Most ERP selections run by an internal team take four to seven months from the day the team is named to an approved decision paper. ERP Research puts a mid-market selection at three to six months and an enterprise selection at six to twelve, and ERP Focus reports an average of seventeen weeks. Where a selection lands in that range is decided by three choices the team makes early. The requirements register is held to about 200 differentiating lines instead of 2,000. No more than three vendors are invited to demo. Two vendors stay in commercial talks until the decision paper is signed. The eight phases, their outputs, owners, time ranges, and common failures follow.
The eight phases, what each produces, and what moves the clock
The table below is the sequence that works for an internal team. The phases overlap on purpose: the long list is built while requirements are still being finished, and reference checks run during the last demo. Time ranges are ERP Research's stage estimates where it gives one, and this site's judgment where it does not, marked as such.
| Phase | What it produces | Who inside the company does it | Typical time and what moves it | Common failure |
|---|---|---|---|---|
| Getting organized and funded | A one-page charter (scope, who signs the decision, budget envelope), the evaluation team roster, knock-out criteria for the long list, and demo days blocked in every scorer's calendar | Sponsor, program manager, IT director, finance lead, procurement, one lead from each core process area | One to two weeks (ERP Research). Stretches to a month or more when the sponsor has not been named or the budget envelope is still being argued. Shortens to days when a replacement decision has already been taken and funded. | The decision maker is a committee. Demo dates are not blocked, so demos slip by a month when a key reviewer is unavailable. |
| Deciding what you need | A requirements register of 150 to 250 lines tagged Must, Differentiator, or Standard; a current-state integration map; a data volume sheet | Process leads run their own area's workshops, IT architect owns integration and data, program manager owns the register | Three to six weeks (ERP Research). Stretches when the register is allowed to list every function (six to eight weeks and still growing), when more than one site must agree on common processes, or when nobody owns the final cut. Shortens when the team starts from the processes that differ by industry and skips the ones every product does. | A 1,500-line register that lists every function the business performs. Every vendor scores 95% and the register tells you nothing. |
| Building the long list | A list of 8 to 10 vendors matched to industry, company size, and deployment preference, with the knock-out criteria written down | Program manager and IT director, with procurement checking SI coverage in your region | Two to three weeks (ERP Research), overlapping the end of requirements. Stretches when the tier question (tier 1 or tier 2) has not been settled, because the list then has to cover both. Shortens when the industry has a known set of contenders. | The list is lifted from an analyst quadrant or a free comparison site. Nobody can say why a vendor is on it. |
| Cutting to a short list | A short-list memo naming 4 to 5 vendors and the reason each other vendor was dropped; a 20-question RFI if the knock-outs leave more than 5 | Program manager, IT director, finance lead | One to two weeks (this site's judgment). Stretches when an RFI is needed and vendors are given three weeks to answer. Shortens to days when the knock-outs do the work. | Vendors are kept "to be fair." Six RFP responses take twice the scoring time of four and tell you no more. |
| RFP and scoring | An issued RFP with the register and your contract terms attached; a scoring model with weights locked before any response arrives; scored responses with indicative five-year pricing and a named SI; three demo finalists | Procurement issues and runs Q&A, process leads answer vendor questions, all scorers score independently and then reconcile | Three to five weeks (ERP Research). Stretches when vendors ask for more time (grant one week, not three) or when the scoring model is built after responses arrive. Shortens when the register is short enough that vendors can answer in two weeks. | Weights are set after responses are read. The model then rewards the vendor the team already liked. |
| Demos | Demo scorecards per process and per scorer; a fit-gap list per vendor; a separate SI scorecard | Full evaluation team, end users for their own processes, program manager running the script | Four to six weeks (ERP Research) for three vendors, two days each, with two weeks' notice for vendors to load your data. Stretches by a week for every vendor beyond three and by two weeks when scorers cannot attend both days. Shortens when the script and data went out with the RFP. | The vendor runs its own standard demo. The team scores presentation skill, not fit. |
| Reference checks | A reference log with the same ten questions asked of every reference; a risk register per finalist; SI team CVs checked | Program manager, IT director, finance lead; the sponsor takes one executive-to-executive call per finalist | Two to three weeks (ERP Research), overlapping the last demo. Stretches when references are only vendor-supplied and have to be chased, or when a site visit is wanted. Shortens when your industry association can put you in touch with recent go-lives directly. | Vendor-supplied references only, 20-minute calls, no site visit. The call confirms what the salesperson said. |
| Commercial negotiation | A five-year TCO per finalist on one template; a draft software agreement and an SI statement of work for the design phase from both finalists; a negotiation log | Procurement leads, legal reviews, finance lead owns the TCO, program manager keeps both vendors moving, sponsor takes escalations | Three to six weeks (ERP Research). Stretches to eight or more for a tier 1 contract, where the vendor's standard agreement runs to hundreds of pages and legal review alone can take three weeks. Shortens when terms were sent with the RFP and both vendors are still in the room. | The runner-up is told it has lost before terms are agreed. The remaining vendor's final price moves up, not down. |
| The decision paper | An approved paper recommending one vendor and one SI, with five-year TCO, the implementation timeline, the top ten risks, the conditions under which the recommendation would change, and every vendor dropped with the reason; a signed software agreement and SI design-phase statement of work | Program manager writes it, finance lead owns the numbers, sponsor signs it, steering committee approves | One to two weeks (this site's judgment). Stretches when the paper has to go to a board that meets quarterly, which is a date to find in the first phase, not the last. | The paper presents a score and a price. It does not say under what conditions the recommendation would change. |
The first two weeks decide whether the other phases are possible
Getting organized means fixing three things before any vendor is contacted: the calendar, the decision rights, and the budget envelope. The right default is to block two days a week for the demo month in the first week, before the requirements work starts. Demos need six to ten people, including the plant controller, the finance director, and the IT architect, in one room for two days at a time. Those people are booked a month or more ahead. When the days are not blocked early, a demo slips by weeks, because each missed day has to be rebooked for every scorer and the vendor's team.
The decision maker has to be one person, usually the sponsor, who signs the decision paper. The evaluation team scores and recommends, but a scoring model cannot make the decision. RAND Group's checklist makes a similar point when it says must-have requirements act as pass or fail gates that a weighted score cannot override. A committee that must reach consensus on a 0.4-point difference between two finalists will take a month longer and still not decide.
The budget envelope does not need to be exact, but it has to be in the right order of magnitude so the long list is built at the right tier. Implementation services run 1x to 3x first-year license or subscription, and five-year total cost runs 3x to 5x license, according to the ERP Research cost breakdown. For published program cost ranges by tier and company size, see the companion article on ERP Executive on what an ERP program really costs. A sponsor who expects a $1M program and is shown an $8M tier 1 program in the decision paper will send the team back to the long list.
Consider a $600M distributor with 1,200 employees, three warehouses, and one ERP covering finance, purchasing, and inventory, whose team is named in the first week of January. Getting organized takes the rest of January. Requirements run to late February, with the long list built in the last two weeks while the register is closed. The short list is cut in the first week of March, the RFP is out by mid March, and responses are back by the end of April. Demos run through May, one vendor every ten days, with reference calls into the first week of June. Commercials run from June into mid July, and the decision paper goes to the sponsor in the last week of July. That is six and a half months. The two places it slips are predictable. A warehouse lead cannot attend demos during the May inventory count, so everything after the demos moves a month. Legal review of the two vendor agreements takes five weeks instead of three. Either slip alone takes the selection to seven and a half months; both take it to eight and a half. Both are fixed in January, by checking the inventory count date and by sending your contract terms out with the RFP so legal starts in March.
About 200 requirements beat 2,000
The requirements register exists to separate vendors, not to describe the business. A general ledger, accounts payable, purchase orders, and a sales order are standard in every product on the long list, so listing them earns no information. The lines that matter are the ones a given vendor might not do well: the rate-case cost allocation a regulated utility needs, the lot genealogy a food manufacturer needs, the revenue recognition rule a project business needs. A register of 150 to 250 such lines, tagged Must, Differentiator, or Standard, lets scorers see daylight between vendors. SelectHub and RAND Group both prioritize requirements in three tiers, but neither sets a ceiling on the count, and that ceiling is what keeps this phase at three weeks instead of eight.
The long register fails in three ways. It takes six to eight weeks to write instead of three. Vendors answer "supported" to 95% of lines, so the RFP scores bunch within two points of each other and the demo has to do all the work. And it carries into the contract, where a 2,000-line appendix of "supported" answers gives the SI grounds to call anything not listed a change order.
Suppose the register reaches 1,400 lines because every workshop added what its area does today. At ten minutes a line, each of five RFP vendors spends about 233 hours answering it, and the three that reach demos have spent 700 hours between them before a single process has been shown. Your own team then reads 7,000 answers; at two minutes each that is another 233 hours of internal time. Now look at what the scores say. If 1,200 of the lines are functions every product on the short list supports, every vendor scores 5 out of 5 on all 1,200, which is 6,000 identical points. The 200 lines that differ carry at most 1,000 points. A vendor that fails 40 of those 200 outright, scoring 1 instead of 5 on each, loses 160 points and finishes at 6,840 out of 7,000, or 97.7%, against a vendor at 100%. On the scoresheet that reads as a tie. Cut the register to the same 200 lines and the same two vendors score 1,000 and 840, or 100% against 84%. The gap was there all along. The 1,200 standard lines hid it and cost five vendors about 1,170 hours between them and your team 233.
Three scripted demos, with the SI presented separately
Three demos, each two days, each on your data, is the limit of what an evaluation team can score well. ERP Research's rule is never to demo more than three products, and ERP Focus's checklist describes the same shape, with the five highest-scoring vendors discussed and three invited in. A fourth demo costs a week of the team's time and lowers the quality of every score, because by the fourth vendor the scorers can no longer remember what the first one showed.
The script is the control. Each vendor receives the same 5 to 8 end-to-end processes two weeks ahead. Three are common to most companies (procure to pay with a three-way match exception, order to cash with a credit hold, month-end close across two entities) and three or four are specific to your industry. Each receives the same sample data to load: a chart of accounts, 200 customers, 500 items or assets, and one month of transactions. The vendor that cannot load the data in two weeks has told you something about its implementation. Scorers score each process on the day, independently, on a 1 to 5 scale against the register lines that process covers.
The SI presents on day two, and is scored on a separate sheet. The software decides what is possible. The SI decides what you actually get, and its fees are 40 to 60% of the implementation cost. For a tier 2 product such as Epicor Kinetic, Acumatica, or SYSPRO, the vendor's partner channel usually implements, and the partner should be in the room. For SAP S/4HANA, Oracle Fusion Cloud ERP, or Microsoft Dynamics 365 Finance and Supply Chain, the SI is a separate firm with its own proposal, and the phases change as described below.
Reference checks happen before price talks, not after
Reference checks run during and just after the last demo, before commercials open, because a bad reference changes who you negotiate with. Three references per finalist is the minimum. At least one should be found through your own network or an industry association rather than supplied by the vendor. All should be go-lives within the last 18 months, the window ERP Research also uses, at a company of similar size and industry. All should be asked the same ten questions so the answers can be compared.
The reference that matters most is about the SI team, not the software. Ask which of the proposed consultants the reference actually worked with, whether the named project manager stayed to go-live, and how far the final fee moved from the signed statement of work. A survey reported by ERP Today found 31% of ERP implementations took longer than expected and 12% had significant delays. The references will tell you which SI teams were in the 31%.
Start commercials with two vendors and finish with one
Commercial negotiation runs with two finalists in parallel, each with the vendor and its SI together, for three to six weeks. The runner-up is not told it is the runner-up until the decision paper is approved. This is the only hold a buyer has over subscription uplift, and the figures say it is worth holding. ERP Research reports first-renewal price rises of 30 to 50% where uplift caps were not negotiated, and about 20% of the planned ERP budget spent fixing a licensing error. Neither cap is on offer once the vendor knows it is alone.
The five-year TCO template is the same for both finalists and is filled in by your team, not the vendor. It carries subscription or license plus maintenance, implementation services, data cleanup and migration, integration, and internal staff time. It also carries external PMO at $5,000 to $20,000 a month if used, dual-run costs for 12 to 24 months (Costbench range, moderate confidence), and post go-live support at 15 to 25% of implementation cost a year (Ecosire). RAND Group's advice to use the same timeframe for every finalist is right; a three-year view flatters subscription products and a ten-year view flatters perpetual licenses.
The SI statement of work negotiated here covers only the design phase, at a fixed fee, with named key staff and a right to replace them. The design phase is 10 to 20% of implementation hours, and committing to the build phase before design is done is where the 20 to 40% budget overruns that ERP Research reports on poorly scoped programs begin.
The decision paper records what you rejected and why
The decision paper is a ten-page document, not a slide of scores. It recommends one vendor and one SI, states the five-year TCO for each finalist on the same template, gives the implementation timeline with the go-live date, and lists the top ten risks with owners. It says under what conditions the recommendation would change, for example if the SI cannot confirm the named project manager, or if the design phase finds more than 15% of requirements need custom code. It also records every vendor dropped from the long list onward and the reason, because that list is what the board asks for two years later when the program is late.
The timeline in the paper is where selections most often mislead their sponsors. Mid-market tier 2 programs run 6 to 18 months (sysgenpro), and energy and utility programs 6 to 24 months (ERP Research). In programs we have advised, a tier 1 program needs 36 months from decision to go-live, which is longer than the published range of 12 to 36 months. The decision paper should carry the planning number, not the proposal number.
A scoresheet can point the wrong way while looking precise. Suppose two finalists finish at 83 and 79 on a 100-point weighted model where functional fit carries 30% of the weight. The leader got there by marking 180 of 200 register lines "supported" against the runner-up's 165. The paper reports the totals. It does not report that 9 of the 20 lines the leader missed were tagged Must, all in intercompany billing and consolidation across the company's 14 legal entities, while 32 of the 35 lines the runner-up missed were tagged Standard. It does not report that the three references were all vendor-supplied and none ran more than four entities, or that the SI was scored on the same sheet as the software and lifted the total by three points. The gap surfaces in the third month of the design phase, when the SI prices custom code for the nine Must lines and the customization share passes 15% of requirements. At that point the 20 to 40% saving that comes from holding customization under 10 to 15% is gone, the design phase runs long, and the program is carrying a change order before build has started. If the gap is found at month-end close after go-live instead, the cost is a year of consolidation by spreadsheet plus a program rescue. On published NetSuite rescue figures from Deliberate Directions (low confidence), a rescue runs 30 to 50% of the original budget. A paper that applied the Must lines as pass or fail gates before adding up the weighted score, as RAND Group's checklist says to, would have shown the runner-up as the only finalist that passed.
How the phases change for tier 1, tier 2, utilities, and manufacturers
The phases above fit a company of 500 to 5,000 employees choosing between tier 1 and tier 2 products with an SI in the picture. Four situations change them. The changes are to the length of specific phases and to what the team must add, not to the order of the phases.
| Situation | What changes | Total time | Why |
|---|---|---|---|
| Tier 1 (SAP S/4HANA, Oracle Fusion Cloud ERP, Microsoft Dynamics 365 Finance and Supply Chain, Workday Financials) | The SI selection becomes a separate track. Issue an SI RFP to 3 or 4 SIs per finalist product during the software RFP, hold SI presentations after the demos, and negotiate software and SI contracts on separate paper. Demos run three days, not two. | Five to eight months | SI fees are 40 to 60% of implementation cost, and on a tier 1 program the SI is often the larger contract. The software vendor's standard agreement runs to hundreds of pages and legal review alone can take three weeks. |
| Tier 2 (Infor CloudSuite, IFS Cloud, Epicor Kinetic, Oracle NetSuite, Acumatica, QAD Adaptive ERP, SYSPRO, Sage Intacct, Sage X3) | The vendor or its named partner implements, so the SI track folds into the vendor track. The long list can start at 6. Demos run two days. Commercials run two to three weeks. | Three to four months | Mid-market programs run 6 to 18 months, contracts are shorter, and the implementing partner is usually fixed by the vendor choice. |
| Utility (investor-owned, municipal, or cooperative) | Add a week or two to requirements for regulatory accounting (FERC or state equivalents), rate-case cost allocation, work and asset management, and the boundary with the customer information and billing system. Add utility-specific processes to the demo script. The long list should include IFS Cloud and Infor alongside SAP S/4HANA, Oracle Fusion, and Dynamics 365. | Four to five months (tier 2) or six to nine months (tier 1) | A utility selection that leaves asset and work management out of scope ends up with two programs. Rate-case timing may also fix the decision date, which fixes the last phase before the first one begins. |
| Manufacturer (discrete or process, one or several plants) | Hold at least one demo day at a plant with shop-floor users. Add quality, traceability, and recall to the script and settle the ERP and MES boundary before the long list. Multi-plant companies add a week or two to requirements to agree which processes will be common. | Four to five months | A manufacturer's differentiating requirements live on the shop floor, and a demo in a conference room cannot score them. |
The tier decision itself should be made when the long list is built, and not left to the demos. A company at $1B revenue with three plants and no plan for more can be well served by Infor CloudSuite Industrial or Epicor Kinetic. A company at $400M with 30 entities and a growth-by-acquisition strategy may be right to go to Oracle Fusion or Dynamics 365. The tier comparison article on this site, S/4HANA vs Oracle Fusion vs Dynamics 365 over $1B, sets out the conditions for a tier 1 shortlist.