Chief Technology Officer / India
What the product could support after the work
- Sewing-equipment retail network
- 44 showrooms
- Fictional Veznori Sewing Retail, based in Coimbatore
- Product-engineering team
- 64 employees
- Eighteen contractor positions are additional; six reporting leaders
- FY2026–27 delegated authority
- ₹26 crore
- ₹18.2 crore operating expense plus ₹7.8 crore approved product-engineering work
Decisions & consequences
01Replacing a model-name assumption with explicit device compatibility
In January 2023, a proposed firmware package was being prepared for sewing equipment sold under one familiar retail model name. The bench review included 100 devices drawn from the relevant supplied revisions. Eight did not match the hardware assumptions of the package. The commercial label had remained stable while an internal revision had changed. No customer release had occurred; these were pre-release findings, not a report of eight damaged machines or customer incidents.
The product sponsor wanted to continue with the planned pilot and explain the exception through support. I did not support asking a customer or a support colleague to discover a compatibility boundary after introduction. We funded ₹62 lakh for the compatibility work, test equipment and implementation of an explicit supported-revision check. The pilot moved by seven weeks. The delay affected demonstration plans, and I kept that consequence visible rather than describe the technical correction as something we could absorb without changing the programme.
Engineering quality head Mira Zervathi worked with the qualified manufacturer representatives on the supported relationships. Our software team did not infer hardware safety from a shared name or treat its own tests as a replacement for required manufacturer approval. The application needed to recognise the relevant revision and refuse an unsupported package through the approved process. Where identification was uncertain, the case returned for authorised review. The aim was to prevent a convenient commercial label from making a technical decision outside its competence.
The later controlled pilot covered 214 eligible devices. Its records showed no incorrect package assignments within that group. That observation did not establish compatibility with every device ever sold under the model name, and the eight original exceptions did not become eligible merely because the pilot proceeded. The supported set and remaining exclusions stayed in the customer and service account. My lesson was to ask for physical-revision evidence before committing the launch date. I had learned that discipline in instruments, but a familiar retail category had made it too easy to assume the distinction was already handled.
Letting eligible local projects open without a live cloud session
A February 2024 service interruption generated ninety-six reports from customers unable to open embroidery projects. Seventy-four concerned eligible files already downloaded to the customer's computer. The other twenty-two involved files not yet present locally and therefore required a different recovery path. Our application had placed an online-session requirement in front of both activities. The shared entry point was simple for the product team but unnecessarily interrupted work that could legitimately continue without fetching a new file.
The first proposal was to improve the cloud service and leave the client behaviour unchanged. Better availability was useful, but it would not answer whether the dependency belonged in that particular action. I approved ₹48 lakh to separate reopening eligible local projects from actions that required an online transaction or permission check. Product and relevant licensing owners defined the rights that had to remain respected. The work did not create a route to acquire unpurchased designs or disregard restrictions on files a customer was not entitled to use.
We tested 180 defined local-project scenarios after the change. One hundred and seventy-four opened successfully without a current cloud session. Six contained damaged files and still required recovery work; they were not network failures disguised as a success. The recovery process preserved the customer's existing copy rather than silently replace it with a different version. The twenty-two original reports involving absent files were handled separately when the service returned. I wanted support colleagues to explain those distinctions instead of promising that the application had become entirely offline.
The result was a narrower and more useful dependency. Existing eligible local work no longer waited for a session merely because the product had one common opening flow. New acquisition, absent content and other genuinely online actions retained their stated requirements. We did not convert the test count into a claim about every future file or the total commercial cost of the outage. I would now review network assumptions when a feature is designed, asking which part of the customer's work actually requires a service and which part has inherited the dependency for our convenience.
Moving preview acceptance from engineering workstations to customer computers
In November 2024, our proposed embroidery preview performed well on the workstations used by its developers. A representative review across twenty-four defined customer-device configurations gave a different account. Seven exceeded the agreed resource limit or failed the twenty-second completion criterion for the fixed project set. The test did not represent every computer in the market, but it showed that the demonstration environment was narrower than the range the product team hoped to support.
The faster commercial proposal was to publish a higher minimum specification and proceed. That could have been a legitimate product choice if explained clearly, but it would exclude customers whose machines were within our intended audience. I recommended a ₹2.1 crore engineering programme to change preview preparation and resource use before settling the support boundary. The investment displaced other feature work. We kept the project set and evaluation conditions documented so a later improvement could be assessed against the same defined work rather than a more convenient demonstration.
The revised programme used staged preview preparation and a bounded resource plan, with engineering quality reviewing the behaviour. Detailed implementation remained the team's responsibility. I asked for evidence that the approach preserved the intended preview and did not merely achieve a time criterion by omitting relevant work without disclosure. Commercial colleagues needed to understand which experience the benchmark measured and what the customer would see while it completed. A faster internal component was not enough if the whole supported action still felt stalled or produced an incomplete result.
Twenty-three of the twenty-four configurations subsequently met the agreed criterion for the fixed set; one remained unsupported. We published that boundary instead of describing the result as universal compatibility. No conversion or revenue benefit was established by the benchmark. The useful outcome was a more defensible supported range and a repeatable account of the work used to assess it. I would bring representative customer equipment into the first product review next time. By waiting until the developer demonstration was persuasive, we made a necessary engineering question feel like an obstacle to a promise that had already gathered momentum.
Requiring a reproducible build before treating source delivery as handover
A supplier transition beginning in September 2025 delivered source archives for eleven software components. In the agreed receiving environment, eight built successfully and three did not. Two depended on commercial libraries whose continuing use needed proper arrangements, and one required a generator version absent from the handover. The archive was real and substantial. It was also insufficient to establish that our team could reproduce and maintain the product under the intended conditions.
The transition sponsor wanted to close the handover milestone because the files had arrived and the supplier's team could demonstrate its existing build. I asked for a receiving-team build and test result instead. Legal and commercial owners handled the relevant rights and supplier arrangements; engineers identified the technical dependencies and evidence needed. Possession of source was not treated as permission to use every accompanying component. The original transition budget was ₹86 lakh, and the work eventually cost ₹94 lakh, an ₹8 lakh overrun that remained in the report.
After the dependency and tooling issues were resolved through the authorised route, all eleven components built in the agreed environment. That did not finish product acceptance. Twenty-four of twenty-six defined device tests passed; two features failed their required checks and remained deferred. Mira Zervathi retained the engineering-quality decision, with manufacturer involvement where the tests concerned hardware relationships. The transition did not become a reason to relax a criterion merely because the supplier relationship was approaching its planned end.
The result was a usable build capability and a clear account of the features still outside release. I did not call source delivery a complete transfer of intellectual-property rights or a promise of indefinite independence from suppliers. The actual agreements and continuing dependencies mattered. My retrospective change is to rehearse the receiving environment earlier, before the final handover date carries commercial pressure. A contract milestone should be supported by evidence of the capability it is meant to provide, and engineering leaders should make that evidence practical enough for colleagues outside the team to understand.