HNTE case studies

SOFTWARE · FIRST-TIME

A clean ledger with nothing behind it

A 120-person scheduling-SaaS company with 400+ customers on annual subscriptions. The bookkeeping was tidy — which is exactly why the gap was obvious: there was no R&D expense account anywhere, and every engineer's salary sat in cost of sales.

The real level was never below the threshold

Reconstructing staff structure put actual R&D intensity above 8% — comfortably over the 5% threshold for companies under RMB 50M. What was missing was not the spend but any auditable trail from person to project to ledger.

So the sequence was reversed: compute the true number first, then build evidence backwards. Companies that start by writing documents end up with documents that contradict the ledger.

Three breakpoints: people, money, proof

People. 68 of 120 staff were technical, but 22 were long-term on customer sites doing implementation and maintenance — under the 183-day rule they cannot count. The remaining 46 included custom-development work that is not product R&D. Roles and hours were split on two axes; anyone below 50% R&D hours in a month was excluded. Final qualifying count: 43 people, 35.8% of headcount.

Money. Cloud and bandwidth only qualify where the environment is R&D or test — production capacity earns revenue, it does not do research. Environment tags isolated about 18% of cloud spend; the rest stayed in cost.

Proof. Nine copyrights said nothing about which release they belonged to. Three major versions were split by module into separate results, each tied to a release date and contract value — 12 commercialisations over three years, four per year on average.

Why 7.8% and not 7.67%

Three-year R&D spend of RMB 6.63M against three-year revenue of RMB 85M gives 7.8%. The annual ratios were 6.6%, 7.6% and 8.8% — an arithmetic mean of 7.67%. The filing system uses the aggregate ratio, not the mean of annual ratios, and reviewers read it the same way.

The denominator is sales revenue: main business plus other operating revenue, excluding non-operating income and investment gains. Adding government grants to the denominator is a common and expensive mistake.

Result. Certified first time, four months end to end, at a 7.8% ratio — 2.8 points of headroom over the threshold.

ADVANCED MANUFACTURING · RENEWAL

The patents were not missing, they were spent

A 380-person non-standard automation builder, RMB 120M revenue, first certified in 2019, renewed in 2022, expiring in 2025. The owner's opening line was: “we have everything — six patents, just file it.”

Class II IP is single-use

Utility models, designs and software copyrights count once in an enterprise's certification history. Those six had already carried two filings — strictly, the 2022 renewal was already defective, it simply was not challenged. By 2025 the balance was zero.

The fix is not to file a batch of patents in the final quarter. Patent grant takes 8–14 months for a utility model and 2–3 years for an invention, so the plan has to start at least a year before expiry.

The second signal: R&D falling 30%

Spend dropped from RMB 6.2M to RMB 4.3M across the cycle. Growth scoring reads revenue and net-asset growth, but a simultaneous fall in R&D spend is read as a company stepping back from technology — and the innovation evaluation is scored by people, not formulas.

Existing work was documented honestly: project initiation records dated to when work actually started, with a note explaining that the file was compiled later. Nothing was back-dated — a back-dated document is the fastest way to fail an on-site check.

What renewal actually tests

Renewal is not a re-submission. It asks whether the company kept innovating for three years, and three years of evidence cannot be manufactured at the end of them. The files that carried this case were timesheets, project reviews and acceptance records from 2022–2024.

Result. Certified on renewal, three months of work, innovation score 78 — with new IP filed in time rather than after the fact.

BIO-PHARMA · FIRST-TIME

Strong technology, silent paperwork

A 90-person chemiluminescence immunoassay developer with RMB 48M revenue and three invention patents. Real R&D intensity was close to 10%, double the 5% threshold — normal for reagent development, where cost sits years ahead of revenue.

A mock review scored 62

Every place that should have held evidence held only a conclusion. Projects had been discussed in group chats but never initiated; clinical trial spend shared an account with production cost; commissioned testing was booked at 100%; and twelve production staff were on the technical staff list.

None of that is a technology problem. It is a documentation problem — and documentation problems are fixable, which is why the mock score mattered more than the raw R&D number.

Rebuilding records from what actually happened

The anchor was the registration test acceptance number for each reagent kit. From there, project start and end dates, participants and stage goals were reconstructed, and chat threads, trial batches and test reports were filed against those dates.

The rule was no back-dating and no embellishment. An initiation document dated to the real start, with a note explaining when the file was compiled, survives an on-site review. A document dated to look tidy does not.

Where the money was leaking

Commissioned testing was booked at 100% of invoice; the working guidelines only allow the portion attributable to R&D. Clinical spend was separated from production by registration project. Re-cutting both moved the reported ratio from an estimated 9.9% to a defensible 9.1% — lower, but supportable line by line.

Result. Certified first time in five months, with a dossier the finance team could defend in an on-site check.

ENERGY & ENVIRONMENT · RENEWAL

0.8 points short — and hiring was the wrong answer

A 200-person industrial wastewater engineering firm, RMB 70M revenue, up for its second renewal. HR counted 18 technical staff out of 195 — 9.2%, 0.8 points under the line. The first instinct was to hire six engineers.

Hiring would not have worked in time

A hire takes three to six months to land, and anyone on site fewer than 183 days in the year still does not count. Hiring in Q4 fixes nothing for that filing year. Recomputing the fraction took a week and added nobody.

The result after recomputation: 22 technical staff out of 193 total — 11.4%.

Three scoping errors, all in the company's favour

Numerator too small. Commissioning and process-tuning engineers work on site, not in the office, and HR had filed them under Engineering. Commissioning and parameter optimisation are related technical innovation activity under the working guidelines, so they qualify.

Affiliate employment. A handful of R&D staff were insured through a related company. Employment was moved back and social-insurance records completed before filing.

Denominator too large. Headcount included a large pool of short-term labour. Recalculated on the 183-day rule, the denominator fell from 195 to 193 — small in absolute terms, decisive at the margin.

Why this case is about scoping, not effort

Most “we are 0.5 points short” situations are scoping errors before they are capability gaps. Recompute both sides before spending money: new hires are the slowest and most expensive lever available, and they only work if they are on payroll long enough to count.

Result. Certified on renewal in three months, at 11.4% technical staff, without a single new hire.