Nothing gets missed
Every mandatory requirement, form, and deadline pulled from the full package — appendices and addenda included — and cited back to its page.
Automation · proposal operations · web experiences
Find the friction.Build the fix.
Object Method designs and deploys practical automation, proposal systems, and high-performing websites around the way your business actually works.
One workflow. A clear recommendation. No obligation.
Founding-client offer · five openings
One workflow. One honest answer. Free.
Bring one process that is repetitive, slow, or too dependent on one person. Object Method will map what happens today, identify where automation could help, and tell you what should stay human.
Request the free audit No obligation. No confidential files required for the first conversation.Free automation audit01 workflow
- 01Workflow map
The real steps, handoffs, tools, and exceptions—not the idealized process.
- 02Opportunity ranking
Up to three improvements ranked by business value, effort, and operational risk.
- 03Recommended first move
A concise action map: automate, simplify, buy an existing tool, or leave it alone.
The real risk
A compliant bid loses to a checkbox.
Public procurement disqualifies on form, not merit. A missing insurance certificate, an unacknowledged addendum, a section over the page limit — and the technical proposal is never read. Your team spent sixty hours on it. Everyone remembers the one that got thrown out.
Nothing gets invented
The system flags what it cannot evidence rather than filling the gap. It never writes project experience or staff credentials it cannot source.
Nothing ships unreviewed
Every output goes to a named approver before it reaches SharePoint, Teams, or your CRM. The approval is recorded with the run.
How it is built
Three layers, and a hard line between them.
Most failed AI projects ask a language model to do work that should have been ordinary code, or hand it a decision that should have stayed with a person. Deciding where each line sits is the actual engineering.
01System layer
Deterministic software
The parts where a wrong answer is unacceptable and the rule is knowable.
- Handles
- File processing, schema validation, date logic, duplicate detection, permissions, routing, calculations, API writes, retry logic, audit logging.
- How
- Written as ordinary code with tests. No model is asked to count pages, parse a closing date, or decide who has write access.
- Proof
- It either passes validation or it stops and tells you why.
ParsingValidationRoutingAudit log
02System layer
Bounded AI judgment
The parts that need reading comprehension, constrained to a strict output schema.
- Handles
- Document classification, nuanced requirement extraction, categorisation, summarisation, comparison against approved content, drafting clarification questions.
- How
- Structured outputs only — never free-form prose into a deliverable. Every important field carries a source citation or it is marked unsupported.
- Proof
- Measured recall and precision against your own past submissions.
ExtractionClassificationCitationsEvals
03System layer
Mandatory human approval
The parts that stay with your people, permanently and by design.
- Handles
- Go/no-go, win themes, project experience, staff credentials, commercial commitments, final technical claims, and submission itself.
- How
- A review interface that shows what the system found, what it could not evidence, and what it wants to change — with a named approver on every action.
- Proof
- Nothing reaches a production system without a name attached to it.
Review UIExceptionsApprovalsAccountability
What you get
Three documents, produced before your kickoff meeting.
Every field traced to a source page. Anything unsupported is marked, not written.
Pilot deliverableHuman-reviewed
Nothing left in the addenda
Requirement Inventory
Every requirement in the package, extracted into a structured record with the page and section it came from. Addenda and appendices included, because that is where the disqualifiers hide.
Full package→Extraction→Structured record→Page citation
Forward-deployed delivery
We start by watching you work.
No discovery deck, no maturity assessment. One workflow, understood properly, then built.
- 01
Embed
Sit with the proposal team and watch a live package get processed. Record the exceptions and decisions nobody ever wrote down.
- 02
Draw the line
Decide what belongs to deterministic code, what belongs to bounded AI judgment, and what must stay with a person. Write it down and agree it.
- 03
Build on your stack
SharePoint, Word, Excel, Teams, and your CRM. No migration, no content library to populate, no per-seat licences.
- 04
Evaluate on your history
Twenty-five to fifty of your past packages become the test set. Recall, precision, citation accuracy, and false negatives are measured before anything touches live work.
- 05
Shadow, then production
The system runs alongside your team without touching anything. When the numbers hold against agreed criteria, it moves into limited production with monitoring.
How to start
A fixed price, and a written answer either way.
The sprint ends in an evaluation report. If the numbers do not clear the acceptance criteria we agreed, the report says so — and you have that in writing for CAD 5,000 instead of finding out after a platform rollout.
01 / Engagement
Free Automation Audit
Find the one workflow worth fixing before you spend money building anything.
- +Review one repetitive workflow end to end
- +Identify the bottleneck, exceptions, and hidden risk
- +Receive a ranked opportunity map and recommended first move
- +An honest no-build answer when automation is the wrong fix
02 / Engagement
Deployment Sprint
One proposal workflow, deployed in about fourteen business days. Half at kickoff, half on agreed acceptance criteria.
- +One workflow, one department, one intake source
- +Evaluated against 25–50 of your historical packages
- +Human-review interface and full audit trail
- +Shadow-mode deployment and a written evaluation report
03 / Engagement
Managed Operations
Keep it accurate as your work and your clients change.
- +Monitoring and evaluation maintenance
- +Correction review and continuous improvement
- +New workflows and refinements
- +User support, adoption, quarterly roadmap
Prices in Canadian dollars. Founding-client pricing applies to the first three deployment sprints. The free audit is open to the first five qualified businesses with one identifiable workflow and a decision-maker who can act on the result.
Non-negotiable
The guardrails are the product.
- 01Never submits a proposal.
- 02Never invents project experience or staff credentials.
- 03Never makes a claim it cannot cite.
- 04Never sets pricing or makes the go/no-go call.
- 05Never writes to a production system without approval.
Qualify yourself
Is this actually for you?
Pick the closest description. One of these answers is that you shouldn’t buy this yet.
Straight answer
You're the fit. Start with a workflow review.
High volume plus undocumented process is exactly where omission risk concentrates. Sixty minutes will tell you whether a pilot is worth CAD 5,000 to you.
Request a workflow reviewWho you are working with
Forward-deployed, not outsourced.
Object Method is Vatsal Patel. Enterprise generative AI adoption and unstructured-data extraction at Canadian Tire, including AI tooling training delivered to more than 200 stakeholders. Author of AIC, an open-source protocol for agent–tool interaction that raised benchmark task success from 33% to 100% while cutting required steps from twelve to three. Builder of Resume2Job, an AI career platform with DOCX-first document parsing at its core — the same class of problem an RFP package presents.
Common questions
Can you do this without seeing confidential proposals?
Partly. We can start with public tenders your firm has already bid, which are published documents. For the evaluation set we sign a mutual NDA first and provide a written data-handling brief covering storage, retention, deletion, and any sub-processors. If your policy requires it, the pilot can run entirely inside your own Microsoft tenant.
We're already evaluating Loopio, Responsive, or Joist.
Those are real products and worth evaluating. They solve content reuse — a searchable library of approved answers. They do not solve the part where the rules governing your submissions live in two or three people's heads and nobody has written them down. If a tool solves your problem, buy the tool. This is for when it doesn't.
Does this replace our proposal coordinator?
No. It removes the administrative reading and cross-referencing so the coordinator spends their time on strategy, win themes, client insight, and technical coordination. Firms that buy this are usually about to hire, not about to cut.
What if the evaluation shows it isn't good enough?
Then the report says so, with the numbers. Acceptance criteria are agreed in writing before work starts, and the second payment is tied to them. A documented negative result for CAD 5,000 is a good outcome compared with the alternative.
Fourteen days seems fast.
It is one workflow, one department, one intake source, one output. That scope discipline is what makes the timeline real. Additional workflows are separate engagements, priced separately.
What happens to our data?
It is used to build and evaluate your system and nothing else. No training on your content, no reuse across clients, defined retention with deletion on request, and every sub-processor named in writing before anything is shared.
Start with one workflow
Show me where the work slows down.
The free audit is a working session, not a sales presentation. We examine one real workflow, locate the costly friction, and determine whether the right next move is automation, simpler software, a process change, or nothing.