Prospect encounters the Starlink Business site. |
| Lead | Enablement | — | Person who may or may not have intent to purchase. |
| Marketing Qualified Lead | Enablement | — | Person who has shown enough interest in Starlink Business to qualify for additional outreach. |
| Agent Qualified Lead | Enablement / Comms | Prospect · Target | Person who has passed inbound lead agent initial verification checks. |
| Sales Accepted Lead | Sales / Reseller Rep | Prospect · Qualify | Person manually vetted as a viable lead by a rep. |
| Sales Qualified Lead | Sales / Reseller Rep | Pursuing · Qualify | Person manually qualified by a rep. |
| Sales Opportunity | Sales / Reseller Rep | Win · Close | Person who has become our customer. |
Table A2. Checks and actions attached to each step.
| Step | Attached |
|---|---|
| Inbound lead form submission | Form Shorten Automation · Apollo Contact Enrichment |
| Clay enrichment agent | Routing Decision · Product Override · Vertical Override · Scalability check · Validate contact + company info · Region availability check · Spam check |
| Gertie script (development required) | Update dashboard · Write to dataHub · Previous Starlink account check |
| Sales rep notification | HubSpot email · Teams channel alert |
| Reach out to lead | Qualification |
| Deal proposing | Solutioning · Quoting / Negotiating |
Systems in the path: Apollo (contact enrichment) · Clay (enrichment and routing) · Gertie (routing script, in development) · HubSpot (email) · Teams (alerting) · Euler (reseller hand-off) · dataHub (write target).
Sample enrichment output dated 2026-08-19, as shown on the map. This is the record on which the three-way routing decision is made.
The classification is a composite tier — kits, locations and growth evidence considered together — rather than a threshold on kit count.
Table C1. Stage definitions to resolve before migration.
| On the map today | The issue | Suggested |
|---|---|---|
| Sales Opportunity — “person who has become our customer” | That describes a customer, not an opportunity | An opportunity is an account actively evaluating a purchase. Customer status begins at mutual commitment. |
| “Lead response” as a stage gate | A reply proves contactability, not commercial potential | Separate engagement from qualification and measure them independently. |
| “Interest in business?” | Undefined, binary, and applied before anything is known | Observable criteria: service availability, geography, use case, intended deployment, existing relationship, channel eligibility. |
| “Deal proposing” | Collapses the whole selection process into one box | Solution configuration, commercial validation, operational feasibility, stakeholder approval, mutual commitment — each measurable. |