Technology

The point of tooling is to spend professional time only on what a person must do.

We do not build general-purpose software. Fourteen in-house products in two groups: six finance and tax tools that grew out of the dirty work recurring on real engagements, frozen and reused on the next client, all running locally with the data staying on the premises; and eight AI modules, which are product modules M09–M16 themselves, organising the daily work that is repetitive, scattered and carried in someone's head.

In-house products

Fourteen in-house products: six finance and tax tools, eight AI modules

This page is about what we build with — the software and systems Yidao writes itself. The six finance and tax tools support forward-deployed delivery of M01–M08; the eight AI modules are product modules M09–M16 in their own right. What all sixteen modules deliver, what the client provides and how many days each takes is on the Services page.

Finance and tax tools

Six tools, every one of them frozen out of a forward-deployed engagement

They all serve the same job: getting the money right. Each grew out of the dirty work that recurred on real engagements and was then reused on the next client. Apart from Anxinshui, which retrieves data once the company authorises it, all run locally and no data leaves the premises.

Big-data tax risk report

Anxinshui

The product works rather like an X-ray. With the company's authorisation it reads filing and invoice data, scans it against standardised risk models, and turns tax risk — something normally sensed only through experience — into a conclusion that can be quantified, compared and re-checked.

264 tax risk points, modularised and standardised into analytical models
Five risk domains: invoicing, VAT, corporate income tax, individual income tax and other taxes
Invoice-level tracing: non-compliant invoices presented across ten dimensions — generally absent from comparable products
Industry-specific reporting, with the assessment basis adjusted to sector characteristics
Detailed and summary editions, with the summary edition available for trial
Suited to self-review, internal audit, external audit, major investment decisions, M&A and pre-planning diligence

Data windowTrailing 36 months

Risk models264

Report versionNow at v10

GenerationAutomatic once authorised

DeliveryPushed to the finance lead

Typical users are companies in higher-risk sectors, mid-sized companies without a dedicated tax function, and advisory firms seeking to raise throughput on client work. The product site carries an anonymised extract of a real client report; a short version can be requested by QR code.

Product siteanxinshui.yidaocaiwu.com ↗(in Chinese)

A management dashboard an owner can actually read

Jingying Tongbao

Feed it the raw tables exported from the accounting system and it builds, locally, a single screen that answers three questions: are we making money, is there enough cash, and who is bleeding us. It does not do bookkeeping — only the handful of things an owner genuinely looks at.

Automatic recognition of nine export types: trial balance, bank statements, invoice detail, sales detail, receivable ageing, inventory, payroll, financial statements and the voucher journal
Order-level margin analysis and profit-and-loss rankings: products and customers ranked by margin, with orders priced below cost and their share of the book identified
Cross-table entity resolution, unifying the same counterparty across the customer, supplier and receivables tables to reveal dual-role counterparties and related-party triangulation
Credit and ageing controls: exposure against limit, days overdue and historical breaches, customer by customer
Cash-flow management: a three-month rolling forecast with break points identified ahead of time
Slow-moving stock analysis: capital tied up by stock age and dead stock identified, alongside purchase price trends, FX exposure and customer concentration

Runs asLocal desktop application

NetworkNo connection required

DataNever leaves the machine

InputAccounting system exports

OutputWeb dashboard / Excel

"The data never leaves the premises" is what makes this class of tool acceptable to a client, and it is guaranteed by the architecture — the program starts no service, makes no connection, and sits on the client's own computer. The product site carries a full anonymised dashboard run from six months of ledgers at a trading company with RMB 300m of annual sales.

Product sitejinyin.yidaocaiwu.com ↗(in Chinese)

Field tool and working-paper engine

Neishentong + CFAudit

Two tools, two users. Neishentong is for the auditor on site — double-click, no installation. CFAudit is the deep analytical engine used by the back-office analyst. They share one threshold definition but deliberately not one runtime.

Working papers as code: each paper is a data model plus assertions, reconciliations become assertions, and a break fails the build
Working paper index and lineage generated automatically; review trail and cross-client reuse handled by version control
Single source of truth: all thresholds and the paper index live in one specification that every layer derives from — no hard-coding anywhere
Golden regression: verified figures are frozen as a baseline, so a changed definition fails first — the purpose is not to prove the audit was right, but to prove the basis was not quietly altered
Related-party graph: common shareholders, shared contact details, adjacent account numbers and near-simultaneous first transactions flag probable common control
Digit distribution testing is used only to rank sampling priority, never as a conclusion of fraud

Field toolDesktop, zero install

Back officeLocal warehouse and model layer

Paper set34 papers across 6 lines (first engagement)

Automation12 fully automated / 11 semi-automated

Self-testBoth source and packaged builds

We do not buy a general GRC platform. The active open-source compliance platforms are all oriented to IT security and handle no financial data, invoice detail or accounting vouchers; commercial internal audit systems likewise do not perform invoice-level analysis under the Chinese tax system. The Neishentong product site sets out the three ways it is used: a trial pack, a per-machine professional licence, and a fully forward-deployed engagement with our own team on site.

Product siteaudit.yidaocaiwu.com ↗(in Chinese)

Customs and tax compliance for cross-border e-commerce

Cross-border Steward

From 2025 a set of public announcements built a look-through supervision regime over cross-border e-commerce, closing the grey space around "borrowed-declaration" exports. This system exists so that a company can still operate under that regime: documents complete, data traceable, refunds actually received.

One-click submission: connects to regional cross-border service platforms for 9610 / 9710 / 9810 / 1210 filing, with batch upload and download of declarations
AI-assisted HS classification combining model inference with product attributes to reduce both difficulty and classification risk
Integrated control across registration, shipment, declaration, invoicing, reconciliation and refund, keeping the data consistent at every step
Full-chain data integration linking declarations, input invoices, output invoices and orders into a traceable refund reconciliation ledger
Pre-submission validation, converting common review errors into automatic checks before filing
Managed service covering qualification filing, identity registration, hardware, network and IT staffing

Regimes9610 / 9710 / 9810 / 1210

InterfacesRegulator / platform / service side

PortsNationwide coverage

EntitiesOne principal, multiple companies

EditionsEnterprise / service provider

Supporting services cover customs registration, electronic port cards, cross-border e-commerce qualification and service platform registration — suitable for companies going cross-border from a standing start, with us alongside throughout. The product site also carries a free 61-point cross-border tax compliance self-check.

Product sitekuajing.yidaocaiwu.com ↗(in Chinese)

R&D super-deduction: capture, cap calculation and risk check

R&D Steward

Since Implementation Guide 2.0, an R&D super-deduction claim has to survive review, not merely be filed. R&D Steward turns an ERP export into the auxiliary ledger, the filing figures and the supporting file list on the local machine, and states in one place which items would be challenged, under which circular, and what to do about them.

Seven expense categories captured automatically — personnel, direct inputs, depreciation, intangible amortisation, new product design fees, other related expenses and commissioned R&D — each entry carrying its classification basis; anything unclear goes to a pending list rather than being guessed
Caps computed on current rules: other related expenses across all projects together under the 2021 No. 28 announcement, domestic commissioned R&D at 80%, overseas commissioned R&D additionally capped at two thirds, with the excess truncated automatically
The 2021 statutory auxiliary ledger and summary sheets align cell by cell with the official templates; every A107012 filing figure traces back to a voucher
24 risk checks, each with the circular it rests on, its impact and the remediation step, plus a supporting-file list with missing items flagged
Automatic recognition of a dozen mainstream ERP export formats: two-row headers, old and new account systems, amounts written as "200k", and web pages saved with an .xls extension all fall back to a working parse
A command-line build takes a folder path and produces the result directly — useful when a firm runs many clients in a batch

CategoriesSeven

Risk checks24

Filing formA107012

Runs asLocal desktop app / CLI

NetworkNone at all

The software is not distributed publicly; it comes with the engagement, with a hands-on introduction. Whether an activity is R&D remains the company's own determination: the output supports management and self-review, it does not replace confirmation with the tax authority, and the high-tech status indicator is a warning light rather than evidence for re-certification.

Product siterd.yidaocaiwu.com ↗(in Chinese)

Six modules · delivered forward-deployed

Sales Enablement and Assessment System

A sales training and assessment system that can be deployed and put live, with six modules of content, a question bank, a case library and a certification mechanism. It is built for organisations that need to turn the experience of senior salespeople into actions a new joiner can repeat.

Module 1 — Industry knowledge: market map, regulatory arc, client segmentation and compliance red lines
Module 2 — Product library: catalogue, delivery boundaries, pricing and discount authority, competitor comparison
Module 3 — End-to-end sales method: staged approach with explicit criteria from lead to repeat business
Module 4 — Skills: discovery, articulation of value, objection handling, negotiation and working with experts
Module 5 — HR framework: role profiles, career levels, compensation and new joiner development
Module 6 — Assessment: tiered question bank, scenario testing, field metrics and licence to practise

Modules6

DeliverySystem + content + run-in support

RolesSales / manager / HR / administrator

AssessmentBank + scenario + field metrics

FitsAdvisory firms / channel networks / sales teams

This is Yidao's organisational product delivered forward-deployed: we supply the system and the content and stay alongside during roll-out, but the judgement and the people management remain inside the client organisation.

Product sitefuneng.yidaocaiwu.com ↗(in Chinese)

Enterprise AI

Eight AI modules — product modules M09–M16

This group is a different animal from the six tools above. It does not serve finance or tax judgement; it organises the part of daily work that is repetitive, scattered and carried in someone's head — it keeps the business moving. The eight modules are bought separately or combined around a business objective; implementation scope, integration and boundaries are stated item by item, and nothing here is sold as one-click. For where each module sits in the process, see Cost and efficiency.

01AI development workflow and implementation serviceEngineering throughput · Stage 01 R&D and project approval
Connects requirements, design, build, test and release into a traceable process. For enterprise IT teams: workflow and repository templates, AI skill packs, standards adaptation and team training. Combined with your existing coding assistant, it gives requirements a basis, changes a record and delivery an acceptance step.
What it is forRequirements and design · AI-assisted development · Independent test and review · Delivery and knowledge capture
How it is usedClarify the requirement → confirm the design → assisted development → test and acceptance → release and write-back
What is deliveredWorkflow and repository templates, skill packs, standards adaptation, training and implementation service
Teams it suitsEnterprise IT teams that need requirements managed more consistently; engineering leads introducing AI-assisted coding; project owners who need a single delivery standard for outsourced work
Pricing basis and integrationPriced by team size, technology-stack adaptation, training and hands-on support; coding assistant subscriptions agreed separately. We assess your existing Git repositories, CI and coding assistant first, then adapt per project
BoundariesNo promise of automatic adaptation to every technology stack. The delivery is the toolkit and the process service; model calls, code transmission and account fees for the coding assistant follow the service you choose.
02Enterprise knowledge baseKnowledge and training · Stage Runs through all seven stages
Turns product, process and business experience into knowledge a team can keep using. Starting from scattered material, it establishes the categories, the content owners and the scope of use, giving enquiries, training, assessment and sales practice a sourced content base.
What it is forKnowledge structuring · Knowledge in use · Content maintenance · Linked to the AI coach
How it is usedTake stock → structure the categories → confirm the content → organise the applications → maintain it
What is deliveredKnowledge structuring service, the content structure itself and the agreed application modules
Teams it suitsCompanies whose material is scattered and inconsistently updated; business teams that need product and role experience captured; training leads who want content maintained over time
Pricing basis and integrationPriced by the scale of the structuring work, number of users, storage capacity and the AI usage selected. Already integrated with the AI coach, and works with online assessment to support knowledge, training and testing
BoundariesContent is confirmed by the company. AI question answering is an application scope that must be agreed separately; answers should cite sources and a route to human verification is kept open.
03AI teaching video generationKnowledge and training · Stage 03 Production and cost · 05 Marketing and content
Turns existing slide decks into training video that can be used again and again. Generating teaching content around the decks you already have cuts repeat recording, and gives onboarding, machine and process SOPs and policy briefings a reusable form.
What it is forDeck to video · Content review · Course reuse
How it is usedPrepare the deck → agree the narration brief → generate the video → review the cut → use it in training
What is deliveredEither the generation tool or finished course video, depending on the service scope selected
Teams it suitsTraining teams that have decks but no time to record; companies that run onboarding repeatedly; multi-team organisations that want one consistent course
Pricing basis and integrationPriced by minutes generated or courses produced; templates, voicing, revision rounds and storage listed separately. Connecting to a course library, assessment or viewing records requires the interface and the extra implementation scope to be confirmed first
BoundariesDigital presenters, voice selection, partial regeneration and learning records are extensions, confirmed item by item in the demonstration and the quotation rather than included by default.
04AI sales call coachKnowledge and training · Stage 04 Sales and collection · 05 Marketing and content
From knowledge preparation through live voice coaching to assessment and analysis. A training manager configures customer scenarios and knowledge packs and schedules training or assessment tasks; the trainee picks a task, completes a device check, holds a live voice conversation with an AI customer, and afterwards sees scores, commentary and a full transcript. Already integrated with the knowledge base and AI online assessment.
What it is forScenarios and knowledge packs · Live voice training · Linked to online assessment · Task management · Scoring and human review · Reports and recording playback
How it is usedPick your task → check the audio devices → live voice training → finish and generate the report → review the result
What is deliveredLive voice coaching application, scenario and task management, knowledge base and assessment integration, result reports and user training
Teams it suitsTeams developing new salespeople; leads who want one consistent training standard; training managers who need training and assessment results tracked
Pricing basis and integrationPriced by team size and the voice minutes included; overage, scenario production and integration listed separately. Knowledge packs are cited into coaching scenarios and assessment tasks sync into the coach
BoundariesThe report keeps both the model's raw score and the reviewed score, and the result follows the final score against the task's pass mark. Recordings and detail views are governed by permissions and retention rules.
05AI online assessmentKnowledge and training · Stage 04 Sales and collection · Organisation and knowledge layer
Connects the question bank, paper assembly, sitting, marking and result analysis. For corporate training and role certification: manage questions and assessment tasks, assist with marking and analyse weak points, so that results feed the next round of training.
What it is forQuestion bank · Paper assembly and sitting · AI-assisted marking · Result analysis
How it is usedPrepare the bank → assemble and issue → sit online → mark and review → analyse and retrain
What is deliveredAssessment administration and candidate applications, agreed question bank import and user training
Teams it suitsCompanies running product knowledge and role assessments; training teams spending heavily on manual assembly and marking; leads who need training results analysed by topic
Pricing basis and integrationPriced by usage scale and assessment requirements; AI marking, analysis usage and question bank preparation listed separately. Already integrated with the AI coach, whose assessment tasks enter the training flow
BoundariesAssembling a paper means selecting from an existing bank; generating new questions from source material is a different capability, agreed separately and reviewed by a person before entering the bank. Free-text answers, disputed items and consequential assessments keep human review.
06AI technical solution researcherBusiness automation · Stage 01 R&D and project approval
Searches for candidate solutions to a business requirement and organises the findings for the person deciding. For engineering, product and pre-sales teams: run the search around the problem at hand and assemble candidate leads with their sources, ready to be verified and compared.
What it is forRequirement framing · Finding candidates · Sources and comparison · Follow-up and capture
How it is usedFrame the requirement → prepare the brief → search for leads → check and compare → follow the chosen solution
What is deliveredThe research application and the agreed business-process integration
Teams it suitsEngineering and pre-sales teams responding to new customer requirements; companies looking for a product or technology direction; leads who want less repeated searching and collation
Pricing basis and integrationPriced by task volume, run frequency and model usage; business system integration listed separately. Requirement management, notifications and knowledge capture are integrated to the agreed scope
BoundariesEvery candidate lead must be checked against its source and its operating conditions — a reachable link, or several models agreeing, proves nothing on its own. What may be sent out, and how briefs are anonymised, is written into the deployment plan.
07AI procurement assistantBusiness automation · Stage 02 Procurement and supply
Reduces the repeated collation in purchase-order follow-up so that people concentrate on what needs judgement. It maps the repetitive work in the follow-up process and helps assemble order progress, correspondence and open items.
What it is forCollating follow-up information · Assisting repetitive work · Keeping the judgement calls with people
How it is usedMap the follow-up process → confirm the data sources → assist daily follow-up → human confirmation → capture the record
What is deliveredThe procurement assistance service and process adaptation within the agreed scope
Teams it suitsTeams with high order volumes and repetitive follow-up; companies where supplier correspondence is scattered and handovers are costly; procurement leads who want better visibility of order progress
Pricing basis and integrationPriced by order or task volume and integration scope; messaging channels, implementation and maintenance listed separately. Whether ERP, corporate email and messaging channels can be connected depends on the actual interfaces, authorisation and process
BoundariesSending messages automatically, writing back to orders, running enquiries and reconciling with finance are not part of the default delivery. What executes automatically and what needs approval must be written into the process item by item — the word “assistant” confers no such permissions.
08AI short video generationContent production · Stage 05 Marketing and content
Organises sustainable video production around material libraries and themes. Existing company footage is used to generate video on different themes, cutting the effort spent reselecting and remaking material.
What it is forMaterial and themes · Video generation · Review and reuse
How it is usedPrepare the material → choose the theme → generate the video → review the cut → use it by channel
What is deliveredVideo generation service, agreed theme configuration and finished output
Teams it suitsMarketing teams with plenty of material and too few people to produce video; companies preparing content across several themes; leads who want product and brand material reused
Pricing basis and integrationPriced by clips or minutes generated; template production, revision rounds and storage listed separately. Material libraries and object storage are assessed against your requirements; publishing platforms and account access confirmed separately
BoundariesAutomatic publishing, platform performance retrieval and one-click resizing for every format are not default commitments. Generating video is not a guarantee of reach — performance also depends on material, theme and channel.

Pricing basis by module

Software, usage and implementation are stated separately. A quotation sets out the functionality included, the allowances, the overage charges, the deliverables and the service period, never a single blended figure.

Module	Main pricing basis	Stated separately
Online assessment / knowledge base	Usage scale, capacity and the AI usage selected	Question bank or knowledge preparation, import, interface implementation
Sales call coach	Team size, voice minutes	Overage, scenario production, system integration
Teaching video / short video	Minutes or clips generated	Templates, voicing, revision rounds, storage
Solution researcher / procurement assistant	Task volume, frequency or order processing scale	Model or messaging usage, channel and system integration
Development workflow service	Team, technology stack, training and hands-on support scope	Coding assistant subscription, subsequent update service

Deployment choice, model and voice service fees, service period and delivery timing follow the confirmed quotation and implementation plan. These eight modules do not include back-end development beyond the systems described here.

Common questions

Do we have to buy a new development platform first?

We assess what you already have. The workflow can be introduced around your existing repositories and coding assistant, with the scope of adaptation set out in the implementation plan.

On measuring the effect: compare effort, delivery time, defects and rework for a comparable scope of work — and do not treat commit days as person-days.

Can we keep using our existing material? How is access to knowledge controlled?

Yes — start by taking stock of what exists and confirming format, quality, permissions and the import route before the structuring work is scheduled.

Access is set out in the plan: personnel permissions, material scope, storage location and audit requirements, verified against the actual delivery.

Is intelligent paper assembly the same as AI writing questions? Can it replace human marking?

They are different capabilities. Assembly selects from an existing bank; generating new questions creates them from source material, which is agreed separately and reviewed by a person before entering the bank.

Multiple-choice and free-text answers are marked differently. Free-text answers, disputed items and consequential assessments keep human review.

Are the coach, the knowledge base and online assessment really connected? Does the trainee type or speak?

They are connected. Knowledge citations and versions flow into coaching scenarios, assessment tasks sync into the training list, and the results feed later analysis.

The trainee speaks with the AI customer through a microphone in real time. The page shows both sides line by line as live subtitles, and a report is produced afterwards for review.

How should the AI score be used?

The report keeps the model's raw score alongside the human review record; the result follows the final score against the task's pass mark, with sub-scores and evidence rounds available for review. Where a person should decide, the tool only assists.

Can the researcher guarantee it finds a usable solution? What is sent to the model?

What the search finds depends on the requirement, the public material and the conditions for verification. The delivery scope covers the research process and its output; it does not promise that every problem has a ready answer.

The deployment plan sets out how briefs are anonymised, what may be sent out and the scope of model services, so that confidential source material is not used directly in external research.

Will the procurement assistant send messages or change orders on our behalf? Do we have to replace our ERP?

What executes automatically and what requires approval must be defined in the process; the name “procurement assistant” does not confer those permissions.

We assess your existing systems and process first. Whether integration is possible, by what route and at what cost is settled before implementation.

Can short video publish to every platform automatically? Can you guarantee views?

Publishing depends on the platform interfaces and account conditions and is confirmed item by item; finished clips can also be published by the marketing team in the usual way.

A video generation service is not a guarantee of reach. Performance also depends on material, theme, channel and how it is run.

Do we need to redo our decks? Can teaching video be bought with assessment?

We assess the structure, readability and material quality of the existing decks first, then confirm whether changes are needed.

They can be bought together. But linking a course to an assessment, issuing it and syncing the records must be agreed explicitly — buying both does not make them connected.

How to start

Talk through the business problem first, then decide whether to take a single module or a combination built around a scenario. The eight modules have no public trial site at present; a demonstration can run on industry samples, and anonymised, authorised material comes later in the conversation.

See where these modules sit in the process

Have a specific problem and want a professional view on it?

Tell us the facts and we will tell you what kind of problem it is, how it is normally handled, and how far it can realistically be pushed on the evidence you have.