Learn how to pick a software development agency that ships—avoid overpromises, hidden costs, and mismatched tech stacks. Fixed-scope criteria for SMBs.
How to evaluate a software development agency gets real when you need a fixed-scope build—not an open-ended retainer. Wholesale distributors on Epicor Prophet 21 or NetSuite, multi-location operators pushing EDI 850/856 through SPS Commerce, and early-stage SaaS founders hiring for React Native field tools or RAG eval harnesses need proof that goes past “we ship MVPs.” Here are 11 criteria, a 30-minute audit, and walk-away signals you can use on the next sales call.
Keyword math: “how to evaluate a software development agency” sits at roughly 80–180 monthly searches (US + EU) with difficulty 35–50. Generic “how to choose” posts dominate; few agencies publish criteria for ERP-adjacent portals, offline mobile, or LLM evals. KPI: 5 qualified organic inquiries in 90 days. Review date: 2026-12-05.
A strong partner delivers a written Scope of Work on time and on budget, extends your system of record (NetSuite, Epicor, SAP Business One), and treats cash like an operator—not a VC. Weak partners sell “digital transformation” with no definition of done, push React + Node.js when your team lives in Excel, or dress retainers up as “MVP sprints.”
You hire for outcomes: a customer portal with branch pricing, a React Native app for Zebra scanners, or a RAG pipeline with retrieval evals. Not a rewrite of your ERP.
Most vendor guides assume a funded SaaS founder shopping for an MVP template. You need fixed-scope delivery in weeks—ERP integration, offline field tools, or RAG evaluation harnesses—not a full-stack rewrite. The right shop will not pitch Salesforce custom objects when you need a portal that syncs customer-specific pricing from Epicor over REST.
If the deck only shows B2C apps and Chrome extensions, keep shopping. [Internal link: custom software development for wholesale distribution]
Score each vendor 0–2 on the criteria below (0 fail, 1 partial, 2 meet). Aim for ≥18/22 before a paid discovery call. Every item has a red flag and a green flag so ops and finance can score the same way.
A good agency extends NetSuite, Epicor Prophet 21, or SAP Business One. They do not propose a rebuild. They ask for ERP schema, API docs, and EDI endpoints before naming a stack. If the first slide says “migrate to PostgreSQL,” they are not a fit for distributor workflows.
Wholesale ops run on credit holds, lot inventory, and branch pricing. Rebuilds wipe years of process. Extensions keep them.
Red flag: Modern-stack rebuild with no review of ERP, WMS, or accounting.
Green flag: Names ERP version, API limits, and integration pattern (REST, SOAP, EDI 850/856) before quoting.
Fixed scope means a Statement of Work with price, timeline, and out-of-scope items in writing. Time-and-materials or “we’ll figure it out” burns cash. Ask for a sample SOW. Confirm the not-building list is longer than the building list.
SMBs and early-stage teams live on cash-flow discipline. Unbounded burn rate blows past budget before month three.
Red flag: “Starting at $X” with no SOW, or “we’ll scope iteratively.”
Green flag: Fixed-price SOW with milestones, acceptance criteria, and a not-building appendix.
Look for NetSuite, Epicor, Fishbowl, or SAP Business One integrations. A React dashboard for a YC startup proves coding skill. It does not prove they can extend an ERP without breaking credit holds or contract pricing.
Distributors need portal logic for ship-to rules, approvals, and EDI order replay. MVP case studies rarely test those constraints.
Red flag: Portfolio of B2C apps, Chrome extensions, or AI marketing toys.
Green flag: Case study with ERP sync, customer portal, or offline field-order app for Zebra scanners.
Defaults should match your team and system of record. For distributors that usually means Next.js or React portals, Python or Node.js pricing services, PostgreSQL on AWS RDS or GCP Cloud SQL, and React Native for offline field tools. One-stack shops (Ruby on Rails or Django for everything) optimize for their bench, not your ops.
Field reps need apps that sync when Wi-Fi drops. AR needs dashboards that pull Epicor pricing. Wrong defaults force retraining.
Red flag: One stack for ERP portals, field apps, and RAG pipelines.
Green flag: Names React, Next.js, Python, React Native, and PostgreSQL—and explains why. [Internal link: React Native offline field apps for distributors]
Integration is not “call the API.” It is idempotent writes, partial-failure handling, and clear read vs write boundaries. Ask how they handle EDI 850/856 via SPS Commerce or TrueCommerce, or REST sync with NetSuite. “We’ll figure it out” usually means duplicate orders in production.
A portal writing orders into Epicor must honor ship-to rules, contract pricing, and credit holds. One bad sync can cost tens of thousands in chargebacks.
Red flag: No idempotency, error handling, or sync boundaries.
Green flag: REST idempotency keys, EDI retry logic, or change-data-capture patterns.
Field teams need React Native apps that sync offline and scan barcodes on Zebra TC-series devices. Progressive web apps fail that constraint. Ask for offline-first patterns and local sync proof.
A PWA cannot scan reliably in a dead zone. React Native with SQLite and offline query sync can.
Red flag: PWA or web-only for field tools.
Green flag: React Native + SQLite + offline sync in a portfolio item.
Value-based pricing ties cost to outcomes—for example, dollars to cut four hours per week of email order entry. Hourly pricing ties cost to their burn rate. Compare a value quote to their hourly rate. If labor savings clear the build cost in under a year, the math is visible.
Open-ended hours break SMB budgets. Fixed value quotes align incentives.
Red flag: Hourly only, no outcome framing.
Green flag: Fixed price tied to an ROI metric (for example, cut 120 hours per year of manual pricing updates).
For RAG work, ask how they measure retrieval quality and when they fine-tune embeddings. “We’ll use LangChain” is not a production plan. Look for eval harnesses, retrieval benchmarks, and production monitoring.
A chatbot that invents customer pricing or inventory damages trust. Models without evals degrade quietly.
Red flag: No evals, retrieval quality, or production monitoring.
Green flag: RAG eval harness or fine-tuning runbook in the portfolio. [Internal link: RAG pipeline evaluation for production agents]
SaaS founders often need product leadership, not only coding. Ask whether they run discovery, prioritize the backlog, and write user stories. “You tell us what to build” is a code shop, not a product partner.
Dashboards without research and prioritization ship features, not value.
Red flag: No discovery, research, or prioritization process.
Green flag: Product strategy engagement or embedded PM case study.
DevOps means Terraform, AWS/GCP, CI/CD, and cost guardrails—not only GitHub Actions. Ask about cost monitoring, secret rotation, and rollbacks. Defaulting to Heroku for ERP-adjacent systems is not production-ready.
A Next.js portal on the wrong host can spike cost and fail under peak load. Terraformed AWS or GCP with cost alerts scales with guardrails.
Red flag: Heroku/Vercel/Netlify as the home for ERP-adjacent systems.
Green flag: Terraform, AWS/GCP, CI/CD, and cost monitoring named in the stack.
Ask for a sample contract before you engage. Look for fixed scope, acceptance criteria, payment milestones, and IP transfer. NDAs plus handshake deals are not enough for a fixed-scope build.
Written scope protects budget and timeline. Pure T&M without guardrails is risk.
Red flag: No sample contract or T&M-only proposals.
Green flag: Sample SOW and contract template shared upfront.
Send a five-question pre-call survey, score vendors on the eleven criteria, then pay for discovery with the top two. A short, structured audit filters decks that look polished but cannot speak ERP, offline mobile, or fixed SOWs. That saves weeks of unpaid “strategy” calls that never produce a not-building list.
Walk away when the pitch pushes a full ERP rewrite, hourly rates with no value framing, zero ERP-adjacent proof, PWA-only field tools, or refusal to share a sample SOW. Those patterns predict scope creep, duplicate orders, and dead-zone field failures—not a fixed-scope delivery your finance lead can defend.
Also leave if they cannot name EDI partners, offline sync patterns, or how they evaluate RAG quality when those are in your RFP.
Plan two weeks: one week for RFPs and surveys, two paid discovery calls, then a scored decision. Longer evaluations usually mean missing SOWs or unclear systems of record—not deeper diligence. Freeze the ERP version and budget ceiling before the first call so vendors quote the same problem.
Yes, if you lack an in-house PM. Embedded product leadership turns vague pain (“pricing emails take forever”) into backlog items with acceptance criteria. Pure code shops wait for specs you do not have time to write, which recreates the T&M trap under a different label.
Demand an eval plan: retrieval benchmarks, failure cases (pricing and inventory hallucinations), and production monitoring. LangChain demos without evals are prototypes. Ask who owns model drift after launch and what “definition of done” looks like for accuracy, not only UI polish.
Usually no. Warehouses with weak signal need React Native (or native) with local storage and barcode support on devices like Zebra TC-series. PWAs work for office portals; they fail when scanning must work offline during receiving or cycle counts.
Wolverine Solution builds web apps (SaaS dashboards, internal tools, customer portals), React Native and native mobile, AI/LLM systems with RAG and evals, UI/UX, Terraform-based AWS/GCP DevOps, and embedded product strategy for early-stage teams—scoped as fixed builds for distributors, multi-location operators, and technical founders.
Bring your ERP version, top workflow, and budget ceiling. We will map them to a fixed-scope outline with a not-building list—or tell you plainly if you should keep the spreadsheet for another quarter. Book a paid discovery audit with Wolverine Solution at https://wolverinesolution.com and use the eleven criteria above on every competing pitch.