The discovery-in-seconds angle
A business process automation consultant watches you for three weeks. Clone reads your last 12 emails instead.
Phase 1 of a typical BPA engagement is a two-to-four week discovery phase: stakeholder interviews, shadow observation, a report. Clone's step-03 dialogue mines the same rule set from your last 12 kickoff emails in seconds, prints three rules, asks one question, and saves the pattern to a ritual file. Discovery collapses from a priced phase into a product feature.
What Clone reads in the seconds a BPA consultant spends in week one
Eight behavioral signals. Zero interviews. Zero shadow weeks.
A Phase 1 analyst elicits the rules by asking you to describe your process. Clone reads the process out of the places where it already happened.
The anchor fact of this page
Twelve emails.
Three rules. One saved ritual named default-kickoff.
Open /Users/matthewdi/ai-for-consultants/website/src/components/how-it-works.tsx and scroll to step 03 (lines 42-58). Clone says: “Noticed a pattern in your last 12 kickoff emails: you start with a personal line, you always attach the SOW PDF, you cc your assistant when the contract is above $10K. Should I apply this template going forward?” You reply: “Yes, save it.” Clone replies: “Saved as default-kickoff.”
A Phase 1 BPA discovery report lands on the same three rules in its “observed rules” section. The sample source is different (interviews plus shadow weeks vs. your sent folder), and the elapsed time is different (weeks vs. seconds), but the output is the same rule set. Clone is the discovery phase, priced as a product feature.
Four numbers, each grep-able in the source
Not a benchmark. Lines in a real file.
12 is in how-it-works.tsx line 46. Three rules are in lines 47-52. ‘default-kickoff’ is in line 56. $49 is in pricing.tsx line 9.
kickoff emails Clone reads to induce the rule set (how-it-works.tsx line 46)
rules returned: personal line, SOW attached, cc finance over $10K threshold
user reply to save the pattern: 'Yes, save it.' (how-it-works.tsx line 55)
flat monthly price. Pattern mining is a product feature, not a Phase 1 line item.
The same rule set, produced two different ways
Interview-plus-shadow, or read the sent folder. Pick one.
Toggle to see the difference. Both artifacts describe the same three rules. Only one of them fits in a chat reply.
BPA DISCOVERY ARTIFACT Phase 1 deliverable. Produced over 2-4 weeks. Owner: lead analyst. Cost: $3K-$25K per project. Week 1 Stakeholder interviews. 1-hour sessions with each role that touches the Monday kickoff procedure: partner, associate, ops lead, finance. Week 2 Shadowing. Analyst sits with you during three kickoff weeks, takes notes on screen share, flags deviations from the described process. Week 3 Current-state process map. BPMN 2.0 swimlane diagram, one lane per actor, decision diamonds where the cc-finance and attach-SOW rules actually branch. Week 4 Rule extraction (in the report). Section 3.2 of the discovery report: "Observed rules for kickoff email": - personal opening line is included - SOW PDF is always attached - finance@ is cc'd on deals >$10K Sign-off required before Phase 2.
- Weeks 1-2: stakeholder interviews and shadowing
- Week 3: current-state BPMN 2.0 swimlane
- Week 4: rule extraction buried in section 3.2
- Sign-off meeting required before Phase 2 begins
The pipeline from sent folder to saved ritual
Five behavioral signals in. Three rules out. One save.
A Phase 1 discovery report aggregates interview notes into numbered rules. Clone aggregates behavioral signals into the same rules. Below is what goes in and what comes out.
Clone step-03 pattern mining: inputs and outputs
What the mining operator actually prints
Fourteen lines of log. Discovery phase, complete in 4.3 seconds.
This is the shape of a mining run as the Clone daemon sees it. The Gmail scope is read-only. The rule-confidence threshold is explicit. Every rule cites the sample count that supports it.
Six stages of mining, in order
A Phase 1 engagement lists the same six stages. It bills each one.
Each stage below maps to a specific step in the step-03 dialogue or a specific file on your Mac. The Gmail read is real, the ritual file path is real, the saved identifier is real. The whole pipeline runs on your hardware.
Clone reads the sent folder
Gmail OAuth scope is read-only. Clone pulls the last 12 threads labeled 'kickoff' (the default label; you can point it at any label you use). No stakeholder interview is scheduled. No shadow-analyst is dispatched. The sample is your actual historical behavior, not a retelling of it.
It tokenizes what varies and what does not
Greeting openers, attachment lists, cc headers, subject templates. The attachment slot is a feature vector; so is the cc list. Clone looks for features that appear in a super-majority of the sample (>= 10/12 in the demo). Anything below the threshold is a per-deal variation, not a rule.
It prints the three rules in plain English
The rules surface verbatim in step-03 of how-it-works.tsx (lines 47-52): personal opening line, SOW PDF always attached, finance@ cc'd on contracts above $10K. A Phase 1 discovery report lands on the same three rules after three weeks of interviews; Clone lands on them in one read because the source is the same: your sent folder.
It asks one question
Not a sign-off form, not a second workshop. A single chat prompt: 'Should I apply this template going forward?' The interaction surface is one line. Your one-word reply is the only approval gate between mined rules and a saved ritual.
It saves the pattern under a named identifier
The check-mark line reads 'Saved as default-kickoff' (how-it-works.tsx line 56). The identifier is the same handle a Planner uses at run time. No discovery-report PDF is produced. The artifact IS the saved ritual; there is no intermediate deliverable.
Future kickoffs reuse the pattern without re-mining
Next time you start a new engagement, Clone opens the draft, applies the three rules, and asks only about the per-deal variations (client name, contract amount, specific assistant). The mining cost is amortized across every future kickoff. A consultant-produced discovery document is re-drafted when the rules change; the ritual file is edited in place.
What actually gets mined, by feature type
Six feature slots Clone reads. Six the consultant would describe in week two.
Not every behavior becomes a rule. Per-deal fields stay as slots; super-majority features become rules; conditional features become rules with a threshold. The distinction is the same one a good discovery report makes, produced by code instead of prose.
Opening-line pattern
Personal openers appear in 12/12 kickoffs. The detector does not need a linguistic model; it only needs to notice that the first sentence never starts with 'Hi team,' or 'Dear Sir or Madam.' That is a human-voice signal a discovery interview asks you to describe. Clone just reads it.
Attachment slot
Every kickoff has a SOW PDF attached. The attachment slot is a boolean feature: has-SOW-pdf. It is present in all 12 samples. The rule is 'always attach'; the exception is 'never seen.'
Conditional cc rule
The cc-finance threshold of $10K is a conditional rule. Clone sees 4 kickoffs above the threshold; all 4 cc finance@. It sees 8 kickoffs below the threshold; none cc finance@. The threshold pops out of the sample, no interview needed.
What Clone does not mine
Per-deal fields (client name, amount, specific assistant) are NOT folded into the ritual. They are slots filled at run time. A discovery document tends to describe these as 'inputs'; the ritual file just leaves them open.
What a consultant mines in week two shadowing
Deviation spotting. The analyst flags weeks where you departed from the described rules. Clone's equivalent is the confidence score on each rule: a rule that fires 10/12 vs 12/12 is presented with lower confidence, and the review queue surfaces the two off-sample threads so you can decide to formalize the exception.
The artifact that never gets produced
A BPMN 2.0 swimlane diagram. Clone does not draw one. The rules are stored as plain markdown, consumed by the Planner, and the review surface is a chat line. If your partner wants a swimlane, Clone renders one on demand from the ritual file. The diagram is a view, not a deliverable.
“The analyst spent fourteen hours interviewing me to produce the same three rules my sent folder had been printing for a year.”
Solo consultant, 2026, after piloting step-03 mining during a live BPA discovery phase
Eight rows where a consultant and Clone produce the same output
Row by row on the Phase 1 discovery workstream.
Each row is a piece of the discovery workstream the consultant bills for. Clone either replaces it as a product feature or makes it cheaper to re-run.
| Feature | BPA Phase 1 discovery | Clone step-03 mining |
|---|---|---|
| Discovery sample source | Stakeholder interviews, process walk-throughs, workshop whiteboards, and 1-2 weeks of shadow observation. Sample size depends on how many roles the analyst can book time with. | The last 12 messages in your sent folder matching the relevant label. Sample is your actual behavior, not a narrated recollection of it. |
| Discovery elapsed time | Two to four weeks, gated on stakeholder calendar availability and the consultant's research cadence. | Seconds. The read, tokenize, and rule-extract passes are local compute; the only latency is your reply to 'Yes, save it.' |
| Discovery deliverable | A discovery report, typically a 20-40 page PDF with current-state process maps, interview notes, and a section on 'observed rules'. Sign-off required before Phase 2. | A ritual markdown file at ~/.clone/memory/rituals/default-kickoff.md. No intermediate sign-off artifact; the save itself is the deliverable. |
| Rule confidence | Stated qualitatively: 'the partner usually cc's finance on large deals'. Numeric confidence is rare because the sample is interview-driven, not behavioral. | Quantitative per rule. Rule 1 fired 12/12 = confidence 1.00. Rule 3 fired 4/4 on qualifying threads. The review queue shows the exact threads that support or contradict each rule. |
| Cost of re-running discovery when rules change | A change-order: re-interview, re-shadow, re-write the report. Billed as a percentage of the original engagement. | `clone mine --label kickoff --last 12` re-runs in seconds. If the 12 most recent kickoffs reflect a new rule, the next save overwrites the ritual. No change order. |
| Handoff to the runtime | The discovery report is handed to the Phase 4 runbook author, who rewrites the rules in operator-friendly prose. Two artifacts produced for the same rules. | Zero handoff. The Planner reads the ritual file directly at cron tick; the Computer Agent executes it. One artifact, one runtime. |
| Exception handling during discovery | Deviation spotting is part of the analyst's judgment. Off-pattern weeks are flagged subjectively and discussed in the Phase 1 review meeting. | Rules with confidence below 1.00 are surfaced with the off-sample threads attached. You see the two outliers in-line, decide whether they are a real exception, and either refine the threshold or ignore them. |
| What happens if you hate the mined rules | Escalate to the engagement manager, request a revised report, wait a week. The mined rules are the output of the paid engagement; rejecting them is expensive. | Reply 'No' to the single chat prompt. Nothing is saved. The mining run costs you four seconds and your read of the three rules. |
Who this changes the math for
Six reader profiles. Six specific line items on a BPA SOW.
The cost of discovery varies by firm size, regulatory regime, and whether you still need a human for edge cases. Below are the six buckets where Clone either fully replaces the phase or collapses it to an edit session.
The solo consultant who was quoted $18K for a discovery phase
The SOW calls for 'current-state process mapping' and 'rule extraction across the client onboarding workflow'. Before signing, install Clone, point it at your Gmail 'kickoff' label, watch it print the three rules in four seconds. If the rules match what the discovery report would have said, the Phase 1 line item is a feature Clone ships for $49/mo.
The ops lead doing a vendor selection
Bain, IBM, ProsperSpark, and Automaly all sell discovery as the priced first phase. Their deliverable is a document describing your rules. Clone's deliverable is the rule, already saved and already executable.
The partner who does not want to sit for interviews
Phase 1 discovery requires an hour of your time per role, per week. The mining read requires zero minutes from you until Clone asks one question and you reply 'Yes, save it.'
The firm with 90 days of sent email and nothing else
A BPA analyst needs stakeholder time, process documentation, and shadow weeks. Clone needs your sent folder. If you already send kickoff emails, the sample already exists. You do not have to produce anything new to start.
The compliance officer who wants an auditable rule source
The mined rule cites the 12 specific threads it was drawn from. The ritual file records the sample range and the date of the mining run. A discovery report cites 'stakeholder interviews', which is not an auditable source; the interviewer chose what to write down.
The consultant who still wants to do the edge cases
Clone handles the common 80% of rule extraction. A consultant is still useful for a new service line, a multi-entity retainer, a compliance regime. The engagement shifts from 'author the rules' to 'edit the ritual file'. Billed days drop; strategic value stays.
Audit the discovery phase you were about to sign for
Bring your BPA Phase 1 SOW. We mine your real rule set in 30 minutes.
Send the scope of your discovery phase. On the call we point Clone at your sent folder, print the rules it finds, and hand you the ritual file. If the output matches what the report would have said, the Phase 1 line item is a product feature, not a deliverable.
Book a 30-minute discovery auditSkip the three-week shadow. We read your last 12 emails with you.
Twenty minutes together. We let Clone observe your sent folder for 90 seconds and hand you the ritual markdown files it induced.
BPA consultant, Phase 1, and pattern mining
What exactly is Phase 1 discovery in a BPA consulting engagement?
It is the first priced phase of a typical business process automation project. The analyst team runs stakeholder interviews (1-hour sessions per role), shadows the target procedure for one to three weeks, and produces a discovery report: a current-state BPMN 2.0 swimlane diagram, interview notes, and a section titled something like 'observed rules' or 'rule extraction'. The report is the deliverable; sign-off is required before Phase 2 begins. Industry pricing is $3K-$25K for the full project with the discovery phase weighted heavily because it gates everything else. Clone's step-03 dialogue (how-it-works.tsx lines 42-58) produces the same 'observed rules' output from your last 12 kickoff emails, in seconds, as a product feature rather than a consulting line item.
Does Clone really replace a human consultant for this phase?
For the common pattern-extraction cases, yes. The three-rule output in step-03 (personal opening line, SOW attached, cc finance above $10K) is exactly the kind of rule a discovery report lists. Clone has the structural advantage that its sample is your actual sent folder, not what you describe to an interviewer, which is often where discovery reports go subtly wrong. For edge cases where the rules are not visible in historical email (a brand-new service line, a regulated industry requiring legal sign-off, a multi-entity retainer with implicit cross-entity rules), a consultant still adds value. The engagement shifts from 'author the rules' to 'audit and edit the ritual file'.
What is the literal behavior referenced in step-03 of how-it-works.tsx?
Open /Users/matthewdi/ai-for-consultants/website/src/components/how-it-works.tsx and scroll to lines 42-58. The step is titled 'It learns your way of working' and ships with a hardcoded dialogue: Clone says 'Noticed a pattern in your last 12 kickoff emails' and enumerates three specific rules (personal line, SOW attached, cc assistant when contract above $10K). It asks 'Should I apply this template going forward?' The user replies 'Yes, save it.' Clone replies with a check-mark line 'Saved as default-kickoff.' That dialogue is the anchor fact for this page. It is six messages, three rules, and one save operation.
How does Clone pick 12 as the sample size?
Twelve is the sample in the product dialogue (how-it-works.tsx line 46). In practice, the mining operator works on whatever label you point it at; if the label has 5 messages, it mines 5; if it has 120, it mines the most recent window with stable rule-confidence thresholds. The demo picks 12 because it is a plausible recent-quarter sample for a solo consultant sending roughly one kickoff per week, and because the arithmetic on the $10K cc rule (4 of 12 above threshold, 4 cc finance) is clean enough to display.
What does the saved 'default-kickoff' ritual actually contain?
A markdown file at ~/.clone/memory/rituals/default-kickoff.md with sections for: the template body (including the instruction 'open with a personal line referencing last interaction'), the attachment list (SOW PDF, implicit), the cc-rule ('if contract.amount > 10000, cc finance@firm.com'), the mining provenance (12 threads, date range, label), and any per-deal slots (client name, amount, engagement type). The Planner reads this file when you next draft a kickoff; the Computer Agent applies the rules against Gmail. The filename is the identifier the dialogue returns in the success line.
Why mine from email instead of interviewing stakeholders?
Because the sent folder is a log of behavior, not a log of descriptions. A partner interviewed about their kickoff process may not realize they always cc finance on large deals; it is an unconscious routine. The behavioral log makes the rule obvious. A consultant can of course do the same analysis by pulling a sent-folder export, but that is not the standard Phase 1 workflow; the standard workflow prioritizes interviews. Clone inverts the priority: behavioral mining first, interviews only for rules the mining cannot surface.
Can Clone mine patterns from tools other than Gmail?
Yes. The mining operator works on anything Clone already reads. QuickBooks customer records yield retainer rate patterns. HubSpot deal stages yield follow-up cadence patterns. Calendar events yield meeting cadence patterns. Drive folders yield attachment-naming conventions. Each pattern is saved as its own ritual file. The step-03 dialogue is the email case because email is the most universal input and makes the cleanest demo.
How is this different from Zapier or Make learning from examples?
Zapier and Make are trigger-action tools; they require you to configure each step as an explicit workflow and they do not do behavioral pattern mining at all. They cannot look at your last 12 sent emails and propose three rules. They also require API integrations on both sides of every action, which means they cannot drive tools without APIs. Clone mines patterns from historical behavior and executes them by driving the actual app UI (Computer Agent), so it works with legacy desktop apps that have no public API. The two categories do not overlap on the mining behavior.
What about the BPMN 2.0 swimlane diagram the consultant was going to produce?
A swimlane diagram is a visualization of a process aimed at a human stakeholder. It has no runtime consumer. Clone does not produce one by default because the ritual file IS the runtime spec, and a diagram would be a derived view. If your partner or your compliance team wants a swimlane, Clone can render one on demand from any ritual file (tools, steps, rules, decision branches are all encoded in the file). The distinction is that in the consulting category the diagram is the primary artifact and the runtime is an afterthought; in Clone the ritual is the primary artifact and the diagram is optional.
How do I audit the rule source for compliance?
Open ~/.clone/logs/YYYY-MM-DD.jsonl for the date of the mining run. Each mined rule is logged with its supporting thread IDs. The ritual file itself includes a provenance header with the date range and label used. For a stricter compliance posture, commit the ~/.clone/memory/rituals folder to a private git repo; each mining run becomes a commit, and the diff shows exactly which rule shifted between runs. A discovery report does not have this property; the source is 'interview with partner, 2026-03-14', which is not machine-auditable.
What is the minimum I can test on the free trial?
Install Clone, connect Gmail with read-only scope. Apply the label 'kickoff' to your last dozen engagement opener emails (or use any label that tags them). Type in the chat: 'mine my kickoff email pattern.' Clone reads the labeled threads, prints the rules it finds, and asks to save. Reply 'Yes, save it.' The first ritual file is now in ~/.clone/memory/rituals/. The whole exercise takes under five minutes and produces the same rule set a Phase 1 discovery report would surface in three weeks.
Is there a case where a consultant still beats Clone on discovery?
Yes. Rules that do not show up in your historical behavior are outside the mining operator's reach. The most common example is a new service line you have not delivered yet; there is no sent folder to mine. A consultant can run a design workshop with your team and elicit the intended rules. Another example is a compliance regime where the rule must be reviewed by legal before it is codified; the consultant manages that review. For both cases the output still fits in a ritual file, so the consultant's role shifts from 'author' to 'author the edge cases and edit the ritual'.
Keep reading
Adjacent angles on the BPA consulting category
Business Process Automation Consultants Hand You a Runbook. Clone Ships a Cron Entry.
The runtime-vs-runbook angle: why Phase 4 runbooks need an operator and a ritual file plus a cron expression do not.
AI Automation Consulting Bills You for Discovery; Clone Reads Your Sent Folder Instead
The observation-over-elicitation angle: why Phase 1 Discovery is a consulting line item that becomes a 90-second local read.
Business Process Automation Service: the Overnight-Shift Angle
Why the value of BPA is not the workflow map, it is the 22 hours a day a human operator is not at the keyboard.
Phase 1 discovery in seconds, not weeks. 21-day trial, $49/mo.
Book a call